In today's high-cost R&D environment, pharma success depends less on cost-cutting and more on evidence-based portfolio decisions, niche-buster strategies, and real-world data-driven indication expansion. [1] The phrase evidence-based decision appears in pharma strategy documents, board presentations, and executive communications with considerable frequency. It appears with considerably less frequency in the actual infrastructure that pharma organisations use to make decisions. There is a gap between the aspiration of evidence-based decision-making and the operational reality: in most pharma organisations, the most consequential decisions, protocol design choices, portfolio investment prioritisation, HTA submission strategy, and payer negotiation positioning, are made with evidence that is assembled manually, sourced inconsistently, and verified inadequately for the consequences those decisions carry.
An evidence based decision pharma is not simply a decision informed by data. It is a decision in which every claim that influenced the decision can be traced to a specific, verified primary source, the provenance of every piece of evidence is documented, and the connection between the evidence and the conclusion is transparent and reproducible. This is a higher standard than most pharma organisations currently apply to their decision-making infrastructure, and it is the standard that regulators, HTA bodies, and payers are increasingly requiring pharma to meet. At Pienomial, we built KnolAI to make the evidence based decision pharma standard operationally achievable for every team in the organisation, not just the ones with the resources to commission manual evidence reviews. [9]
1. What an Evidence-Based Decision Actually Requires
The term evidence-based originates in clinical medicine, where it refers to the integration of the best available research evidence with clinical expertise and patient values. In the pharmaceutical organisational context, an evidence based decision pharma has three specific requirements that most strategic decision processes do not fully satisfy. [1]
Requirement 1, Every claim has a source: Every factual assertion that influences the decision, whether a clinical efficacy figure, a competitive landscape claim, an HTA precedent, or a payer coverage characterisation, must be traceable to a specific primary source. A claim that is remembered from a prior analysis, summarised from a secondary report, or generated by a language model without source attribution does not satisfy this requirement. The source is the foundation of the claim's credibility, and without it, the claim is an assumption.
Requirement 2, The source is verified: A cited source must actually exist, contain the claimed information, and be the appropriate level of primary evidence for the decision context. This requirement is now more pressing than ever: as the fabricated citation crisis in scientific literature demonstrates, cited sources increasingly contain references that do not exist or do not support the cited claim. Verification means independently confirming that the source exists and says what the citation claims. [2]
Requirement 3, The decision process is transparent and reproducible: Another analyst, reviewer, or decision-maker must be able to follow the chain from evidence to conclusion and reach the same understanding of what the evidence supports. This requires that the decision process itself, not just the outcome, is documented in a form that can be reviewed and challenged. [7] Decisions made from institutional memory, hallway conversations, or AI-generated summaries without source attribution are not reproducible. They are judgment calls presented as evidence-based decisions.
2. The Cost of Decisions That Are Not Evidence-Based
The pharmaceutical industry is making decisions of extraordinary consequence and cost from evidence bases that frequently do not satisfy the three requirements above. The consequences of this gap are measurable. [4]
At the portfolio investment level, companies are approaching $300 billion in annual revenue at risk from patent expirations between 2025 and 2030. The portfolio decisions that will determine which companies navigate this successfully require precise evidence about pipeline productivity, competitive landscape, and market access probability. A portfolio investment decision made from an inaccurate competitive intelligence claim, an outdated HTA precedent, or an unsourced clinical comparison does not just waste analysis time. It allocates hundreds of millions of development dollars to the wrong asset.
At the clinical protocol level, a Phase III protocol designed without accurate evidence of the HTA body's comparator requirements, endpoint patient-relevance standards, or subgroup analysis expectations may produce trial data that is technically excellent and commercially insufficient. The cost of this evidence gap is not the cost of the wrong analysis. It is the cost of the trial: hundreds of millions of dollars and years of development time that produced an evidence package that cannot achieve the commercial outcome it was designed to support. [5]
At the AI deployment level, 44% of organisations have reported experiencing negative consequences from generative AI use, with average financial losses of $4.4 million per incident. The vast majority of these consequences occurred because AI outputs were used to inform decisions without verifying the evidence those outputs were based on. When the AI output contains a hallucinated clinical citation or a misattributed efficacy figure, and that output is used to inform a submission, a negotiation, or a portfolio decision, the financial consequence is the cost of acting on a claim that was presented as evidence but was not.
3. Where Evidence-Based Decisions Break Down in Pharma Practice
Five specific failure patterns account for most evidence-based decision failures in pharma organisations. Each reflects a gap between the aspiration of evidence-based decision-making and the operational infrastructure available to achieve it.
Failure 1, Claim sourcing from secondary reports: A HEOR analyst builds an evidence summary by reading a competitor's HTA submission overview published by a consulting firm. The claims in the overview are referenced to the original dossier, which the analyst does not have access to. The decision is based on a secondary source interpretation of a primary source document. If the consulting firm's characterisation was incorrect or incomplete, the decision is based on a wrong premise with no mechanism to detect it.
Failure 2, Institutional memory as evidence: A market access team meets to discuss the payer evidence requirements for a submission. The head of market access recalls from a prior project that NICE accepts PFS as a surrogate endpoint in this indication based on a prior appraisal. The recollection is not verified against the actual appraisal document. NICE's methodology guidance has since been updated. The strategy is built around an incorrect premise that no one checked.
Failure 3, AI-generated summaries used as primary evidence: A competitive intelligence team uses a general-purpose AI tool to generate a landscape summary. The AI output contains clinical trial results with specific efficacy figures for competitor products. The figures sound precise and are formatted like citations. Two of the figures are from the wrong trials. The CI brief goes to leadership and influences a portfolio prioritisation decision.[7]
Failure 4, Outdated evidence presented as current: A regulatory strategy team uses an evidence summary from a prior submission to inform the evidence architecture for a new product in the same indication. The prior summary was accurate at the time it was prepared. The competitive landscape has since changed: two new approvals have shifted the standard of care and the HTA comparator definition. The regulatory strategy is built around a landscape that no longer exists.
Failure 5, Cross-functional evidence inconsistency: The HEOR team's evidence summary for a NICE submission uses one set of efficacy figures. The CI team's competitive landscape briefing uses different figures for the same products because they sourced from different databases. The inconsistency is not identified before the submission, and a NICE reviewer spots it.
4. The Evidence Chain: What Sourced Claims Look Like in Practice
An evidence based decision pharma requires that every claim influencing the decision has what we call a complete evidence chain: a traceable path from the claim in the decision document to the specific primary source that supports it.
In practice, a complete evidence chain for a competitive intelligence claim looks like this: the claim states that nivolumab achieved a median OS of 12.2 months in second-line NSCLC in the CheckMate 017 trial. The evidence chain links: the claim, to the specific data extraction from the CheckMate 017 publication, to the specific table in that publication, to the PubMed-indexed journal article with confirmed author list, DOI, and publication date, to the knowledge graph triple that was validated during ingestion and flagged as the primary source for this specific efficacy result.
An HTA precedent claim with a complete evidence chain states that NICE accepted PFS as a primary endpoint in second-line NSCLC for nivolumab in TA403, based on an established correlation with OS documented in the assessment documentation. The evidence chain links: the claim, to the specific NICE assessment document TA403, to the specific section of that document that documents the endpoint acceptability judgement, to the confirmed NICE website source with the official assessment publication date.
This is not a theoretical standard. It is what KnolAI delivers by default for every claim it generates, because the architecture stores facts as sourced entity-relationship triples and constrains output generation to those retrieved triples. Every KnolAI output is an evidence chain by construction.
5. Evidence-Based Decisions in Portfolio Strategy
Portfolio strategy is one of the highest-stakes decision contexts in pharma, and one of the most evidence-dependent. Companies generating 70% or more of revenues from their top two therapeutic areas have delivered 65% higher total shareholder returns over the past decade compared to those with broader, less focused portfolios. The decision of which therapeutic areas to concentrate in requires accurate evidence about competitive positioning, regulatory precedent, HTA probability, and market access landscape across multiple indications simultaneously.
Portfolio decisions made from incomplete or inaccurate evidence carry the full commercial cost of the assets they misdirect. A go decision on an asset whose clinical differentiation claim was based on a misattributed competitive efficacy figure commits development resources that cannot be redirected once Phase III is underway. A no decision on an asset whose HTA probability was underestimated because the HTA precedent analysis was based on outdated comparator data foregoes commercial opportunity that cannot be recovered.
KnolAI combines AI expert intelligence with multi-domain research capabilities, covering clinical, regulatory, HTA, competitive, and commercial intelligence from a single unified knowledge layer. This provides the evidence foundation that portfolio strategy decisions require: current, sourced, verified intelligence across all five decision-relevant domains, generated simultaneously in a single structured output with claim-level attribution for every element.
6. Evidence-Based Decisions in Clinical Protocol Design
The protocol design decision is the most consequential evidence-based decision in drug development, because it determines the quality of the evidence that all subsequent decisions will be based on. A Phase III trial designed with accurate evidence of HTA body comparator requirements, endpoint patient-relevance standards, and subgroup pre-specification expectations produces trial data that satisfies payer evidence requirements at submission. A trial designed without this evidence produces data that may be scientifically excellent and commercially insufficient.
Evidence-based protocol design requires intelligence from multiple domains: the clinical evidence precedent for endpoint acceptance in the indication, the regulatory precedent for label language linked to specific endpoint evidence, the HTA precedent for the comparator definitions that NICE, G-BA, and HAS have applied to analogous products, the competitive landscape of endpoint strategies that competitors have used and the HTA outcomes those choices produced, and the patient-relevance judgements that JCA assessors have made for surrogate endpoints in the indication.
KnolAI retrieves this multi-domain protocol design intelligence from the Knolens knowledge layer in a single structured output. The protocol design team receives a sourced intelligence brief covering all five domains, with every claim traceable to its primary source, before the statistical analysis plan is finalised. The evidence foundation for the protocol design decision is complete, current, and verified.
7. Evidence-Based Decisions in HTA Submissions
HTA submissions are formal evidence-based decisions presented to external reviewers with the authority to determine reimbursement access. Every claim in a NICE submission, a G-BA dossier, or an ICER evidence report must satisfy the sourcing, verification, and transparency requirements of an evidence-based decision to survive the review process. HTA reviewers are specifically trained to identify unsourced claims, verify cited sources against primary documents, and challenge evidence interpretations that are not supported by the evidence presented.
The evidence-based decision standard for HTA submissions is operationally demanding. Every efficacy figure must link to a specific trial result in a specific publication. Every comparator characterisation must link to the HTA body's own prior assessment documents. Every ITC methodology claim must link to the specific methodology guidance document it claims to follow. Every RWE conclusion must link to a specific data source, analysis methodology, and result.
KnolAI and KnolComposer, working together within the Knolens platform, generate HTA submission content with this claim-level evidence chain built in by architecture. The HEOR team reviewing the KnolComposer output is reviewing sourced claims, not checking unsourced AI-generated prose against primary sources. The verification step is replaced by a structured review of evidence that is already traceable.
8. AI's Role in Evidence-Based Decisions: Intelligence Engine, Not Oracle
The arrival of generative AI in pharma has created a specific and dangerous confusion: the confusion between AI that generates plausible-sounding intelligence and AI that retrieves verified evidence. The first category produces outputs that satisfy the surface form of an evidence-based claim, a confident assertion with specific numbers, without satisfying the substance: a verified primary source that supports those numbers.
The correct role of AI in evidence-based decision-making is as an intelligence engine that retrieves verified evidence from a governed knowledge base, synthesises it across domains, and presents it with the source attribution and audit documentation that makes the evidence chain transparent. This is how KnolAI is designed to function: not as an oracle that generates answers, but as an intelligence engine that retrieves facts from verified sources and presents them in the format that decision-makers need.
The distinction is architectural. An AI system that generates answers is structurally incapable of providing the evidence chain that evidence-based decisions require, regardless of how accurate its outputs are on average. An AI system that retrieves from a governed, sourced knowledge base is structurally incapable of generating an unsourced claim, because the generation is constrained to retrieved triples that each carry their source. This is why we built KnolAI on a knowledge graph architecture rather than a language model architecture for the research intelligence function: the architecture determines whether evidence-based decision-making is achievable, not the quality of the language model.
9. How Fast Can Your Team Build Evidence-Based Decision Infrastructure with KnolAI?
Building the evidence infrastructure that genuine evidence based decision pharma requires does not mean rebuilding your organisation's research workflows from scratch. KnolAI integrates into existing HEOR, CI, regulatory, and market access workflows as the governed evidence layer that makes every claim traceable and every decision auditable. Most pharma teams are producing their first fully sourced, evidence-chained intelligence outputs within two weeks of onboarding.
Sprint 1, Weeks 1 to 2, Evidence-chained outputs from day one: KnolAI is connected to your indication scope and primary decision use cases. The Knolens knowledge layer is populated with validated clinical, regulatory, HTA, and competitive intelligence for your therapeutic area. Every KnolAI output from the first session contains claim-level source links. The evidence chain is structural, not a post-generation annotation exercise.
Sprint 2, Weeks 3 to 4, Decision-specific evidence architectures configured: Evidence architecture briefs for your primary decision contexts, whether protocol design, HTA submission, payer negotiation, or portfolio prioritisation, are configured for your indication and target bodies. KnolPersona assessor challenge simulation is activated to stress-test the evidence supporting each decision before it is acted on.
Sprint 3, Weeks 5 to 6, Cross-functional evidence consistency live: The same Knolens knowledge layer serves HEOR, CI, regulatory, and market access teams simultaneously. Evidence consistency across functions is structural: every function draws from the same sourced knowledge base, so the clinical efficacy figures in the NICE submission are identical to those in the CI brief and the board portfolio presentation, attributed to the same sources. Cross-functional evidence inconsistency is eliminated by architecture.
Conclusion
An evidence based decision pharma is not a decision informed by data. It is a decision in which every claim influencing the decision is traceable to a verified primary source, the evidence chain is documented, and the decision process is transparent and reproducible. This standard is what regulators require, what HTA bodies enforce, and what the most commercially successful pharma portfolio decisions are built on.
At Pienomial, we built KnolAI as the evidence intelligence engine that makes this standard achievable at scale: not by asking teams to do more manual verification, but by making source verification structural, claim-level attribution automatic, and cross-functional evidence consistency a property of the knowledge architecture rather than a coordination challenge. Every KnolAI output is an evidence chain. Every decision made from KnolAI outputs is, by construction, an evidence-based decision.
CTA: See how KnolAI makes every pharma decision evidence-based by architecture. Book a demo with the Pienomial team.
