There is a specific question that almost every biotech commercial model answers incorrectly — not because the team is careless, but because the data that would answer it correctly does not exist in any source that an AI tool, a published guideline, or a market research report can access. The question is: what will the oncologists who matter most in your target market actually prescribe, given how they are practising medicine right now?
Not what the guidelines say they should prescribe. Not what the clinical trial data suggests the optimal treatment is. Not what the key opinion leaders said at the last ASCO presentation. What they will actually do, in their actual practice, with their actual patient population, under their actual prior authorisation and formulary constraints, when your therapy is available and competing against everything else on their desk.
This is the prescriber intent gap. And it is the most expensive assumption in biotech commercial planning.
What the Most Recent Prescriber Data Actually Shows
The 2026 Oncology Drug Report — the most comprehensive annual survey of prescribing behaviour among US oncologists — contains a finding that every commercial team building a launch model should have read but most have not internalised. In 2025, only 35% of oncologists surveyed reported frequently prescribing a newly approved drug, down from 44% in 2024. This is not a marginal shift. It is an 11-point decline in aggressive adoption in a single year.
The same survey found that oncologists are becoming more selective — not less engaged with innovation, but more deliberate about it. Nearly 50% are now prescribing just one to three new drugs annually. The era of reflexive early adoption, where approval equals uptake, is ending. Oncologists are now discriminating between therapies based on a set of practical factors that never appear in a clinical trial report and are never captured by AI tools reading published literature.
The IQVIA data makes the problem concrete in a way that guidelines and trial data cannot. When a physician writes a prescription, the commercial model counts it. But 27% of those prescriptions — driven overwhelmingly by payer rejections, prior authorisation failures, and cost-related abandonment — never result in a patient receiving the therapy. Every commercial model built on prescription intent without accounting for fill rate is structurally overestimating the accessible market from day one of launch.
And this is before considering that prior authorisation is now reshaping prescribing intent before the prescription is even written. The same oncology drug report found that physicians are increasingly engaging in anticipatory avoidance: proactively not prescribing certain therapies because of anticipated administrative burden — before they even encounter a denial. This behaviour does not appear anywhere in published clinical data. It only surfaces when you ask the prescribers directly.
The Four Forces That AI Cannot Read — But Prescribers Can Tell You
AI tools trained on published medical literature, clinical trial data, and treatment guidelines are reading a description of ideal prescribing behaviour in controlled settings. The four forces that actually determine commercial prescribing happen after the guidelines are written — in the operational reality of clinical practice — and none of them are captured in any published source.
Which specific insurers, PBMs, and integrated delivery networks cover your therapy at which tier, with what prior authorisation criteria, and with what administrative burden attached. Three in four oncologists say cost and reimbursement issues affect their prescribing. Not in theory — in practice, for specific therapies, specific patient populations, and specific coverage scenarios. This is not in any guideline. It is in the prescribers' daily experience.
The actual sequence in which physicians use therapies in practice differs systematically from what guidelines recommend. Tumour board consensus, local institutional norms, peer influence patterns, and the specific clinical experience of the practice shape sequencing decisions that guidelines describe in the abstract. These informal hierarchies are invisible to any tool that reads published data — and they determine exactly where in the treatment sequence your therapy will sit, which directly determines your addressable market.
Not every practice that is clinically qualified to prescribe a therapy has the infrastructure to administer it. The 2026 oncology drug report found that nearly one in four oncologists reported being slightly or not at all prepared for subcutaneous drug administration — with rural practices showing readiness at half the rate of urban ones. A launch model that projects uniform uptake across practice types is wrong before it leaves the spreadsheet.
How prescribers actually perceive your therapy relative to alternatives — based on their own clinical experience with their own patient populations, not on published trial comparisons — determines where it fits in their mental hierarchy. A therapy that outperforms on paper in a controlled trial population may underperform in the specific patient mix of a community oncology practice. Prescribers know this. The clinical literature does not capture it. Only a primary survey does.
The cumulative effect of these four forces is a systematic divergence between what clinical data and AI-synthesised guidelines predict about prescribing behaviour and what actually happens in practice. That divergence is not random noise. It is structured and predictable — and it can be measured before launch, in a 72-hour primary survey of 500 oncologists who are actively practising in the indications most relevant to your therapy.
What Happens When the Commercial Model Gets This Wrong
The consequences of building a commercial model on guideline-adherence assumptions — rather than validated prescriber intent data — are concrete, predictable, and expensive. They follow a consistent pattern across launch types.
The pattern is consistent: a commercial model anchored to guideline-eligible patient population produces a TAM estimate. That estimate is then systematically eroded at each step of the actual prescribing pathway by forces that the guidelines do not describe and that AI tools cannot access. By the time a therapy reaches actual sustained use in a real patient, the accessible market is often a fraction of the guideline-eligible population.
The teams that discover this after launch spend the subsequent 12 months explaining variance to investors and revising commercial plans. The teams that discover this before launch build the commercial model around what will actually happen — and allocate sales force, market access, and patient support resources accordingly.
Why AI Cannot Close This Gap — and What It Can Do
The failure mode is not that AI tools are bad at their job. They are excellent at synthesising published data at speed. The failure mode is a category error: teams are using AI synthesis as a substitute for primary prescriber intelligence, because AI synthesis is fast and cheap and available now.
AI tools trained on published literature will tell you what the evidence base says about your therapy. They will summarise the NCCN guidelines accurately. They will identify relevant KOLs by publication history. They will map the competitive landscape as it was described in the most recent analyst reports. They will do all of this in minutes, at essentially no cost.
None of that is the prescriber intent data you need to build a credible commercial model.
The distinction matters because the commercial model built on AI synthesis and the commercial model built on primary prescriber data produce materially different revenue projections — and one is considerably more likely to reflect commercial reality. The history of biotech commercial launches contains consistent examples of teams who discovered this gap after launch rather than before it.
The Three Prescriber Intelligence Questions That Change Every Commercial Model
A primary prescriber survey designed around your specific therapy and indication produces validated answers to three questions that no other data source can resolve:
1. What proportion of the oncologists who matter most to your launch will actually prescribe your therapy — segmented by practice type, payer environment, and patient population mix? Not the guideline-eligible universe. The prescribers who will actually write the prescriptions that drive year-one revenue. The survey identifies them, segments them, and tells you exactly what their intent is and what conditions must be met for that intent to become a prescription.
2. Where in the treatment sequence will your therapy sit — and what factors determine whether it is used first-line, second-line, or reserved for specific patient subgroups? Sequencing drives market size far more than approval language does. A therapy approved for first-line use that is functionally prescribed second-line at 60% of practices has a structurally different commercial model than the approval label implies. This gap is invisible in published data and measurable in a prescriber survey.
3. What are the specific access barriers — prior authorisation criteria, formulary tier, patient cost-sharing — that will most significantly reduce prescription-to-fill conversion in your target markets? With 27% of written prescriptions not filled across all therapies, and oncology access barriers structurally higher than average, the fill rate assumption in a commercial model is the number most likely to be wrong. Primary surveys of prescribers and pharmacy benefit managers, designed around your specific therapy and payer mix, produce the fill rate estimates that change the commercial projection.
Tell us the prescriber intent question your launch model currently cannot answer. We design the study that resolves it.
By submitting, you agree to our Privacy Policy. We process data in accordance with applicable data protection laws.