Quantifying the Return on Investment of Medical Affairs in the AI Era

Quantifying the Return on Investment of Medical Affairs in the AI Era | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Quantifying the Return on Investment of Medical Affairs in the AI Era Georgios Bakalos, Thomas Buechele This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9193637/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 4 You are reading this latest preprint version Abstract Artificial intelligence (AI) is entering pharmaceutical Medical Affairs (MA) across medical writing, evidence generation, medical information, medical education, scientific communications, and field medical operations, changing how the function creates and delivers value. Yet MA has never had a standardised financial framework for measuring return on investment (ROI) comparable to those established in R&D and Commercial. Without such a framework, the additional value generated through AI adoption, and whether it justifies the associated investment, cannot be quantified. This paper presents the Medical Affairs Value Index (MAVI), a tool to quantify MA's ROI and assess how AI modifies it. MAVI integrates a Positive ROI model capturing value creation through revenue acceleration, cost avoidance, pipeline de-risking, and strategic option value, operationalised as a benefit–cost ratio, with a Reverse ROI model that quantifies the enterprise cost of MA underinvestment through probability-weighted loss estimation. Attribution coefficients for both upside (αMA) and downside (βMA) are introduced with governance safeguards. The primary output is a two-dimensional value map. AI is unlikely to materially reduce the cost of MA operations; it reallocates spending from manual activities to AI-augmented workflows, keeping the cost denominator approximately flat while the benefit numerator grows. An illustrative scenario demonstrates that under evidence-based assumptions, MAVI reveals a Positive ROI above breakeven before AI adoption, with AI further amplifying the return. The Reverse ROI, capturing enterprise losses MA helps prevent, remains broadly stable across both scenarios. MAVI provides the financial architecture to quantify both dimensions and track how AI modifies them over time. Medical Affairs artificial intelligence return on investment cost of inaction enterprise risk governance attribution pharmaceutical management Figures Figure 1 Figure 2 KEY POINTS 1) Medical Affairs (MA) lacks a standardised financial framework for measuring return on investment, a gap now compounded by AI’s capacity to amplify MA output without proportionally reducing cost. 2) The Medical Affairs Value Index (MAVI) integrates a Positive ROI model (benefit–cost ratio capturing value creation) with a Reverse ROI model (probability-weighted enterprise losses prevented), reported as a two-dimensional value map. 3) Attribution coefficients on both upside (αMA) and downside (βMA) govern how value is assigned to MA relative to co-producing functions, a requirement that becomes more demanding when AI is a co-producer. 4) AI is unlikely to materially reduce the cost of MA operations; it reallocates spending from manual tasks to AI-augmented workflows while introducing new cost lines for tools and governance. The cost denominator remains approximately flat while the benefit numerator grows. MAVI provides the framework to quantify this shift and assess its impact on both value creation and risk prevention. 5) Without a financial measurement framework, neither the baseline value MA creates nor the additional value AI enables is visible to decision-makers. MAVI provides the architecture to quantify both 1. INTRODUCTION 1.1 AI IN MEDICAL AFFAIRS Artificial intelligence is entering pharmaceutical operations across the product lifecycle. The McKinsey Global Institute estimates that generative AI could generate US $ 60–110 billion annually in economic value for the pharmaceutical and medical-product industries, with applications spanning research, clinical development, operations, commercial, and medical affairs.[ 1 ] AI adoption is already visible in adjacent functions: a landscape analysis of FDA regulatory submissions documented a growing number of drug and biologic applications incorporating AI and machine learning components between 2016 and 2021.[ 2 ] Within MA, AI is being applied across medical writing and publications, evidence generation and real-world evidence analysis, medical information inquiry handling, medical education, scientific communications, and field medical operations including MSL preparation and insight capture.[ 3 – 5 ] Taken together, these developments are shifting the operational profile of MA functions, changing which tasks require human effort, how much output a given team can produce, and what skill mix is needed. Without a framework for measuring the financial impact of these changes, organisations lack the basis for informed decisions about MA investment, workforce planning, and the governance of AI-augmented workflows. 1.2 THE MEASUREMENT GAP MA has been recognised as a strategic function alongside R&D and Commercial,[ 6 , 7 ] with growing attention to the value it brings to patients and to the organisations that invest in it.[ 8 ] Unlike its peer functions, however, MA has never had a standardised financial model for demonstrating its contribution. For example, Commercial can quantify value through revenue attribution and market share analysis, while R&D uses net present value (NPV) and probability-of-success models to justify investment decisions. MA, by contrast, typically reports activity-based metrics, like MSL interaction counts, publication numbers, medical information response volumes, or qualitative narratives without financial granularity.[ 9 , 10 ] This asymmetry becomes more consequential as AI enters MA operations. AI automates the tasks that produce countable output (literature reviews, response documents, evidence summaries) but the strategic contributions that drive organisational value (scientific judgement, regulatory credibility, evidence strategy) remain unmeasured.[ 11 ] The result is that AI increases what MA produces without providing a mechanism to quantify what that increased production is worth. Without a financial framework, the increased output cannot be translated into the financial terms required for resource-allocation decisions Industry bodies have developed increasingly sophisticated approaches,[ 9 , 12 , 13 ] including multi-stakeholder value models[ 11 , 14 ]and validated workforce instruments.[ 15 ] None, however, provides a financial framework for quantifying the value MA creates, nor a basis for understanding how AI may amplify or transform that value. Six gaps persist: (1) incomplete value capture, particularly the cost of underinvestment; (2) attribution challenges in co-created value; (3) misalignment between multi-year returns and annual budget cycles; (4) insufficient financial rigour; (5) sector-specific variation; and (6) AI-readiness. 1.3 OBJECTIVE This paper presents the Medical Affairs Value Index (MAVI), a tool developed to quantify MA’s return on investment and to explore how AI is impacting its ROI. The central proposition is that AI can increase MA’s Positive ROI, not by reducing cost, but by increasing the output the same cost produces, while strengthening the risk-governance case. However, this additional value is not captured without a measurement framework that governs attribution. MAVI provides both the measurement architecture and the attribution governance. MAVI is deliberately structured as a multi-component index rather than a single score: its principal output is a two-dimensional value map that allows organisations to assess MA’s contribution on both value creation and risk prevention simultaneously. The framework is conceptual and designed for progressive organisational calibration; its immediate purpose is to provide a structured basis for investment discussions, not a definitive decision threshold. Although illustrated with pharmaceutical examples, MAVI is intended for MA functions across biopharma and MedTech. 2. MAVI: MEASURING MA ROI Figure 1 provides a conceptual overview of the MAVI framework. The notation glossary (Table 1 ) defines all symbols. The evidence search methodology is documented in ESM File S2. Table 1 MAVI framework notation glossary Symbol Name Definition Units / Range αMA Upside attribution Fraction of revenue value attributable to MA; set via governed process 0 < αMA ≤ 1 βMA,k Downside attribution Fraction of loss category k preventable by MA 0 < βMA ≤ 1 ΔRevt Incremental revenue Revenue acceleration vs. counterfactual in period t US $ per period CAt Cost avoided Expenditures averted through MA in period t US $ per period ΔrNPVt Risk-adjusted NPVΔ Change in risk-adjusted NPV from MA de-risking US $ per period SOVt Strategic option value Option premium from MA-enabled future choices US $ per period OverlapAdjt Overlap adjustment Deduction for double-counted value (Eq. 1a) US $ per period r Discount rate Organisation’s WACC 8–12% T Evaluation horizon Periods for discounted analysis 3–5 years P(Lossk,t) Loss probability Event probability absent adequate MA 0–1 CorrAdj Correlation adj. Deduction for non-independence (Eq. 2a; PV-denominated; applied after discounting) US $ 2.1 Positive ROI model The Positive ROI captures measurable financial value attributable to MA over a T-period evaluation horizon using discounted cash-flow methodology, operationalised as a benefit–cost ratio (BCR = PV(Benefits)/PV(Cost)). Net ROI = BCR − 1 is reported alongside all BCR values for interpretability. Equation 1a , Present value of MA-attributable benefits: PV(BenefitsMA) = ΣTt = 1 [(αMA × ΔRevt) + CAt + ΔrNPVt + SOVt − OverlapAdjt] × (1 + r) − t Equation 1b , Present value of MA investment: PV(CostMA) = ΣTt = 1 [MA_Costt] × (1 + r) − t Equation 1c , Positive ROI: Positive ROI = PV(BenefitsMA) / PV(CostMA) Each component (Table 2 ) represents a distinct value channel. The coefficient αMA (0 < αMA ≤ 1) is applied to ΔRevenue, where cross-functional co-creation with Commercial is most pronounced; ΔRev should be operationalised as incremental contribution margin, not gross revenue. The remaining components (CA (cost avoided), ΔrNPV (risk-adjusted NPV change from pipeline de-risking), and SOV (strategic option value)) capture only the MA-attributable portion by construction, subject to the documentation requirements in Box 1. Table 3 proposes indicative αMA ranges by activity type. Table 2 Positive ROI component definitions and benchmarks Component Definition Operationalisation Key Benchmark αMA × ΔRevt Incremental net sales (MA-attributed fraction) Comparator launch matching, time-to-adoption IQVIA [ 19 ] (grey; vendor; illustrative) CAt MA-attributable expenditures prevented (Box 1) Historical cost tracking; MA ownership documented ~ $ 535K/Phase III amendment [ 20 ] ΔrNPVt MA-attributable change in asset value (Box 1) ΔPoS × NPV × discount; documented assumptions PoS data [ 21 , 22 ] SOVt MA-attributable option premium (Box 1) Binomial lattice at decision gates; incremental Real options methodology OverlapAdjt Deduction for double-counted value Mapping rules or 15–25% haircut Organisation-specific Reference numbers correspond to the main manuscript reference list. BOX 1. Operational Rules for "Attributable by Construction" Components CA, ΔrNPV, and SOV bypass αMA only when each claimed item satisfies all four criteria: (i) Documented MA ownership: a named MA individual or team led or co-led the activity, evidenced by project records, approval chains, or RACI matrices. (ii) Counterfactual plausibility: management can articulate why the outcome would not have occurred (or would have been materially worse) without MA involvement. (iii) Exclusion of shared credit: if another function has an equal or greater claim, the item must be resolved by: (a) applying an explicit attribution factor agreed with the co-owning function (analogous to αMA), (b) splitting the item by documented agreement and assigning each portion to the appropriate component, or (c) excluding it entirely. Items should not be reclassified into a different component category to circumvent shared ownership. (iv) Annual audit: items classified as "attributable by construction" are reviewed annually by the cross-functional governance panel (including Finance and Compliance). Items failing any criterion are reclassified. (v) AI-generated outputs (added in v2): when AI tools contribute to an output claimed under CA, ΔrNPV, or SOV, the documentation must specify: (a) which AI tool was used and for which task; (b) the nature and extent of human review and modification; and (c) whether the strategic direction (which questions to answer, which populations to study, which endpoints to prioritise) was set by MA. AI-assisted execution does not transfer attribution to IT or data science if MA directed the strategy and validated the output. Example of a failed Box 1 item: an MSL-facilitated investigator-initiated trial application where Commercial independently contacted the same investigator is classified as "shared credit" under criterion (iii) and receives an explicit attribution factor agreed between MA and Commercial, rather than claiming full CA attribution. Items that do not meet these criteria should either receive an explicit attribution factor or be excluded from the calculation. Table 3 Indicative αMA ranges by MA activity type MA Activity Category Indicative αMA Rationale KOL engagement and medical education 0.15–0.30 Substantial Commercial co-contribution Protocol design and trial feasibility 0.20–0.40 MA contribution more isolable; R&D co-creation RWE generation and evidence planning 0.30–0.50 MA typically leads; Market Access co-benefits Post-market surveillance and safety 0.30–0.50 Relatively direct causal link to regulatory outcomes IIT-driven label expansion 0.25–0.40 MA initiates; R&D executes submission Medical information and congress 0.10–0.20 Diffuse impact; attribution difficult Expert-informed starting points for cross-functional calibration, not empirically validated parameters. 2.2 Reverse ROI model The Reverse ROI quantifies expected enterprise losses when MA is eliminated or materially under-resourced, framing MA investment as risk governance. Equation 2a , Present value of MA-preventable expected losses: PV(LossesNoMA) = Σk ΣTt = 1 [βMA,k × P(Lossk,t) × Impactk,t × Durationk,t] × (1 + r) − t − CorrAdj Equation 2b , Present value of foregone MA budget: PV(Foregone MA Budget) = ΣTt = 1 [MA_Costt] × (1 + r) − t Equation 2c , Reverse ROI: Reverse ROI = PV(LossesNoMA) / PV(Foregone MA Budget) The downside attribution coefficient βMA,k (0 < βMA,k ≤ 1) represents the fraction of each loss category that MA is credibly positioned to prevent. Table 4 Panel B proposes indicative βMA ranges. Minimum reporting standards require documentation of probability sources, impact definitions, duration logic, and sensitivity analysis. The probabilities P(Lossk,t) represent industry-level event frequencies, not conditional probabilities specific to MA underinvestment, which is a recognised limitation. Box 2 describes staged implementation for probability calibration. The correlation safeguard (CorrAdj) prevents double counting of co-occurring events. Table 4 Evidence credibility and Reverse ROI loss categories Panel A: Evidence source credibility Benchmark Source Type Credibility Key Limitation Ref Protocol amendment cost (~ $ 535K) Peer-reviewed High US-centric sample [ 20 ] Delay day value ( $ 500K– $ 800K) Peer-reviewed High Median varies by TA [ 22 ] Clinical PoS rates Peer-reviewed High Historical; varies by TA [ 21 , 22 ] R&D costs ( $ 2.6B capitalised) Peer-reviewed High Industry-funded study [ 23 ] Launch benchmarks Grey—vendor Low–Medium Potential commercial bias; illustrative only [ 18 ] Regulatory sanctions ( $ 1–10M) Industry estimates Low Wide range; no single robust source; illustrative Product recall costs Press / regulatory Medium Single-case examples [ 24 ] EU-MDR penalties Regulation / legal High (structure) Monetary exposure jurisdiction-specific; applies to MedTech/combination-product MA [ 25 ] Panel B: Reverse ROI loss categories with indicative βMA BOX 2. Implementation Upgrade Path for Probability Modelling (ΔP Calibration) Stage 1, Baseline frequency + conservative βMA Use unconditional industry-level event frequencies from published sources (Table 4). Apply conservative βMA values from Table 4 Panel B. This is the lowest-data-requirement entry point and yields a pragmatic lower bound. Stage 2, Internal incident log + empirical priors Retrospectively analyse the organisation’s own loss events over 3–5 years. Use these internal frequencies as Bayesian updates to the Stage 1 priors. For example, if the company has experienced two protocol-amendment events attributable to insufficient MA input in five years, this empirical rate (2/5 = 0.40/year) replaces the industry benchmark for that category. Stage 3, Capability-specific ΔP estimation Model partial under-resourcing rather than binary MA removal. Estimate the relative risk reduction (ΔP) associated with specific MA capabilities (e.g., pharmacovigilance staffing level, evidence-planning maturity score). This allows modelling of investment increments (e.g., "+5 MSLs reduces launch delay probability by X%") rather than all-or-nothing scenarios. Structured expert elicitation (Delphi panels with Finance and Compliance) is the recommended calibration method at this stage. AI-era note (added in v2): AI adoption may accelerate the transition from Stage 1 to Stage 2 by enabling systematic tracking of AI-augmented outputs and their outcomes. However, AI also introduces new loss categories (e.g., regulatory findings from inadequately reviewed AI-generated content) that may not have Stage 1 industry benchmarks. Organisations should document these emerging categories with conservative probability estimates until empirical data are available. Organisations should document which Stage they are operating at and disclose this alongside reported Reverse ROI values. Migration from Stage 1 to Stage 3 is expected over 2–3 annual cycles. 2.3 The two-dimensional value map MAVI is reported as a two-dimensional value map (Fig. 2 ), with Positive ROI (BCR) on the horizontal axis and Reverse ROI (βMA-adjusted) on the vertical axis. This format preserves the distinct information each dimension carries and avoids combining non-commensurable ratios into a single score. Dashed reference lines at 1.0× on both axes mark breakeven: a Positive ROI above 1.0× indicates that MA-attributable benefits exceed MA cost; a Reverse ROI above 1.0× indicates that expected losses prevented exceed the MA budget that would be foregone. Organisations positioned below breakeven on both axes face a diagnostic question: does the result reflect underperformance, or conservative measurement? Those above breakeven on risk prevention but below on value creation have a quantified case for MA as a governance function. The value map thus supports investment discussions grounded in evidence rather than in assumptions. To prevent double counting, each value stream is assigned to a single dimension (Table 5 ). Table 5 Cross-panel mutual exclusivity: mapping rules and edge cases Value Stream Home Panel Excluded From Edge Case Resolution Protocol amendments avoided Positive ROI (CA) Reverse ROI Residual amendment risk beyond MA input Cap residual at 25% in Reverse ROI; document Launch delay prevention Reverse ROI Positive ROI (ΔRev) Faster launch = revenue acceleration? Revenue acceleration goes to ΔRev; delay prevention stays in Reverse ROI as distinct pathway Regulatory compliance costs avoided Positive ROI (CA) Reverse ROI Sanctions from non-compliance? Compliance costs avoided → CA; sanctions risk (distinct from CA) → Reverse ROI with βMA Competitive displacement Reverse ROI Positive ROI (ΔRev) Evidence differentiation drives revenue? Revenue from evidence → ΔRev; chronic competitor gap (absent MA) → Reverse ROI Product recall prevention Reverse ROI Positive ROI (CA) Signal detection cost savings? Operational savings → CA; catastrophic recall risk → Reverse ROI Evidence-enabled label expansion Positive ROI (ΔrNPV) Reverse ROI Missed expansion = lost revenue? NPV appreciation → ΔrNPV; forfeited option excluded (subsumed in ΔrNPV counterfactual) When both panels have a legitimate claim, the upstream (more direct) pathway takes the home assignment. Downstream (indirect) pathways are capped at 25–50% and documented. The overriding principle: no dollar of claimed value appears twice in the two-dimensional value map. Table 6 Illustrative scenario: pre-AI and AI-augmented (mid-size pharmaceutical company, T = 5 years, r = 10%) Parameter Pre-AI AI-Era Source / Basis Portfolio revenue US $ 1B US $ 1B Assumed Annual MA investment US $ 18M (60 FTEs) US $ 18M (55 FTEs + $ 2.5M AI) Reallocation, not reduction Time horizon (T) 5 years 5 years Design choice Discount rate (r) 10% 10% Pharma WACC midpoint αMA (ΔRev) 0.30 0.30 Evidence-based; Table 3 range OverlapAdj 15% 25% Higher: more co-producing functions PV(BenefitsMA) $ 73.5M ~ $ 83M Eq. 1a PV(CostMA) $ 68.2M $ 68.2M Eq. 1b Positive ROI (BCR) 1.08× ~ 1.22× Eq. 1c Net ROI (BCR − 1) + 0.08 + 0.22 Weighted βMA 0.50 0.55 Residual risk more human-dependent PV(Gross Expected Losses, 5 year) $ 50M ( $ 35–78M) ~ $ 47M Eq. 2a before βMA; AI reduces P(Loss) ~ 15% PV(βMA-adjusted Losses, 5 year) $ 25M ~ $ 27M Eq. 2a PV(Foregone MA Budget, 5 year) $ 68.2M $ 68.2M Eq. 2b (matches Eq. 1b) Reverse ROI (gross) 0.73× ~ 0.69× Eq. 2c Reverse ROI (βMA-adjusted) 0.37× ~ 0.40× Eq. 2c Value map position (Fig. 2 ) Point A Point B Above breakeven on Positive ROI; Reverse ROI stable Probability stage (Box 2) Stage 1 Stage 1 Baseline industry frequencies Positive ROI (BCR) exceeds 1× under evidence-based parameters (αMA = 0.30). Under conservative assumptions (αMA = 0.25, BCR = 0.87×), see ESM File S3. AI-era estimates reflect plausible assumptions about output multiplication at constant cost; they are illustrative, not validated. Benchmark inputs sourced from Table 4 Panel A. Gross Reverse ROI represents total enterprise risk exposure; βMA-adjusted represents the MA-preventable portion 3. HOW AI MODIFIES THE ROI 3.1 THE COST DENOMINATOR: REALLOCATION RATHER THAN REDUCTION AI adoption in MA does not materially reduce total function cost.[ 3 ] Increased output volume requires proportionally greater expert review capacity, and new cost lines emerge: AI platform licences, validation infrastructure, and governance overhead. FTE profiles shift (fewer staff perform manual tasks, more serve as AI-literate medical scientists and senior reviewers) but aggregate cost remains broadly stable. Under a representative scenario, a pre-AI MA function of 60 FTEs at US $ 18 million per year transitions to 55 FTEs plus US $ 2–3 million in AI tools and infrastructure, yielding a comparable annual cost of approximately US $ 18 million. The PV(CostMA) denominator in Equations 1c and 2c remains approximately US $ 68.2M over five years at a 10% discount rate. Any improvement in ROI must therefore come from the numerator. 3.2 THE BENEFIT NUMERATOR: OUTPUT MULTIPLICATION AT CONSTANT COST AI increases what MA produces for the same investment across its principal activities (Table 7). Medical writing and publications. AI-assisted drafting enables higher publication output, faster turnaround on evidence summaries, and broader congress presence from the same editorial team. The resulting gains affect CA (cost avoided through faster medical information responses and fewer compliance gaps from delayed answers) and SOV (more evidence assets available at key decision gates). Evidence generation. AI-accelerated real-world evidence analysis compresses timelines for post-market studies, supports faster label-expansion dossiers, and increases the volume of evidence packages for health technology assessment submissions. This directly affects ΔrNPV: more evidence generated faster de-risks pipeline assets and advances the revenue they generate. Medical information. AI-enabled triage and response drafting reduce response times while maintaining accuracy under mandatory medical review[ 3 , 4 ] lowering the probability of compliance breaches (a CA component). MSL interactions and field insights. AI-augmented preparation and structured insight-capture systems improve the quality and traceability of the αMA × ΔRevenue component (the pathway from scientific engagement to downstream treatment adoption). Medical education and scientific communications. AI-personalised content extends MA’s reach across geographies and stakeholder segments without proportional FTE growth, generating new strategic options (SOV). Estimated magnitude. If AI increases the Positive ROI numerator by approximately 10–15% through cumulative gains across these activities while the denominator remains constant, PV(BenefitsMA) rises from approximately US $ 74M (pre-AI) to approximately US $ 84M. This is partially offset by a higher OverlapAdj (increasing from 15% to 20% as more functions contribute to AI-augmented outcomes). The net result: Positive ROI (BCR) improves from approximately 1.08× to approximately 1.22×. 3.3 THE REVERSE ROI: CONCENTRATION OF HUMAN-DEPENDENT RISK On the risk-prevention side, AI modifies loss probabilities. AI-augmented literature surveillance reduces the probability of missing safety-relevant publications. AI-assisted regulatory document preparation lowers submission delay risk. AI-enabled medical information triage reduces compliance exposure. At the same time, AI introduces new risk categories. AI systems in medical contexts exhibit variable error rates depending on governance architecture; systems with retrieval-augmented generation and mandatory human oversight achieve substantially lower error rates than unguided implementations.[ 3 ] The FDA has emphasised that AI outputs intended to support regulatory decision-making require credibility assessment frameworks including human validation.[ 4 , 5 ] AI-generated evidence or communications released without adequate medical review could result in regulatory findings, reputational harm, or patient safety concerns (risks with no pre-AI equivalent). The net effect on βMA runs against expectations. AI generally reduces the probability of routine adverse events, but the residual risk (the risk AI cannot mitigate without human oversight) concentrates in medical judgement: signal escalation decisions, evidence strategy prioritisation, and the scientific credibility that sustains regulator and payer trust. βMA for these residual categories may therefore increase from 0.50 to 0.55–0.60 in an AI-augmented environment, even as the total risk pool decreases. 3.4 THE MEASUREMENT IMPERATIVE The analysis above points to a structural concern. AI amplifies MA output and the benefit numerator grows, but this improvement is invisible without a financial framework to capture it. MA continues to report the same activity-based metrics it used before AI adoption, even as the value those activities generate has materially increased. The function produces more, but in the absence of a financial framework, the return on AI investment within MA cannot be demonstrated MAVI addresses this through governed αMA and βMA coefficients, Box 1 documentation requirements, and cross-functional calibration panels. These provide the measurement architecture that converts AI-amplified output into a quantified financial case for continued MA investment. 4. ILLUSTRATIVE APPLICATION This scenario demonstrates framework mechanics, not validated estimates. All parameters are derived from published benchmarks classified by credibility (Table 4 , Panel A).[ 18 – 25 ] Low-credibility benchmarks are flagged and should not inform investment decisions. 4.1 COMPANY PROFILE Mid-size pharmaceutical company, US $ 1 billion portfolio revenue, three marketed products, two label-expansion candidates. T = 5 years, r = 10%. Year-by-year calculations are provided in ESM File S1. 4.2 PRE-AI BASELINE Annual MA investment: US $ 18M (60 FTEs, no AI tools). PV(CostMA) = US $ 68.2M. With evidence-based parameters (αMA = 0.30, reflecting documented MA contribution to evidence-mediated adoption; CA including regulatory query avoidance; OverlapAdj = 15%), PV(BenefitsMA) = US $ 74M. Positive ROI (BCR) = 1.08× (Net ROI = + 0.08). PV(Gross expected losses) = US $ 50M ( $ 35–78M range). Weighted βMA = 0.50. PV(βMA-adjusted losses) = US $ 25M. Reverse ROI (βMA-adjusted) = 0.37×; Reverse ROI (gross) = 0.73×. Value map position: above breakeven on Positive ROI; below breakeven on Reverse ROI (Fig. 2 , Point A). 4.3 AI-ERA SCENARIO Annual MA investment: US $ 18M (55 FTEs + US $ 2.5M AI tools and infrastructure). PV(CostMA) = US $ 68.2M (unchanged). AI-augmented benefits: ΔrNPV + 30% (accelerated evidence generation and additional label-expansion dossiers); CA + 15% (faster medical information turnaround, fewer protocol amendments through AI-assisted design); SOV doubles (AI-enabled strategic options including adaptive trial designs and predictive analytics); αMA × ΔRev + 10% (AI-augmented MSL engagement). OverlapAdj increases from 15% to 20%. PV(BenefitsMA) ≈ US $ 84M. Positive ROI (BCR) ≈ 1.22× (Net ROI ≈ + 0.22). Reverse ROI: P(Loss) decreases approximately 15% for routine surveillance categories; βMA increases modestly from 0.50 to 0.60 for residual risk categories. PV(βMA-adjusted losses) ≈ US $ 27M. Reverse ROI (βMA-adjusted) ≈ 0.40×. The Reverse ROI remains broadly stable, confirming that AI’s primary impact is on value creation rather than risk prevention. Value map position: Positive ROI increases from 1.08× to 1.22× while Reverse ROI moves from 0.37× to 0.40× (Fig. 2 , Point B). 4.4 INTERPRETATION The illustrative scenario demonstrates three properties of the framework. First, the result is sensitive to attribution governance: with evidence-based parameters (αMA = 0.30), Positive ROI exceeds breakeven (Fig. 2 , Point A); under conservative assumptions (αMA = 0.25, ESM File S3), it does not. The difference is measurement calibration, not MA performance. Second, the Reverse ROI dimension reveals a contribution that single-dimension analysis would miss: MA prevents expected enterprise losses worth a material fraction of its own budget, providing a risk-governance case that is invisible without two-dimensional measurement. Third, AI amplifies the Positive ROI through output multiplication at constant cost, while the Reverse ROI remains broadly stable (Fig. 2 , arrow from Point A to Point B), a shift that is quantifiable only because the framework captures both dimensions." This addresses a longstanding perception that MA’s ROI is inherently unmeasurable. MAVI challenges this by applying standard corporate finance methodology, discounted cash flow, real options, probability-weighted loss estimation, to MA’s specific value channels. The illustrative results suggest that the difficulty has not been that MA lacks ROI, but that it has lacked a framework for quantifying it. ESM File S3 presents a stress test using conservative parameters (αMA = 0.25, BCR = 0.87×), demonstrating that even when Positive ROI falls below breakeven, the Reverse ROI dimension sustains the investment case. These figures are illustrative and should not be treated as validated estimates. The underlying organisational question remains whether the enterprise is prepared to accept the unmitigated downside exposure that would follow from materially reducing MA capability. Events such as product recalls (US $ 100M–US $ 1B+)24 may be low-probability but can be existential for mid-size companies. A further consideration is progressive deskilling. As AI assumes routine tasks, junior MA professionals have fewer opportunities to develop domain expertise, potentially weakening the human oversight layer on which βMA depends. 5. IMPLICATIONS AND PATH FORWARD 5.1 MAVI IN THE AI ERA The analysis presented here suggests two conclusions. First, the financial case for MA is sensitive to measurement approach: when a structured framework is applied with governed attribution, the result differs materially from what activity-based reporting alone would suggest. Second, AI modifies the Positive ROI through output multiplication at approximately constant cost, while the risk-governance dimension remains stable, but this effect is quantifiable only with a framework that captures both dimensions. MAVI provides that framework. In a pre-AI environment, it offers a structured basis for MA investment discussions. In an AI-augmented environment, it enables organisations to assess whether AI adoption is generating a measurable return Zdon et al. characterised MA as a strategic asset in launch execution, corporate reputation, and innovation.[ 26 ] MAVI offers a quantitative approach to evaluating that characterisation, by enabling organisations to assess MA's contribution on both value creation and risk prevention within a single measurement architecture. MAVI is not a health technology assessment framework. It draws on corporate finance concepts (NPV, real options, attribution) and is intended for senior leadership and finance decision-makers rather than reimbursement authorities. The structural challenge MAVI addresses is not unique to MA. Any knowledge-intensive function that co-creates value across organisational boundaries (regulatory affairs, quality assurance, health economics, corporate affairs) faces analogous measurement difficulties that AI is likely to amplify. The two-dimensional approach (positive return plus risk prevention) and the attribution governance architecture presented here may serve as a template for the financial evaluation of such functions more broadly. 5.2 COMPLIANCE BOUNDARIES The following are structural requirements: (i) only non-promotional activities are eligible for MAVI quantification; (ii) promotional activities are excluded; (iii) calibration panels for αMA and βMA must include compliance representation; (iv) MAVI should not serve as a basis for MA personnel compensation linked to revenue outcomes; (v) revenue modelling should be evidence-mediated. AI-generated outputs used within MAVI quantification require human validation before being counted as MA-attributable. 5.3 LIMITATIONS AND NEXT STEPS The proposed αMA and βMA ranges are expert-informed rather than empirically derived. Prospective calibration through cross-functional Delphi panels would strengthen the evidence base. Several benchmark inputs rely on heterogeneous sources of variable quality (Table 4 , Panel A). The framework has not been prospectively validated and does not currently include patient outcomes as a direct value dimension. Three AI-era priorities emerge. First, organisations should track how AI adoption changes αMA and βMA over time through longitudinal within-company calibration. Second, human-AI attribution protocols are needed: when AI and a medical scientist together produce an evidence package that supports a label expansion, the attribution must be governed and documented. Third, industry collaboration to develop AI-adjusted parameter libraries would reduce adoption burden and improve comparability. A phased implementation programme, beginning with cost-avoided tracking, then adding ΔrNPV and Reverse ROI, and subsequently integrating AI-adjusted parameters, may better match organisational readiness than simultaneous deployment 6. CONCLUSION MAVI provides a structured financial framework for quantifying the return on investment of Medical Affairs across two dimensions: value creation and risk prevention. Applied with evidence-based parameters, it reveals a Positive ROI above breakeven before AI adoption and demonstrates that AI further amplifies this return through output multiplication at approximately constant cost, while the risk-governance dimension remains stable. The framework addresses a measurement gap that has left MA unable to articulate its financial contribution in terms comparable to those used by R&D and Commercial functions. As AI reshapes pharmaceutical operations, the need for such a framework extends beyond MA to any knowledge-intensive function whose value is co-created, diffuse, and difficult to attribute. Empirical calibration through prospective, cross-functional implementation is the essential next step. Declarations Author contributions: G.B. Conceptualization, Methodology, Literature Synthesis, Writing Original Draft, Implementation Science Integration, MAVI Framework Development; T.B. Literature Synthesis, Framework Development, Review & Editing; All authors approved the final manuscript and accept responsibility for its content. Conflicts of interest: G.B. is employed by Helsinn Healthcare SA, Switzerland; T.B. is employed by Miltenyi Biomedicine, Germany. The views expressed are those of the authors and do not necessarily represent the official positions of their affiliated institutions. Funding: This work received no external funding. Ethics approval: Not applicable (no primary data collection). Data Availability: This article is based on published literature, publicly available regulatory guidance, and expert synthesis. No original data were generated. The year-by-year present-value workbook is provided as ESM File S1. The literature research strategy is provided as ESM File S2. Ethics and AI Use Statement: This article does not involve human participants or patient data. Generative AI tools were used to assist with literature research, reference formatting and structural editing (Claude, Anthropic). Figure 1, 2 and Supplementary Figure 1 were generated programmatically using Python (matplotlib). All AI-generated content was critically reviewed, verified and substantially modified by the authors. All citations were manually validated for accuracy through database searches and DOI verification. The authors accept full responsibility for the accuracy, scientific integrity, and intellectual content of this work.. References Shah B, Adabala Viswa C, Zurkiya D, Leydon E, Bleys J. Generative AI in the pharmaceutical industry: moving from hype to reality. McKinsey & Company; 2024 Jan 9. Available from: https://www.mckinsey.com/industries/life-sciences/our-insights/generative-ai-in-the-pharmaceutical-industry-moving-from-hype-to-reality . Accessed 2026 Mar. Liu Q, Huang R, Hsieh J, et al. Landscape analysis of the application of artificial intelligence and machine learning in regulatory submissions for drug development from 2016 to 2021. Clin Pharmacol Ther. 2023;113(4):771–4. 10.1002/cpt.2668 . Bate A, Hobbiger SF. Artificial intelligence, real-world automation and the safety of medicines. Drug Saf. 2021;44(2):125–32. 10.1007/s40264-020-01001-7 . US Food and Drug Administration. Using artificial intelligence & machine learning in the development of drug and biological products: discussion paper. Silver Spring (MD): FDA; 2023 May (revised 2025 Feb). Available from: https://www.fda.gov/media/167973/download . Accessed 2026 Mar. US Food and Drug Administration. Considerations for the use of artificial intelligence to support regulatory decision-making for drug and biological products: draft guidance for industry. Silver Spring (MD): FDA; 2025 Jan. Federal Register docket FDA-2024-D-4689. Available from: https://www.federalregister.gov/documents/2025/01/07/2024-31542/considerations-for-the-use-of-artificial-intelligence-to-support-regulatory-decision-making-for-drug . Accessed 2026 Mar. Darino L, Knepp A, Mills N, Tinkoff D. How pharma can accelerate business impact from advanced analytics. McKinsey & Company; 2018 Jan 8. Available from: https://www.mckinsey.com/industries/life-sciences/our-insights/how-pharma-can-accelerate-business-impact-from-advanced-analytics . Accessed 2026 Mar. Algazy J, Garcia A, Ryan S, Westra A, Zemp A. A vision for medical affairs 2030: five priorities for patient impact. McKinsey & Company; 2023 Oct. Available from: https://www.mckinsey.com/industries/life-sciences/our-insights/a-vision-for-medical-affairs-2030-five-priorities-for-patient-impact . Accessed 2026 Mar. Farrington AD, Frøstrup AG, Dahl P. The value and deliverables of medical affairs: affiliate perspectives and future expectations. Pharmaceut Med. 2023;37(6):417–24. 10.1007/s40290-023-00501-y . Medical Affairs Professional Society (MAPS). Measuring value and impact in medical affairs: standards and guidance document. Golden (CO): MAPS. 2025. Available from: https://medicalaffairs.org/wp-content/uploads/2025/01/UPDATE_1.21.25-STRATEGY_Value-and-Impact-in-Medical-Affairs_SG-pwp.pdf . Accessed 2026 Mar. Dyer S, Hyder C, Kraemer J. Challenges of key performance indicators and metrics for measuring Medical Science Liaison performance: insights from a global survey. Pharmacy. 2025;13(2):51. 10.3390/pharmacy13020051 . Jandhyala R. Development and validation of the Medical Affairs Pharmaceutical Physician Value (MAPPval) Instrument. Pharmaceut Med. 2022;36(1):47–57. 10.1007/s40290-021-00413-9 . Kaplan RS, Norton DP. The balanced scorecard: translating strategy into action. Boston (MA): Harvard Business School Press; 1996. Kaplan RS, Norton DP. Strategy maps: converting intangible assets into tangible outcomes. Boston (MA): Harvard Business School Press; 2004. Jandhyala R. Development of a definition for medical affairs using the Jandhyala method for observing consensus opinion among medical affairs pharmaceutical physicians. Front Pharmacol. 2022;13:842431. 10.3389/fphar.2022.842431 . Jandhyala R. Development, validation and implementation of the medical affairs pharmaceutical physician work-related quality of life instrument. Curr Med Res Opin. 2023;39(12):1567–74. 10.1080/03007995.2023.2174747 . Saleem M, Cesario L, Wilcox L, et al. Evaluating metrics applied to the Medical Science Liaison (MSL) role: a survey-based study of Canadian MSL leaders. Ther Innov Regul Sci. 2021;55(5):954–65. 10.1007/s43441-021-00291-y . Lucid Group. Share of Scientific Voice (SoSV): methodology and benchmarking. White Paper. Published 2023. Available from: https://wearelucidgroup.com/white-papers/medical-affairs-kpis . Accessed 2026 Mar. IQVIA. In pursuit of Medical Launch Excellence: lessons for medical affairs from IQVIA’s Launch Excellence research (White paper). Durham (NC): IQVIA; 2023 Nov 30. Available from: https://www.iqvia.com/-/media/iqvia/pdfs/library/white-papers/in-pursuit-of-medical-launch-excellence.pdf . Accessed 2026 Mar. Getz KA, Stergiopoulos S, Marlborough M, Whitehill J, Curran M, Kaitin KI. Quantifying the magnitude and cost of collecting extraneous protocol data. Am J Ther. 2015;22(2):117–24. 10.1097/MJT.0b013e31826fc4aa . Hay M, Thomas DW, Craighead JL, Economides C, Rosenthal J. Clinical development success rates for investigational drugs. Nat Biotechnol. 2014;32(1):40–51. 10.1038/nbt.2786 . Wong CH, Siah KW, Lo AW. Estimation of clinical trial success rates and related parameters. Biostatistics. 2019;20(2):273–86. 10.1093/biostatistics/kxx069 . Smith ZP, DiMasi JA, Getz KA. New estimates on the cost of a delay day in drug development. Ther Innov Regul Sci. 2024;58(5):855–62. 10.1007/s43441-024-00667-w . DiMasi JA, Grabowski HG, Hansen RW. Innovation in the pharmaceutical industry: new estimates of R&D costs. J Health Econ. 2016;47:20–33. 10.1016/j.jhealeco.2016.01.012 . Associated Press. Philips will pay $ 1.1 billion to resolve US lawsuits over breathing machines that expel debris. AP News. 2024 Apr 29. Available from: https://apnews.com/article/philips-sleep-apnea-settlement-recall-safety-fc5cc1509656645d7f5679e0f1b34b85 . Accessed 2026 Mar. European Parliament and of the Council. Regulation (EU) 2017/745 of 5 April 2017 on medical devices. OJ L 117, 5.5.2017, pp. 1–175. https://eur-lex.europa.eu/eli/reg/2017/745/oj/eng Zdon J, Chatgilaou GJ, Henderson D, et al. How can general managers best leverage medical affairs now and in the future? Pharmaceut Med. 2024;38(4):277–90. 10.1007/s40290-024-00528-9 . Additional Declarations Competing interest reported. G.B. is employed by Helsinn Healthcare SA, Switzerland; T.B. is employed by Miltenyi Biomedicine, Germany. Supplementary Files MAVIESMS1.xlsx MAVIESMS3.docx MAVIESMS2.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviewers invited by journal 16 May, 2026 Editor assigned by journal 25 Mar, 2026 Submission checks completed at journal 24 Mar, 2026 First submitted to journal 22 Mar, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9193637","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":612093811,"identity":"ca39404c-6978-4c94-bf52-05857743b1fa","order_by":0,"name":"Georgios Bakalos","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABEElEQVRIiWNgGAWjYDACZihtwMDGwPCAgUEOxDnwgGgtCQwMxmAtCcTYBtOS2ADi4dMi78587OOXijoGc/ZjiR8SKurS54cdfgi0xU5OtwG7FsPDbMmzZc4cZrDsSTsskXDmcO7G22kGQC3JxmYHcGhp5jFmlmw7wGBwIL1BIrHtQO7G2QkgLQcSt+HXUsdgcP5584/Etrp0w9npH/BqkWfmMWb82MbMYHAj7RjQFuYEeekc/LYYMLMlMzOcOcxjOeNZmgXQL4YbpHMKDiQY4PaLfP/hw4w/KurkzPnTjG98qKiTl5+dvvnDhwo7OVxaDIDizDwMDDwoIqBowgnkGxgYGH+gi4yCUTAKRsEoQAYAZ0xgX8umf5IAAAAASUVORK5CYII=","orcid":"","institution":"Helsinn Healthcare SA","correspondingAuthor":true,"prefix":"","firstName":"Georgios","middleName":"","lastName":"Bakalos","suffix":""},{"id":612093812,"identity":"8c3ebaf4-adff-4a0d-9837-b8156a5cf2c8","order_by":1,"name":"Thomas Buechele","email":"","orcid":"","institution":"Miltenyi Biomedicine (Germany)","correspondingAuthor":false,"prefix":"","firstName":"Thomas","middleName":"","lastName":"Buechele","suffix":""}],"badges":[],"createdAt":"2026-03-22 20:54:07","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9193637/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9193637/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":106865157,"identity":"852a3a35-fe6a-4baf-90a4-baa971abb62e","added_by":"auto","created_at":"2026-04-14 08:57:25","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":120929,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eConceptual overview of MAVI architecture.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eThe left panel shows the Positive ROI pathway (Equation 1): four benefit components (αMA × ΔRev, CA, ΔrNPV, SOV) are summed, reduced by an overlap adjustment (OverlapAdj), discounted to present value, and divided by PV(CostMA) to yield the benefit–cost ratio. The right panel shows the Reverse ROI pathway (Equation 2): six loss categories, each weighted by βMA,k × P(Loss) × Impact × Duration, are summed, reduced by a correlation adjustment (CorrAdj), discounted, and divided by PV(Foregone MA Budget). Colour coding indicates data maturity: green = established methodology; blue = benchmark-populated; amber = requires calibration. A mutual-exclusivity boundary (Table 5) ensures no value stream is counted in both panels. Both outputs converge in the two-dimensional value map (Figure 2). MAVI, Medical Affairs Value Index; PV, present value; BCR, benefit–cost ratio; MA, Medical Affairs.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-9193637/v1/292ed77dd529915807a307f5.png"},{"id":106865175,"identity":"a3499de0-5184-47de-8cb1-51b0ac5c7ae5","added_by":"auto","created_at":"2026-04-14 08:57:31","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":167419,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eMAVI value map: pre-AI and AI-augmented scenarios\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003ePositive ROI (BCR) is plotted on the horizontal axis and Reverse ROI (βMA-adjusted) on the vertical axis. Point A represents the pre-AI baseline under evidence-based parameters (αMA = 0.30; BCR = 1.08×; Reverse ROI = 0.37×). Point B represents the AI-augmented scenario at approximately constant total MA cost (BCR = 1.22×; Reverse ROI = 0.40×). The arrow indicates the shift achieved through AI-driven output multiplication. Dashed lines at 1.0× indicate breakeven on each dimension. Shaded ellipses represent sensitivity ranges under parameter variation. Full year-by-year calculations are provided in ESM File S1. A conservative-assumptions variant of the value map is provided in ESM File S3 (Figure S1). BCR, benefit–cost ratio; MA, Medical Affairs.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-9193637/v1/0ecbf6f3759bbe88545f2db5.png"},{"id":106965975,"identity":"25054808-a730-4da4-9415-b868d43a28f4","added_by":"auto","created_at":"2026-04-15 09:57:59","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1532496,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9193637/v1/6fd45f7c-e5a6-46a5-bb42-38aa2d749bc3.pdf"},{"id":106961247,"identity":"74938588-bffb-42e5-b324-27cb93d85441","added_by":"auto","created_at":"2026-04-15 09:24:50","extension":"xlsx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":24200,"visible":true,"origin":"","legend":"","description":"","filename":"MAVIESMS1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-9193637/v1/c0cb171f96dd8016730fa963.xlsx"},{"id":106865172,"identity":"764ca33a-2c25-48e9-8f76-15c70892afff","added_by":"auto","created_at":"2026-04-14 08:57:31","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":27535,"visible":true,"origin":"","legend":"","description":"","filename":"MAVIESMS3.docx","url":"https://assets-eu.researchsquare.com/files/rs-9193637/v1/e9757d482a4df87fbf881926.docx"},{"id":106865221,"identity":"98abc9cb-007a-4c84-9dc4-963cdb8fe613","added_by":"auto","created_at":"2026-04-14 08:57:44","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":23783,"visible":true,"origin":"","legend":"","description":"","filename":"MAVIESMS2.docx","url":"https://assets-eu.researchsquare.com/files/rs-9193637/v1/793146d159705aac72319984.docx"}],"financialInterests":"Competing interest reported. G.B. is employed by Helsinn Healthcare SA, Switzerland; \nT.B. is employed by Miltenyi Biomedicine, Germany.","formattedTitle":"Quantifying the Return on Investment of Medical Affairs in the AI Era","fulltext":[{"header":"KEY POINTS","content":"\u003cp\u003e1) Medical Affairs (MA) lacks a standardised financial framework for measuring return on investment, a gap now compounded by AI\u0026rsquo;s capacity to amplify MA output without proportionally reducing cost.\u003c/p\u003e\u003cp\u003e2) The Medical Affairs Value Index (MAVI) integrates a Positive ROI model (benefit\u0026ndash;cost ratio capturing value creation) with a Reverse ROI model (probability-weighted enterprise losses prevented), reported as a two-dimensional value map.\u003c/p\u003e\u003cp\u003e3) Attribution coefficients on both upside (αMA) and downside (βMA) govern how value is assigned to MA relative to co-producing functions, a requirement that becomes more demanding when AI is a co-producer.\u003c/p\u003e\u003cp\u003e4) AI is unlikely to materially reduce the cost of MA operations; it reallocates spending from manual tasks to AI-augmented workflows while introducing new cost lines for tools and governance. The cost denominator remains approximately flat while the benefit numerator grows. MAVI provides the framework to quantify this shift and assess its impact on both value creation and risk prevention.\u003c/p\u003e\u003cp\u003e5) Without a financial measurement framework, neither the baseline value MA creates nor the additional value AI enables is visible to decision-makers. MAVI provides the architecture to quantify both\u003c/p\u003e"},{"header":"1. INTRODUCTION","content":"\u003cdiv id=\"Sec2\" class=\"Section2\"\u003e \u003ch2\u003e1.1 AI IN MEDICAL AFFAIRS\u003c/h2\u003e \u003cp\u003eArtificial intelligence is entering pharmaceutical operations across the product lifecycle. The McKinsey Global Institute estimates that generative AI could generate US\u003cspan\u003e$\u003c/span\u003e60\u0026ndash;110\u0026nbsp;billion annually in economic value for the pharmaceutical and medical-product industries, with applications spanning research, clinical development, operations, commercial, and medical affairs.[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e] AI adoption is already visible in adjacent functions: a landscape analysis of FDA regulatory submissions documented a growing number of drug and biologic applications incorporating AI and machine learning components between 2016 and 2021.[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e] Within MA, AI is being applied across medical writing and publications, evidence generation and real-world evidence analysis, medical information inquiry handling, medical education, scientific communications, and field medical operations including MSL preparation and insight capture.[\u003cspan additionalcitationids=\"CR4\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eTaken together, these developments are shifting the operational profile of MA functions, changing which tasks require human effort, how much output a given team can produce, and what skill mix is needed. Without a framework for measuring the financial impact of these changes, organisations lack the basis for informed decisions about MA investment, workforce planning, and the governance of AI-augmented workflows.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e1.2 THE MEASUREMENT GAP\u003c/h2\u003e \u003cp\u003eMA has been recognised as a strategic function alongside R\u0026amp;D and Commercial,[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e] with growing attention to the value it brings to patients and to the organisations that invest in it.[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e] Unlike its peer functions, however, MA has never had a standardised financial model for demonstrating its contribution. For example, Commercial can quantify value through revenue attribution and market share analysis, while R\u0026amp;D uses net present value (NPV) and probability-of-success models to justify investment decisions. MA, by contrast, typically reports activity-based metrics, like MSL interaction counts, publication numbers, medical information response volumes, or qualitative narratives without financial granularity.[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eThis asymmetry becomes more consequential as AI enters MA operations. AI automates the tasks that produce countable output (literature reviews, response documents, evidence summaries) but the strategic contributions that drive organisational value (scientific judgement, regulatory credibility, evidence strategy) remain unmeasured.[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e] The result is that AI increases what MA produces without providing a mechanism to quantify what that increased production is worth. Without a financial framework, the increased output cannot be translated into the financial terms required for resource-allocation decisions\u003c/p\u003e \u003cp\u003eIndustry bodies have developed increasingly sophisticated approaches,[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e] including multi-stakeholder value models[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]and validated workforce instruments.[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e] None, however, provides a financial framework for quantifying the value MA creates, nor a basis for understanding how AI may amplify or transform that value. Six gaps persist: (1) incomplete value capture, particularly the cost of underinvestment; (2) attribution challenges in co-created value; (3) misalignment between multi-year returns and annual budget cycles; (4) insufficient financial rigour; (5) sector-specific variation; and (6) AI-readiness.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e1.3 OBJECTIVE\u003c/h2\u003e \u003cp\u003eThis paper presents the Medical Affairs Value Index (MAVI), a tool developed to quantify MA\u0026rsquo;s return on investment and to explore how AI is impacting its ROI. The central proposition is that AI can increase MA\u0026rsquo;s Positive ROI, not by reducing cost, but by increasing the output the same cost produces, while strengthening the risk-governance case. However, this additional value is not captured without a measurement framework that governs attribution. MAVI provides both the measurement architecture and the attribution governance.\u003c/p\u003e \u003cp\u003eMAVI is deliberately structured as a multi-component index rather than a single score: its principal output is a two-dimensional value map that allows organisations to assess MA\u0026rsquo;s contribution on both value creation and risk prevention simultaneously. The framework is conceptual and designed for progressive organisational calibration; its immediate purpose is to provide a structured basis for investment discussions, not a definitive decision threshold. Although illustrated with pharmaceutical examples, MAVI is intended for MA functions across biopharma and MedTech.\u003c/p\u003e \u003c/div\u003e"},{"header":"2. MAVI: MEASURING MA ROI","content":"\u003cp\u003eFigure \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e provides a conceptual overview of the MAVI framework. The notation glossary (Table \u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e) defines all symbols. The evidence search methodology is documented in ESM File S2.\u003c/p\u003e\n\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003cbr\u003e\u003c/div\u003e\u0026nbsp;\u0026nbsp;\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eMAVI framework notation glossary\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eSymbol\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eName\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eDefinition\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eUnits / Range\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003e\u0026alpha;MA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eUpside attribution\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eFraction of revenue value attributable to MA; set via governed process\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e0\u0026thinsp;\u0026lt;\u0026thinsp;\u0026alpha;MA\u0026thinsp;\u0026le;\u0026thinsp;1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003e\u0026beta;MA,k\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eDownside attribution\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eFraction of loss category k preventable by MA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e0\u0026thinsp;\u0026lt;\u0026thinsp;\u0026beta;MA\u0026thinsp;\u0026le;\u0026thinsp;1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003e\u0026Delta;Revt\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eIncremental revenue\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eRevenue acceleration vs. counterfactual in period t\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eUS\u003cspan\u003e$\u003c/span\u003e per period\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eCAt\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eCost avoided\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eExpenditures averted through MA in period t\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eUS\u003cspan\u003e$\u003c/span\u003e per period\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003e\u0026Delta;rNPVt\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eRisk-adjusted NPV\u0026Delta;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eChange in risk-adjusted NPV from MA de-risking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eUS\u003cspan\u003e$\u003c/span\u003e per period\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eSOVt\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eStrategic option value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eOption premium from MA-enabled future choices\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eUS\u003cspan\u003e$\u003c/span\u003e per period\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eOverlapAdjt\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eOverlap adjustment\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eDeduction for double-counted value (Eq.\u0026nbsp;1a)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eUS\u003cspan\u003e$\u003c/span\u003e per period\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003er\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eDiscount rate\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eOrganisation\u0026rsquo;s WACC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e8\u0026ndash;12%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eEvaluation horizon\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003ePeriods for discounted analysis\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e3\u0026ndash;5 years\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eP(Lossk,t)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eLoss probability\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eEvent probability absent adequate MA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e0\u0026ndash;1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eCorrAdj\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eCorrelation adj.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eDeduction for non-independence (Eq.\u0026nbsp;2a; PV-denominated; applied after discounting)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eUS\u003cspan\u003e$\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\n \u003ch2\u003e2.1 Positive ROI model\u003c/h2\u003e\n \u003cp\u003eThe Positive ROI captures measurable financial value attributable to MA over a T-period evaluation horizon using discounted cash-flow methodology, operationalised as a benefit\u0026ndash;cost ratio (BCR\u0026thinsp;=\u0026thinsp;PV(Benefits)/PV(Cost)). Net ROI\u0026thinsp;=\u0026thinsp;BCR\u0026thinsp;\u0026minus;\u0026thinsp;1 is reported alongside all BCR values for interpretability.\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eEquation 1a\u003c/strong\u003e, Present value of MA-attributable benefits:\u003c/p\u003e\n \u003cp\u003e\u003cem\u003ePV(BenefitsMA) = \u0026Sigma;Tt\u0026thinsp;=\u0026thinsp;1 [(\u0026alpha;MA\u0026thinsp;\u0026times;\u0026thinsp;\u0026Delta;Revt)\u0026thinsp;+\u0026thinsp;CAt\u0026thinsp;+\u0026thinsp;\u0026Delta;rNPVt\u0026thinsp;+\u0026thinsp;SOVt\u0026thinsp;\u0026minus;\u0026thinsp;OverlapAdjt] \u0026times; (1\u0026thinsp;+\u0026thinsp;r)\u0026thinsp;\u0026minus;\u0026thinsp;t\u003c/em\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eEquation 1b\u003c/strong\u003e, Present value of MA investment:\u003c/p\u003e\n \u003cp\u003e\u003cem\u003ePV(CostMA) = \u0026Sigma;Tt\u0026thinsp;=\u0026thinsp;1 [MA_Costt] \u0026times; (1\u0026thinsp;+\u0026thinsp;r)\u0026thinsp;\u0026minus;\u0026thinsp;t\u003c/em\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eEquation 1c\u003c/strong\u003e, Positive ROI:\u003c/p\u003e\n \u003cp\u003e\u003cem\u003ePositive ROI\u0026thinsp;=\u0026thinsp;PV(BenefitsMA) / PV(CostMA)\u003c/em\u003e\u003c/p\u003e\n \u003cp\u003eEach component (Table \u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e) represents a distinct value channel. The coefficient \u0026alpha;MA (0\u0026thinsp;\u0026lt;\u0026thinsp;\u0026alpha;MA\u0026thinsp;\u0026le;\u0026thinsp;1) is applied to \u0026Delta;Revenue, where cross-functional co-creation with Commercial is most pronounced; \u0026Delta;Rev should be operationalised as incremental contribution margin, not gross revenue. The remaining components (CA (cost avoided), \u0026Delta;rNPV (risk-adjusted NPV change from pipeline de-risking), and SOV (strategic option value)) capture only the MA-attributable portion by construction, subject to the documentation requirements in Box 1. Table \u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e proposes indicative \u0026alpha;MA ranges by activity type.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\u0026nbsp;\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003ePositive ROI component definitions and benchmarks\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"4\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eComponent\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eDefinition\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eOperationalisation\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eKey Benchmark\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003e\u0026alpha;MA\u0026thinsp;\u0026times;\u0026thinsp;\u0026Delta;Revt\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eIncremental net sales (MA-attributed fraction)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eComparator launch matching, time-to-adoption\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eIQVIA [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e] (grey; vendor; illustrative)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eCAt\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eMA-attributable expenditures prevented (Box 1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eHistorical cost tracking; MA ownership documented\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e~\u003cspan\u003e$\u003c/span\u003e535K/Phase III amendment [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003e\u0026Delta;rNPVt\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eMA-attributable change in asset value (Box 1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e\u0026Delta;PoS \u0026times; NPV \u0026times; discount; documented assumptions\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003ePoS data [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eSOVt\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eMA-attributable option premium (Box 1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eBinomial lattice at decision gates; incremental\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eReal options methodology\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eOverlapAdjt\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eDeduction for double-counted value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eMapping rules or 15\u0026ndash;25% haircut\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eOrganisation-specific\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003ctfoot\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\"\u003e\u003cem\u003eReference numbers correspond to the main manuscript reference list.\u003c/em\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tfoot\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cdiv class=\"gridtable\"\u003e\n \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003cp\u003e\u003cstrong\u003eBOX 1. Operational Rules for \u0026quot;Attributable by Construction\u0026quot;\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eComponents CA, \u0026Delta;rNPV, and SOV bypass \u0026alpha;MA only when each claimed item satisfies all four criteria:\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(i) Documented MA ownership:\u0026nbsp;\u003c/strong\u003ea named MA individual or team led or co-led the activity, evidenced by project records, approval chains, or RACI matrices.\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(ii) Counterfactual plausibility:\u0026nbsp;\u003c/strong\u003emanagement can articulate why the outcome would not have occurred (or would have been materially worse) without MA involvement.\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(iii) Exclusion of shared credit:\u0026nbsp;\u003c/strong\u003eif another function has an equal or greater claim, the item must be resolved by: (a) applying an explicit attribution factor agreed with the co-owning function (analogous to \u0026alpha;MA), (b) splitting the item by documented agreement and assigning each portion to the appropriate component, or (c) excluding it entirely. Items should not be reclassified into a different component category to circumvent shared ownership.\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(iv) Annual audit:\u0026nbsp;\u003c/strong\u003eitems classified as \u0026quot;attributable by construction\u0026quot; are reviewed annually by the cross-functional governance panel (including Finance and Compliance). Items failing any criterion are reclassified.\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(v) AI-generated outputs (added in v2):\u0026nbsp;\u003c/strong\u003ewhen AI tools contribute to an output claimed under CA, \u0026Delta;rNPV, or SOV, the documentation must specify: (a) which AI tool was used and for which task; (b) the nature and extent of human review and modification; and (c) whether the strategic direction (which questions to answer, which populations to study, which endpoints to prioritise) was set by MA. AI-assisted execution does not transfer attribution to IT or data science if MA directed the strategy and validated the output.\u003c/p\u003e\n \u003cp\u003e\u003cem\u003eExample of a failed Box 1 item: an MSL-facilitated investigator-initiated trial application where Commercial independently contacted the same investigator is classified as \u0026quot;shared credit\u0026quot; under criterion (iii) and receives an explicit attribution factor agreed between MA and Commercial, rather than claiming full CA attribution.\u003c/em\u003e\u003c/p\u003e\n \u003cp\u003e\u003cem\u003eItems that do not meet these criteria should either receive an explicit attribution factor or be excluded from the calculation.\u003c/em\u003e\u003c/p\u003e\u0026nbsp;\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eIndicative \u0026alpha;MA ranges by MA activity type\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"3\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eMA Activity Category\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eIndicative \u0026alpha;MA\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eRationale\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eKOL engagement and medical education\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\n \u003cp\u003e0.15\u0026ndash;0.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eSubstantial Commercial co-contribution\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eProtocol design and trial feasibility\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\n \u003cp\u003e0.20\u0026ndash;0.40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eMA contribution more isolable; R\u0026amp;D co-creation\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eRWE generation and evidence planning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\n \u003cp\u003e0.30\u0026ndash;0.50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eMA typically leads; Market Access co-benefits\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003ePost-market surveillance and safety\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\n \u003cp\u003e0.30\u0026ndash;0.50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eRelatively direct causal link to regulatory outcomes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eIIT-driven label expansion\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\n \u003cp\u003e0.25\u0026ndash;0.40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eMA initiates; R\u0026amp;D executes submission\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eMedical information and congress\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\n \u003cp\u003e0.10\u0026ndash;0.20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eDiffuse impact; attribution difficult\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003ctfoot\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"3\"\u003e\u003cem\u003eExpert-informed starting points for cross-functional calibration, not empirically validated parameters.\u003c/em\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tfoot\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\n \u003ch2\u003e2.2 Reverse ROI model\u003c/h2\u003e\n \u003cp\u003eThe Reverse ROI quantifies expected enterprise losses when MA is eliminated or materially under-resourced, framing MA investment as risk governance.\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eEquation 2a\u003c/strong\u003e, Present value of MA-preventable expected losses:\u003c/p\u003e\n \u003cp\u003ePV(LossesNoMA) = \u0026Sigma;k \u0026Sigma;Tt\u0026thinsp;=\u0026thinsp;1 [\u0026beta;MA,k \u0026times; P(Lossk,t) \u0026times; Impactk,t \u0026times; Durationk,t] \u0026times; (1\u0026thinsp;+\u0026thinsp;r)\u0026thinsp;\u0026minus;\u0026thinsp;t\u0026thinsp;\u0026minus;\u0026thinsp;CorrAdj\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eEquation 2b\u003c/strong\u003e, Present value of foregone MA budget:\u003c/p\u003e\n \u003cp\u003ePV(Foregone MA Budget) = \u0026Sigma;Tt\u0026thinsp;=\u0026thinsp;1 [MA_Costt] \u0026times; (1\u0026thinsp;+\u0026thinsp;r)\u0026thinsp;\u0026minus;\u0026thinsp;t\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eEquation 2c\u003c/strong\u003e, Reverse ROI:\u003c/p\u003e\n \u003cp\u003e\u003cem\u003eReverse ROI\u0026thinsp;=\u0026thinsp;PV(LossesNoMA) / PV(Foregone MA Budget)\u003c/em\u003e\u003c/p\u003e\n \u003cp\u003eThe downside attribution coefficient \u0026beta;MA,k (0\u0026thinsp;\u0026lt;\u0026thinsp;\u0026beta;MA,k\u0026thinsp;\u0026le;\u0026thinsp;1) represents the fraction of each loss category that MA is credibly positioned to prevent. Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e Panel B proposes indicative \u0026beta;MA ranges. Minimum reporting standards require documentation of probability sources, impact definitions, duration logic, and sensitivity analysis. The probabilities P(Lossk,t) represent industry-level event frequencies, not conditional probabilities specific to MA underinvestment, which is a recognised limitation. Box 2 describes staged implementation for probability calibration. The correlation safeguard (CorrAdj) prevents double counting of co-occurring events.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\u0026nbsp;\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eEvidence credibility and Reverse ROI loss categories Panel A: Evidence source credibility\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"5\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eBenchmark\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eSource Type\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eCredibility\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eKey Limitation\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c5\"\u003e\n \u003cp\u003eRef\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eProtocol amendment cost (~\u003cspan\u003e$\u003c/span\u003e535K)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003ePeer-reviewed\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eHigh\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eUS-centric sample\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\n \u003cp\u003e[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eDelay day value (\u003cspan\u003e$\u003c/span\u003e500K\u0026ndash;\u003cspan\u003e$\u003c/span\u003e800K)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003ePeer-reviewed\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eHigh\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eMedian varies by TA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\n \u003cp\u003e[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eClinical PoS rates\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003ePeer-reviewed\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eHigh\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eHistorical; varies by TA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\n \u003cp\u003e[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eR\u0026amp;D costs (\u003cspan\u003e$\u003c/span\u003e2.6B capitalised)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003ePeer-reviewed\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eHigh\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eIndustry-funded study\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\n \u003cp\u003e[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eLaunch benchmarks\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eGrey\u0026mdash;vendor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eLow\u0026ndash;Medium\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003ePotential commercial bias; illustrative only\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\n \u003cp\u003e[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eRegulatory sanctions (\u003cspan\u003e$\u003c/span\u003e1\u0026ndash;10M)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eIndustry estimates\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eLow\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eWide range; no single robust source; illustrative\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eProduct recall costs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003ePress / regulatory\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eMedium\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eSingle-case examples\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\n \u003cp\u003e[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eEU-MDR penalties\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eRegulation / legal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eHigh (structure)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eMonetary exposure jurisdiction-specific; applies to MedTech/combination-product MA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\n \u003cp\u003e[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003ctfoot\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"5\"\u003e\u003cstrong\u003ePanel B: Reverse ROI loss categories with indicative \u0026beta;MA\u003c/strong\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tfoot\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003e\u003cstrong\u003eBOX 2. Implementation Upgrade Path for Probability Modelling (\u0026Delta;P Calibration)\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eStage 1, Baseline frequency + conservative \u0026beta;MA\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eUse unconditional industry-level event frequencies from published sources (Table 4). Apply conservative \u0026beta;MA values from Table 4 Panel B. This is the lowest-data-requirement entry point and yields a pragmatic lower bound.\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eStage 2, Internal incident log + empirical priors\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eRetrospectively analyse the organisation\u0026rsquo;s own loss events over 3\u0026ndash;5 years. Use these internal frequencies as Bayesian updates to the Stage 1 priors. For example, if the company has experienced two protocol-amendment events attributable to insufficient MA input in five years, this empirical rate (2/5 = 0.40/year) replaces the industry benchmark for that category.\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eStage 3, Capability-specific \u0026Delta;P estimation\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eModel partial under-resourcing rather than binary MA removal. Estimate the relative risk reduction (\u0026Delta;P) associated with specific MA capabilities (e.g., pharmacovigilance staffing level, evidence-planning maturity score). This allows modelling of investment increments (e.g., \u0026quot;+5 MSLs reduces launch delay probability by X%\u0026quot;) rather than all-or-nothing scenarios. Structured expert elicitation (Delphi panels with Finance and Compliance) is the recommended calibration method at this stage.\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eAI-era note (added in v2):\u0026nbsp;\u003c/strong\u003eAI adoption may accelerate the transition from Stage 1 to Stage 2 by enabling systematic tracking of AI-augmented outputs and their outcomes. However, AI also introduces new loss categories (e.g., regulatory findings from inadequately reviewed AI-generated content) that may not have Stage 1 industry benchmarks. Organisations should document these emerging categories with conservative probability estimates until empirical data are available.\u003c/p\u003e\u003cem\u003eOrganisations should document which Stage they are operating at and disclose this alongside reported Reverse ROI values. Migration from Stage 1 to Stage 3 is expected over 2\u0026ndash;3 annual cycles.\u003c/em\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\n \u003ch2\u003e2.3 The two-dimensional value map\u003c/h2\u003e\n \u003cp\u003eMAVI is reported as a two-dimensional value map (Fig. \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e), with Positive ROI (BCR) on the horizontal axis and Reverse ROI (\u0026beta;MA-adjusted) on the vertical axis. This format preserves the distinct information each dimension carries and avoids combining non-commensurable ratios into a single score. Dashed reference lines at 1.0\u0026times; on both axes mark breakeven: a Positive ROI above 1.0\u0026times; indicates that MA-attributable benefits exceed MA cost; a Reverse ROI above 1.0\u0026times; indicates that expected losses prevented exceed the MA budget that would be foregone.\u003c/p\u003e\n \u003cp\u003eOrganisations positioned below breakeven on both axes face a diagnostic question: does the result reflect underperformance, or conservative measurement? Those above breakeven on risk prevention but below on value creation have a quantified case for MA as a governance function. The value map thus supports investment discussions grounded in evidence rather than in assumptions. To prevent double counting, each value stream is assigned to a single dimension (Table \u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\n\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eCross-panel mutual exclusivity: mapping rules and edge cases\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"5\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eValue Stream\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eHome Panel\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eExcluded From\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eEdge Case\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c5\"\u003e\n \u003cp\u003eResolution\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eProtocol amendments avoided\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003ePositive ROI (CA)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eReverse ROI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eResidual amendment risk beyond MA input\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\n \u003cp\u003eCap residual at 25% in Reverse ROI; document\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eLaunch delay prevention\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eReverse ROI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003ePositive ROI (\u0026Delta;Rev)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eFaster launch\u0026thinsp;=\u0026thinsp;revenue acceleration?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\n \u003cp\u003eRevenue acceleration goes to \u0026Delta;Rev; delay prevention stays in Reverse ROI as distinct pathway\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eRegulatory compliance costs avoided\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003ePositive ROI (CA)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eReverse ROI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eSanctions from non-compliance?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\n \u003cp\u003eCompliance costs avoided \u0026rarr; CA; sanctions risk (distinct from CA) \u0026rarr; Reverse ROI with \u0026beta;MA\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eCompetitive displacement\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eReverse ROI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003ePositive ROI (\u0026Delta;Rev)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eEvidence differentiation drives revenue?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\n \u003cp\u003eRevenue from evidence \u0026rarr; \u0026Delta;Rev; chronic competitor gap (absent MA) \u0026rarr; Reverse ROI\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eProduct recall prevention\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eReverse ROI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003ePositive ROI (CA)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eSignal detection cost savings?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\n \u003cp\u003eOperational savings \u0026rarr; CA; catastrophic recall risk \u0026rarr; Reverse ROI\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eEvidence-enabled label expansion\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003ePositive ROI (\u0026Delta;rNPV)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eReverse ROI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eMissed expansion\u0026thinsp;=\u0026thinsp;lost revenue?\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c5\"\u003e\n \u003cp\u003eNPV appreciation \u0026rarr; \u0026Delta;rNPV; forfeited option excluded (subsumed in \u0026Delta;rNPV counterfactual)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003ctfoot\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"5\"\u003e\u003cem\u003eWhen both panels have a legitimate claim, the upstream (more direct) pathway takes the home assignment. Downstream (indirect) pathways are capped at 25\u0026ndash;50% and documented. The overriding principle: no dollar of claimed value appears twice in the two-dimensional value map.\u003c/em\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tfoot\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cdiv class=\"gridtable\"\u003e\n \u003c/div\u003e\u0026nbsp;\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eIllustrative scenario: pre-AI and AI-augmented (mid-size pharmaceutical company, T\u0026thinsp;=\u0026thinsp;5 years, r\u0026thinsp;=\u0026thinsp;10%)\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"4\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eParameter\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003ePre-AI\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eAI-Era\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eSource / Basis\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003ePortfolio revenue\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eUS\u003cspan\u003e$\u003c/span\u003e1B\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eUS\u003cspan\u003e$\u003c/span\u003e1B\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eAssumed\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eAnnual MA investment\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eUS\u003cspan\u003e$\u003c/span\u003e18M (60 FTEs)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eUS\u003cspan\u003e$\u003c/span\u003e18M (55 FTEs + \u003cspan\u003e$\u003c/span\u003e2.5M AI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eReallocation, not reduction\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eTime horizon (T)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e5 years\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e5 years\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eDesign choice\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eDiscount rate (r)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e10%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e10%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003ePharma WACC midpoint\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003e\u0026alpha;MA (\u0026Delta;Rev)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e0.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e0.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eEvidence-based; Table \u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e range\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eOverlapAdj\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e15%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e25%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eHigher: more co-producing functions\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003ePV(BenefitsMA)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e\u003cspan\u003e$\u003c/span\u003e73.5M\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e~\u003cspan\u003e$\u003c/span\u003e83M\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eEq.\u0026nbsp;1a\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003ePV(CostMA)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e\u003cspan\u003e$\u003c/span\u003e68.2M\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e\u003cspan\u003e$\u003c/span\u003e68.2M\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eEq.\u0026nbsp;1b\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003e\u003cstrong\u003ePositive ROI (BCR)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e\u003cstrong\u003e1.08\u0026times;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e\u003cstrong\u003e~\u0026thinsp;1.22\u0026times;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eEq.\u0026nbsp;1c\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eNet ROI (BCR\u0026thinsp;\u0026minus;\u0026thinsp;1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e+\u0026thinsp;0.08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e+\u0026thinsp;0.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eWeighted \u0026beta;MA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e0.50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e0.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eResidual risk more human-dependent\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003ePV(Gross Expected Losses, 5\u0026nbsp;year)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e\u003cspan\u003e$\u003c/span\u003e50M (\u003cspan\u003e$\u003c/span\u003e35\u0026ndash;78M)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e~\u003cspan\u003e$\u003c/span\u003e47M\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eEq.\u0026nbsp;2a before \u0026beta;MA; AI reduces P(Loss)\u0026thinsp;~\u0026thinsp;15%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003ePV(\u0026beta;MA-adjusted Losses, 5\u0026nbsp;year)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e\u003cspan\u003e$\u003c/span\u003e25M\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e~\u003cspan\u003e$\u003c/span\u003e27M\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eEq.\u0026nbsp;2a\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003ePV(Foregone MA Budget, 5\u0026nbsp;year)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e\u003cspan\u003e$\u003c/span\u003e68.2M\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e\u003cspan\u003e$\u003c/span\u003e68.2M\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eEq.\u0026nbsp;2b (matches Eq.\u0026nbsp;1b)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003e\u003cstrong\u003eReverse ROI (gross)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.73\u0026times;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e\u003cstrong\u003e~\u0026thinsp;0.69\u0026times;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eEq.\u0026nbsp;2c\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003e\u003cstrong\u003eReverse ROI (\u0026beta;MA-adjusted)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.37\u0026times;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e\u003cstrong\u003e~\u0026thinsp;0.40\u0026times;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eEq.\u0026nbsp;2c\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eValue map position (Fig. \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003ePoint A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003ePoint B\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eAbove breakeven on Positive ROI; Reverse ROI stable\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eProbability stage (Box 2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eStage 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eStage 1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eBaseline industry frequencies\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003ctfoot\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\"\u003e\u003cem\u003ePositive ROI (BCR) exceeds 1\u0026times; under evidence-based parameters (\u0026alpha;MA\u0026thinsp;=\u0026thinsp;0.30). Under conservative assumptions (\u0026alpha;MA\u0026thinsp;=\u0026thinsp;0.25, BCR\u0026thinsp;=\u0026thinsp;0.87\u0026times;), see ESM File S3. AI-era estimates reflect plausible assumptions about output multiplication at constant cost; they are illustrative, not validated. Benchmark inputs sourced from\u003c/em\u003e Table \u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e \u003cem\u003ePanel A. Gross Reverse ROI represents total enterprise risk exposure; \u0026beta;MA-adjusted represents the MA-preventable portion\u003c/em\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tfoot\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003c/div\u003e"},{"header":"3. HOW AI MODIFIES THE ROI","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.1 THE COST DENOMINATOR: REALLOCATION RATHER THAN REDUCTION\u003c/h2\u003e \u003cp\u003eAI adoption in MA does not materially reduce total function cost.[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e] Increased output volume requires proportionally greater expert review capacity, and new cost lines emerge: AI platform licences, validation infrastructure, and governance overhead. FTE profiles shift (fewer staff perform manual tasks, more serve as AI-literate medical scientists and senior reviewers) but aggregate cost remains broadly stable.\u003c/p\u003e \u003cp\u003eUnder a representative scenario, a pre-AI MA function of 60 FTEs at US\u003cspan\u003e$\u003c/span\u003e18\u0026nbsp;million per year transitions to 55 FTEs plus US\u003cspan\u003e$\u003c/span\u003e2\u0026ndash;3\u0026nbsp;million in AI tools and infrastructure, yielding a comparable annual cost of approximately US\u003cspan\u003e$\u003c/span\u003e18\u0026nbsp;million. The PV(CostMA) denominator in Equations 1c and 2c remains approximately US\u003cspan\u003e$\u003c/span\u003e68.2M over five years at a 10% discount rate. Any improvement in ROI must therefore come from the numerator.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.2 THE BENEFIT NUMERATOR: OUTPUT MULTIPLICATION AT CONSTANT COST\u003c/h2\u003e \u003cp\u003eAI increases what MA produces for the same investment across its principal activities (Table\u0026nbsp;7).\u003c/p\u003e \u003cp\u003e \u003cb\u003eMedical writing and publications.\u003c/b\u003e AI-assisted drafting enables higher publication output, faster turnaround on evidence summaries, and broader congress presence from the same editorial team. The resulting gains affect CA (cost avoided through faster medical information responses and fewer compliance gaps from delayed answers) and SOV (more evidence assets available at key decision gates).\u003c/p\u003e \u003cp\u003e \u003cb\u003eEvidence generation.\u003c/b\u003e AI-accelerated real-world evidence analysis compresses timelines for post-market studies, supports faster label-expansion dossiers, and increases the volume of evidence packages for health technology assessment submissions. This directly affects ΔrNPV: more evidence generated faster de-risks pipeline assets and advances the revenue they generate.\u003c/p\u003e \u003cp\u003e \u003cb\u003eMedical information.\u003c/b\u003e AI-enabled triage and response drafting reduce response times while maintaining accuracy under mandatory medical review[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e] lowering the probability of compliance breaches (a CA component).\u003c/p\u003e \u003cp\u003e \u003cb\u003eMSL interactions and field insights.\u003c/b\u003e AI-augmented preparation and structured insight-capture systems improve the quality and traceability of the αMA\u0026thinsp;\u0026times;\u0026thinsp;ΔRevenue component (the pathway from scientific engagement to downstream treatment adoption).\u003c/p\u003e \u003cp\u003e \u003cb\u003eMedical education and scientific communications.\u003c/b\u003e AI-personalised content extends MA\u0026rsquo;s reach across geographies and stakeholder segments without proportional FTE growth, generating new strategic options (SOV).\u003c/p\u003e \u003cp\u003e \u003cb\u003eEstimated magnitude.\u003c/b\u003e If AI increases the Positive ROI numerator by approximately 10\u0026ndash;15% through cumulative gains across these activities while the denominator remains constant, PV(BenefitsMA) rises from approximately US\u003cspan\u003e$\u003c/span\u003e74M (pre-AI) to approximately US\u003cspan\u003e$\u003c/span\u003e84M. This is partially offset by a higher OverlapAdj (increasing from 15% to 20% as more functions contribute to AI-augmented outcomes). The net result: Positive ROI (BCR) improves from approximately 1.08\u0026times; to approximately 1.22\u0026times;.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.3 THE REVERSE ROI: CONCENTRATION OF HUMAN-DEPENDENT RISK\u003c/h2\u003e \u003cp\u003eOn the risk-prevention side, AI modifies loss probabilities. AI-augmented literature surveillance reduces the probability of missing safety-relevant publications. AI-assisted regulatory document preparation lowers submission delay risk. AI-enabled medical information triage reduces compliance exposure.\u003c/p\u003e \u003cp\u003eAt the same time, AI introduces new risk categories. AI systems in medical contexts exhibit variable error rates depending on governance architecture; systems with retrieval-augmented generation and mandatory human oversight achieve substantially lower error rates than unguided implementations.[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e] The FDA has emphasised that AI outputs intended to support regulatory decision-making require credibility assessment frameworks including human validation.[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e] AI-generated evidence or communications released without adequate medical review could result in regulatory findings, reputational harm, or patient safety concerns (risks with no pre-AI equivalent).\u003c/p\u003e \u003cp\u003eThe net effect on βMA runs against expectations. AI generally reduces the probability of routine adverse events, but the residual risk (the risk AI cannot mitigate without human oversight) concentrates in medical judgement: signal escalation decisions, evidence strategy prioritisation, and the scientific credibility that sustains regulator and payer trust. βMA for these residual categories may therefore increase from 0.50 to 0.55\u0026ndash;0.60 in an AI-augmented environment, even as the total risk pool decreases.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.4 THE MEASUREMENT IMPERATIVE\u003c/h2\u003e \u003cp\u003eThe analysis above points to a structural concern. AI amplifies MA output and the benefit numerator grows, but this improvement is invisible without a financial framework to capture it. MA continues to report the same activity-based metrics it used before AI adoption, even as the value those activities generate has materially increased. The function produces more, but in the absence of a financial framework, the return on AI investment within MA cannot be demonstrated\u003c/p\u003e \u003cp\u003eMAVI addresses this through governed αMA and βMA coefficients, Box 1 documentation requirements, and cross-functional calibration panels. These provide the measurement architecture that converts AI-amplified output into a quantified financial case for continued MA investment.\u003c/p\u003e \u003c/div\u003e"},{"header":"4. ILLUSTRATIVE APPLICATION","content":"\u003cp\u003eThis scenario demonstrates framework mechanics, not validated estimates. All parameters are derived from published benchmarks classified by credibility (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, Panel A).[\u003cspan additionalcitationids=\"CR19 CR20 CR21 CR22 CR23 CR24\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] Low-credibility benchmarks are flagged and should not inform investment decisions.\u003c/p\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e4.1 COMPANY PROFILE\u003c/h2\u003e \u003cp\u003eMid-size pharmaceutical company, US\u003cspan\u003e$\u003c/span\u003e1\u0026nbsp;billion portfolio revenue, three marketed products, two label-expansion candidates. T\u0026thinsp;=\u0026thinsp;5 years, r\u0026thinsp;=\u0026thinsp;10%. Year-by-year calculations are provided in ESM File S1.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e4.2 PRE-AI BASELINE\u003c/h2\u003e \u003cp\u003eAnnual MA investment: US\u003cspan\u003e$\u003c/span\u003e18M (60 FTEs, no AI tools). PV(CostMA)\u0026thinsp;=\u0026thinsp;US\u003cspan\u003e$\u003c/span\u003e68.2M. With evidence-based parameters (αMA\u0026thinsp;=\u0026thinsp;0.30, reflecting documented MA contribution to evidence-mediated adoption; CA including regulatory query avoidance; OverlapAdj\u0026thinsp;=\u0026thinsp;15%), PV(BenefitsMA)\u0026thinsp;=\u0026thinsp;US\u003cspan\u003e$\u003c/span\u003e74M. Positive ROI (BCR)\u0026thinsp;=\u0026thinsp;1.08\u0026times; (Net ROI\u0026thinsp;=\u0026thinsp;+\u0026thinsp;0.08). PV(Gross expected losses)\u0026thinsp;=\u0026thinsp;US\u003cspan\u003e$\u003c/span\u003e50M (\u003cspan\u003e$\u003c/span\u003e35\u0026ndash;78M range). Weighted βMA\u0026thinsp;=\u0026thinsp;0.50. PV(βMA-adjusted losses)\u0026thinsp;=\u0026thinsp;US\u003cspan\u003e$\u003c/span\u003e25M. Reverse ROI (βMA-adjusted)\u0026thinsp;=\u0026thinsp;0.37\u0026times;; Reverse ROI (gross)\u0026thinsp;=\u0026thinsp;0.73\u0026times;. Value map position: above breakeven on Positive ROI; below breakeven on Reverse ROI (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, Point A).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e4.3 AI-ERA SCENARIO\u003c/h2\u003e \u003cp\u003eAnnual MA investment: US\u003cspan\u003e$\u003c/span\u003e18M (55 FTEs\u0026thinsp;+\u0026thinsp;US\u003cspan\u003e$\u003c/span\u003e2.5M AI tools and infrastructure). PV(CostMA)\u0026thinsp;=\u0026thinsp;US\u003cspan\u003e$\u003c/span\u003e68.2M (unchanged). AI-augmented benefits: ΔrNPV\u0026thinsp;+\u0026thinsp;30% (accelerated evidence generation and additional label-expansion dossiers); CA\u0026thinsp;+\u0026thinsp;15% (faster medical information turnaround, fewer protocol amendments through AI-assisted design); SOV doubles (AI-enabled strategic options including adaptive trial designs and predictive analytics); αMA\u0026thinsp;\u0026times;\u0026thinsp;ΔRev\u0026thinsp;+\u0026thinsp;10% (AI-augmented MSL engagement). OverlapAdj increases from 15% to 20%. PV(BenefitsMA)\u0026thinsp;\u0026asymp;\u0026thinsp;US\u003cspan\u003e$\u003c/span\u003e84M. Positive ROI (BCR)\u0026thinsp;\u0026asymp;\u0026thinsp;1.22\u0026times; (Net ROI\u0026thinsp;\u0026asymp;\u0026thinsp;+\u0026thinsp;0.22).\u003c/p\u003e \u003cp\u003eReverse ROI: P(Loss) decreases approximately 15% for routine surveillance categories; βMA increases modestly from 0.50 to 0.60 for residual risk categories. PV(βMA-adjusted losses)\u0026thinsp;\u0026asymp;\u0026thinsp;US\u003cspan\u003e$\u003c/span\u003e27M. Reverse ROI (βMA-adjusted)\u0026thinsp;\u0026asymp;\u0026thinsp;0.40\u0026times;. The Reverse ROI remains broadly stable, confirming that AI\u0026rsquo;s primary impact is on value creation rather than risk prevention. Value map position: Positive ROI increases from 1.08\u0026times; to 1.22\u0026times; while Reverse ROI moves from 0.37\u0026times; to 0.40\u0026times; (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, Point B).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e4.4 INTERPRETATION\u003c/h2\u003e \u003cp\u003eThe illustrative scenario demonstrates three properties of the framework. First, the result is sensitive to attribution governance: with evidence-based parameters (αMA\u0026thinsp;=\u0026thinsp;0.30), Positive ROI exceeds breakeven (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, Point A); under conservative assumptions (αMA\u0026thinsp;=\u0026thinsp;0.25, ESM File S3), it does not. The difference is measurement calibration, not MA performance. Second, the Reverse ROI dimension reveals a contribution that single-dimension analysis would miss: MA prevents expected enterprise losses worth a material fraction of its own budget, providing a risk-governance case that is invisible without two-dimensional measurement. Third, AI amplifies the Positive ROI through output multiplication at constant cost, while the Reverse ROI remains broadly stable (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, arrow from Point A to Point B), a shift that is quantifiable only because the framework captures both dimensions.\"\u003c/p\u003e \u003cp\u003eThis addresses a longstanding perception that MA\u0026rsquo;s ROI is inherently unmeasurable. MAVI challenges this by applying standard corporate finance methodology, discounted cash flow, real options, probability-weighted loss estimation, to MA\u0026rsquo;s specific value channels. The illustrative results suggest that the difficulty has not been that MA lacks ROI, but that it has lacked a framework for quantifying it. ESM File S3 presents a stress test using conservative parameters (αMA\u0026thinsp;=\u0026thinsp;0.25, BCR\u0026thinsp;=\u0026thinsp;0.87\u0026times;), demonstrating that even when Positive ROI falls below breakeven, the Reverse ROI dimension sustains the investment case.\u003c/p\u003e \u003cp\u003eThese figures are illustrative and should not be treated as validated estimates. The underlying organisational question remains whether the enterprise is prepared to accept the unmitigated downside exposure that would follow from materially reducing MA capability. Events such as product recalls (US\u003cspan\u003e$\u003c/span\u003e100M\u0026ndash;US\u003cspan\u003e$\u003c/span\u003e1B+)24 may be low-probability but can be existential for mid-size companies.\u003c/p\u003e \u003cp\u003eA further consideration is progressive deskilling. As AI assumes routine tasks, junior MA professionals have fewer opportunities to develop domain expertise, potentially weakening the human oversight layer on which βMA depends.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. IMPLICATIONS AND PATH FORWARD","content":"\u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e5.1 MAVI IN THE AI ERA\u003c/h2\u003e \u003cp\u003eThe analysis presented here suggests two conclusions. First, the financial case for MA is sensitive to measurement approach: when a structured framework is applied with governed attribution, the result differs materially from what activity-based reporting alone would suggest. Second, AI modifies the Positive ROI through output multiplication at approximately constant cost, while the risk-governance dimension remains stable, but this effect is quantifiable only with a framework that captures both dimensions. MAVI provides that framework. In a pre-AI environment, it offers a structured basis for MA investment discussions. In an AI-augmented environment, it enables organisations to assess whether AI adoption is generating a measurable return\u003c/p\u003e \u003cp\u003eZdon et al. characterised MA as a strategic asset in launch execution, corporate reputation, and innovation.[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] MAVI offers a quantitative approach to evaluating that characterisation, by enabling organisations to assess MA's contribution on both value creation and risk prevention within a single measurement architecture.\u003c/p\u003e \u003cp\u003eMAVI is not a health technology assessment framework. It draws on corporate finance concepts (NPV, real options, attribution) and is intended for senior leadership and finance decision-makers rather than reimbursement authorities.\u003c/p\u003e \u003cp\u003eThe structural challenge MAVI addresses is not unique to MA. Any knowledge-intensive function that co-creates value across organisational boundaries (regulatory affairs, quality assurance, health economics, corporate affairs) faces analogous measurement difficulties that AI is likely to amplify. The two-dimensional approach (positive return plus risk prevention) and the attribution governance architecture presented here may serve as a template for the financial evaluation of such functions more broadly.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003e5.2 COMPLIANCE BOUNDARIES\u003c/h2\u003e \u003cp\u003eThe following are structural requirements: (i) only non-promotional activities are eligible for MAVI quantification; (ii) promotional activities are excluded; (iii) calibration panels for αMA and βMA must include compliance representation; (iv) MAVI should not serve as a basis for MA personnel compensation linked to revenue outcomes; (v) revenue modelling should be evidence-mediated. AI-generated outputs used within MAVI quantification require human validation before being counted as MA-attributable.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003e5.3 LIMITATIONS AND NEXT STEPS\u003c/h2\u003e \u003cp\u003eThe proposed αMA and βMA ranges are expert-informed rather than empirically derived. Prospective calibration through cross-functional Delphi panels would strengthen the evidence base. Several benchmark inputs rely on heterogeneous sources of variable quality (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, Panel A). The framework has not been prospectively validated and does not currently include patient outcomes as a direct value dimension.\u003c/p\u003e \u003cp\u003eThree AI-era priorities emerge. First, organisations should track how AI adoption changes αMA and βMA over time through longitudinal within-company calibration. Second, human-AI attribution protocols are needed: when AI and a medical scientist together produce an evidence package that supports a label expansion, the attribution must be governed and documented. Third, industry collaboration to develop AI-adjusted parameter libraries would reduce adoption burden and improve comparability.\u003c/p\u003e \u003cp\u003eA phased implementation programme, beginning with cost-avoided tracking, then adding ΔrNPV and Reverse ROI, and subsequently integrating AI-adjusted parameters, may better match organisational readiness than simultaneous deployment\u003c/p\u003e \u003c/div\u003e"},{"header":"6. CONCLUSION","content":"\u003cp\u003eMAVI provides a structured financial framework for quantifying the return on investment of Medical Affairs across two dimensions: value creation and risk prevention. Applied with evidence-based parameters, it reveals a Positive ROI above breakeven before AI adoption and demonstrates that AI further amplifies this return through output multiplication at approximately constant cost, while the risk-governance dimension remains stable. The framework addresses a measurement gap that has left MA unable to articulate its financial contribution in terms comparable to those used by R\u0026amp;D and Commercial functions. As AI reshapes pharmaceutical operations, the need for such a framework extends beyond MA to any knowledge-intensive function whose value is co-created, diffuse, and difficult to attribute. Empirical calibration through prospective, cross-functional implementation is the essential next step.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAuthor contributions:\u0026nbsp;\u003c/strong\u003eG.B. Conceptualization, Methodology, Literature Synthesis, Writing Original Draft, Implementation Science Integration, MAVI Framework Development; T.B. Literature Synthesis, Framework Development, Review \u0026amp; Editing; All authors approved the final manuscript and accept responsibility for its content.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflicts of interest:\u0026nbsp;\u003c/strong\u003eG.B. is employed by Helsinn Healthcare SA, Switzerland; T.B. is employed by Miltenyi Biomedicine, Germany. The views expressed are those of the authors and do not necessarily represent the official positions of their affiliated institutions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding:\u003c/strong\u003e This work received no external funding.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval:\u0026nbsp;\u003c/strong\u003eNot applicable (no primary data collection).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability:\u0026nbsp;\u003c/strong\u003eThis article is based on published literature, publicly available regulatory guidance, and expert synthesis. No original data were generated. The year-by-year present-value workbook is provided as ESM File S1. The literature research strategy is provided as ESM File S2.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics and AI Use Statement:\u0026nbsp;\u003c/strong\u003eThis article does not involve human participants or patient data. Generative AI tools were used to assist with literature research, reference formatting and structural editing (Claude, Anthropic). Figure 1, 2 and Supplementary Figure 1 were generated programmatically using Python (matplotlib). All AI-generated content was critically reviewed, verified and substantially modified by the authors. All citations were manually validated for accuracy through database searches and DOI verification. The authors accept full responsibility for the accuracy, scientific integrity, and intellectual content of this work..\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eShah B, Adabala Viswa C, Zurkiya D, Leydon E, Bleys J. Generative AI in the pharmaceutical industry: moving from hype to reality. McKinsey \u0026amp; Company; 2024 Jan 9. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.mckinsey.com/industries/life-sciences/our-insights/generative-ai-in-the-pharmaceutical-industry-moving-from-hype-to-reality\u003c/span\u003e\u003cspan address=\"https://www.mckinsey.com/industries/life-sciences/our-insights/generative-ai-in-the-pharmaceutical-industry-moving-from-hype-to-reality\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Accessed 2026 Mar.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu Q, Huang R, Hsieh J, et al. Landscape analysis of the application of artificial intelligence and machine learning in regulatory submissions for drug development from 2016 to 2021. Clin Pharmacol Ther. 2023;113(4):771\u0026ndash;4. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/cpt.2668\u003c/span\u003e\u003cspan address=\"10.1002/cpt.2668\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBate A, Hobbiger SF. Artificial intelligence, real-world automation and the safety of medicines. Drug Saf. 2021;44(2):125\u0026ndash;32. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s40264-020-01001-7\u003c/span\u003e\u003cspan address=\"10.1007/s40264-020-01001-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUS Food and Drug Administration. Using artificial intelligence \u0026amp; machine learning in the development of drug and biological products: discussion paper. Silver Spring (MD): FDA; 2023 May (revised 2025 Feb). Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.fda.gov/media/167973/download\u003c/span\u003e\u003cspan address=\"https://www.fda.gov/media/167973/download\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Accessed 2026 Mar.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUS Food and Drug Administration. Considerations for the use of artificial intelligence to support regulatory decision-making for drug and biological products: draft guidance for industry. Silver Spring (MD): FDA; 2025 Jan. Federal Register docket FDA-2024-D-4689. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.federalregister.gov/documents/2025/01/07/2024-31542/considerations-for-the-use-of-artificial-intelligence-to-support-regulatory-decision-making-for-drug\u003c/span\u003e\u003cspan address=\"https://www.federalregister.gov/documents/2025/01/07/2024-31542/considerations-for-the-use-of-artificial-intelligence-to-support-regulatory-decision-making-for-drug\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Accessed 2026 Mar.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDarino L, Knepp A, Mills N, Tinkoff D. How pharma can accelerate business impact from advanced analytics. McKinsey \u0026amp; Company; 2018 Jan 8. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.mckinsey.com/industries/life-sciences/our-insights/how-pharma-can-accelerate-business-impact-from-advanced-analytics\u003c/span\u003e\u003cspan address=\"https://www.mckinsey.com/industries/life-sciences/our-insights/how-pharma-can-accelerate-business-impact-from-advanced-analytics\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Accessed 2026 Mar.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlgazy J, Garcia A, Ryan S, Westra A, Zemp A. A vision for medical affairs 2030: five priorities for patient impact. McKinsey \u0026amp; Company; 2023 Oct. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.mckinsey.com/industries/life-sciences/our-insights/a-vision-for-medical-affairs-2030-five-priorities-for-patient-impact\u003c/span\u003e\u003cspan address=\"https://www.mckinsey.com/industries/life-sciences/our-insights/a-vision-for-medical-affairs-2030-five-priorities-for-patient-impact\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Accessed 2026 Mar.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFarrington AD, Fr\u0026oslash;strup AG, Dahl P. The value and deliverables of medical affairs: affiliate perspectives and future expectations. Pharmaceut Med. 2023;37(6):417\u0026ndash;24. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s40290-023-00501-y\u003c/span\u003e\u003cspan address=\"10.1007/s40290-023-00501-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMedical Affairs Professional Society (MAPS). Measuring value and impact in medical affairs: standards and guidance document. Golden (CO): MAPS. 2025. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://medicalaffairs.org/wp-content/uploads/2025/01/UPDATE_1.21.25-STRATEGY_Value-and-Impact-in-Medical-Affairs_SG-pwp.pdf\u003c/span\u003e\u003cspan address=\"https://medicalaffairs.org/wp-content/uploads/2025/01/UPDATE_1.21.25-STRATEGY_Value-and-Impact-in-Medical-Affairs_SG-pwp.pdf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Accessed 2026 Mar.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDyer S, Hyder C, Kraemer J. Challenges of key performance indicators and metrics for measuring Medical Science Liaison performance: insights from a global survey. Pharmacy. 2025;13(2):51. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/pharmacy13020051\u003c/span\u003e\u003cspan address=\"10.3390/pharmacy13020051\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJandhyala R. Development and validation of the Medical Affairs Pharmaceutical Physician Value (MAPPval) Instrument. Pharmaceut Med. 2022;36(1):47\u0026ndash;57. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s40290-021-00413-9\u003c/span\u003e\u003cspan address=\"10.1007/s40290-021-00413-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKaplan RS, Norton DP. The balanced scorecard: translating strategy into action. Boston (MA): Harvard Business School Press; 1996.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKaplan RS, Norton DP. Strategy maps: converting intangible assets into tangible outcomes. Boston (MA): Harvard Business School Press; 2004.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJandhyala R. Development of a definition for medical affairs using the Jandhyala method for observing consensus opinion among medical affairs pharmaceutical physicians. Front Pharmacol. 2022;13:842431. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fphar.2022.842431\u003c/span\u003e\u003cspan address=\"10.3389/fphar.2022.842431\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJandhyala R. Development, validation and implementation of the medical affairs pharmaceutical physician work-related quality of life instrument. Curr Med Res Opin. 2023;39(12):1567\u0026ndash;74. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1080/03007995.2023.2174747\u003c/span\u003e\u003cspan address=\"10.1080/03007995.2023.2174747\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSaleem M, Cesario L, Wilcox L, et al. Evaluating metrics applied to the Medical Science Liaison (MSL) role: a survey-based study of Canadian MSL leaders. Ther Innov Regul Sci. 2021;55(5):954\u0026ndash;65. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s43441-021-00291-y\u003c/span\u003e\u003cspan address=\"10.1007/s43441-021-00291-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLucid Group. Share of Scientific Voice (SoSV): methodology and benchmarking. White Paper. Published 2023. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://wearelucidgroup.com/white-papers/medical-affairs-kpis\u003c/span\u003e\u003cspan address=\"https://wearelucidgroup.com/white-papers/medical-affairs-kpis\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Accessed 2026 Mar.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIQVIA. In pursuit of Medical Launch Excellence: lessons for medical affairs from IQVIA\u0026rsquo;s Launch Excellence research (White paper). Durham (NC): IQVIA; 2023 Nov 30. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.iqvia.com/-/media/iqvia/pdfs/library/white-papers/in-pursuit-of-medical-launch-excellence.pdf\u003c/span\u003e\u003cspan address=\"https://www.iqvia.com/-/media/iqvia/pdfs/library/white-papers/in-pursuit-of-medical-launch-excellence.pdf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Accessed 2026 Mar.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGetz KA, Stergiopoulos S, Marlborough M, Whitehill J, Curran M, Kaitin KI. Quantifying the magnitude and cost of collecting extraneous protocol data. Am J Ther. 2015;22(2):117\u0026ndash;24. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/MJT.0b013e31826fc4aa\u003c/span\u003e\u003cspan address=\"10.1097/MJT.0b013e31826fc4aa\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHay M, Thomas DW, Craighead JL, Economides C, Rosenthal J. Clinical development success rates for investigational drugs. Nat Biotechnol. 2014;32(1):40\u0026ndash;51. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/nbt.2786\u003c/span\u003e\u003cspan address=\"10.1038/nbt.2786\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWong CH, Siah KW, Lo AW. Estimation of clinical trial success rates and related parameters. Biostatistics. 2019;20(2):273\u0026ndash;86. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/biostatistics/kxx069\u003c/span\u003e\u003cspan address=\"10.1093/biostatistics/kxx069\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSmith ZP, DiMasi JA, Getz KA. New estimates on the cost of a delay day in drug development. Ther Innov Regul Sci. 2024;58(5):855\u0026ndash;62. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s43441-024-00667-w\u003c/span\u003e\u003cspan address=\"10.1007/s43441-024-00667-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDiMasi JA, Grabowski HG, Hansen RW. Innovation in the pharmaceutical industry: new estimates of R\u0026amp;D costs. J Health Econ. 2016;47:20\u0026ndash;33. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jhealeco.2016.01.012\u003c/span\u003e\u003cspan address=\"10.1016/j.jhealeco.2016.01.012\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAssociated Press. Philips will pay \u003cspan\u003e$\u003c/span\u003e1.1 billion to resolve US lawsuits over breathing machines that expel debris. AP News. 2024 Apr 29. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://apnews.com/article/philips-sleep-apnea-settlement-recall-safety-fc5cc1509656645d7f5679e0f1b34b85\u003c/span\u003e\u003cspan address=\"https://apnews.com/article/philips-sleep-apnea-settlement-recall-safety-fc5cc1509656645d7f5679e0f1b34b85\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Accessed 2026 Mar.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEuropean Parliament and of the Council. Regulation (EU) 2017/745 of 5 April 2017 on medical devices. OJ L 117, 5.5.2017, pp. 1\u0026ndash;175. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://eur-lex.europa.eu/eli/reg/2017/745/oj/eng\u003c/span\u003e\u003cspan address=\"https://eur-lex.europa.eu/eli/reg/2017/745/oj/eng\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZdon J, Chatgilaou GJ, Henderson D, et al. How can general managers best leverage medical affairs now and in the future? Pharmaceut Med. 2024;38(4):277\u0026ndash;90. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s40290-024-00528-9\u003c/span\u003e\u003cspan address=\"10.1007/s40290-024-00528-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"therapeutic-innovation-and-regulatory-science","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"tirs","sideBox":"Learn more about [Therapeutic Innovation \u0026 Regulatory Science](https://link.springer.com/journal/43441)","snPcode":"43441","submissionUrl":"https://www.editorialmanager.com/tirs/default.aspx","title":"Therapeutic Innovation \u0026 Regulatory Science","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Medical Affairs, artificial intelligence, return on investment, cost of inaction, enterprise risk governance, attribution, pharmaceutical management","lastPublishedDoi":"10.21203/rs.3.rs-9193637/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9193637/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eArtificial intelligence (AI) is entering pharmaceutical Medical Affairs (MA) across medical writing, evidence generation, medical information, medical education, scientific communications, and field medical operations, changing how the function creates and delivers value. Yet MA has never had a standardised financial framework for measuring return on investment (ROI) comparable to those established in R\u0026amp;D and Commercial. Without such a framework, the additional value generated through AI adoption, and whether it justifies the associated investment, cannot be quantified.\u003c/p\u003e\n\u003cp\u003eThis paper presents the Medical Affairs Value Index (MAVI), a tool to quantify MA's ROI and assess how AI modifies it. MAVI integrates a Positive ROI model capturing value creation through revenue acceleration, cost avoidance, pipeline de-risking, and strategic option value, operationalised as a benefit–cost ratio, with a Reverse ROI model that quantifies the enterprise cost of MA underinvestment through probability-weighted loss estimation. Attribution coefficients for both upside (αMA) and downside (βMA) are introduced with governance safeguards. The primary output is a two-dimensional value map.\u003c/p\u003e\n\u003cp\u003eAI is unlikely to materially reduce the cost of MA operations; it reallocates spending from manual activities to AI-augmented workflows, keeping the cost denominator approximately flat while the benefit numerator grows. An illustrative scenario demonstrates that under evidence-based assumptions, MAVI reveals a Positive ROI above breakeven before AI adoption, with AI further amplifying the return. The Reverse ROI, capturing enterprise losses MA helps prevent, remains broadly stable across both scenarios. MAVI provides the financial architecture to quantify both dimensions and track how AI modifies them over time.\u003c/p\u003e","manuscriptTitle":"Quantifying the Return on Investment of Medical Affairs in the AI Era","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-14 08:55:00","doi":"10.21203/rs.3.rs-9193637/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewersInvited","content":"","date":"2026-05-16T20:43:29+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-25T14:24:44+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-24T10:19:13+00:00","index":"","fulltext":""},{"type":"submitted","content":"Therapeutic Innovation \u0026 Regulatory Science","date":"2026-03-22T20:49:57+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"therapeutic-innovation-and-regulatory-science","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"tirs","sideBox":"Learn more about [Therapeutic Innovation \u0026 Regulatory Science](https://link.springer.com/journal/43441)","snPcode":"43441","submissionUrl":"https://www.editorialmanager.com/tirs/default.aspx","title":"Therapeutic Innovation \u0026 Regulatory Science","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"6bb77f22-3e99-4e2d-bb8d-0ba5acd0b309","owner":[],"postedDate":"April 14th, 2026","published":true,"recentEditorialEvents":[{"type":"reviewersInvited","content":"100","date":"2026-05-16T20:43:29+00:00","index":"","fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-16T20:54:08+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-14 08:55:00","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9193637","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9193637","identity":"rs-9193637","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}