A Launch Does Not Fail in the Forecast. It Fails in the Exits.

Half of a launch class misses its first-year forecast. The post-mortems almost always blame the forecast. In my experience the forecast was close to right about the patients. It was wrong about how many of them would make it all the way to a paid dispense.

New product commercialization fails in a specific, repeatable way. A forecast quietly assumes commercial infrastructure that a newly approved product does not have yet: a formulary position, a working authorization pathway, dispensers who know the product, an enrolled support base, and claims history to reconcile against. For the first two or three quarters the product is sold into a market that has none of that, while bridge supply masks the shortfall. By the time the first clean Gross-to-Net read arrives, the decisions that shaped it are already locked.

Half the class misses, and the diagnosis is usually wrong

Trinity Life Sciences found that half of the 2023 United States launch class underperformed its pre-launch first-year forecast, with 39 percent above it. That split has been roughly stable for years, which tells you something. If the problem were forecasting skill, a decade of better analytics would have moved it. It has not moved much, because the miss is not mostly an estimating error. It is an execution gap that shows up in the same line of the P&L as an estimating error.

A forecast is a statement about patients: how many exist, how many get diagnosed, how many get treated, what share this product wins, what it nets per unit. Every one of those is defensible when the model is built. What the model does not carry is the operational state of the market on the day the product actually arrives.

Read that right column again. Four of those five gaps close over the first year with ordinary work. The fifth does not close at all. There is no claims history for a product that has never been dispensed, so every early accrual is an assumption, and every variance explanation is a guess dressed as an analysis.

The exits are widest exactly when the product is newest

The 10 Exits of Therapy™ describe the ten places a patient can leave between diagnosis and a first dose. At launch, four of them are wider than they will ever be again. Coverage has not been established. Authorization criteria are new or absent, so denials are the default rather than the exception. Affordability programs are still enrolling. Fulfillment runs through dispensers who have never handled the product.

The scale of the authorization problem is now documented. A Johns Hopkins analysis published in JAMA in July 2026, covering more than two million fill attempts for single-source brand drugs, found initial rejections rose from 24.3 percent in 2018 to 40.7 percent in 2024. Of those rejected attempts, 48.4 percent were never followed within 90 days by a fill of the prescribed drug or any drug in the same class. The patients who did get treated waited an average of 12 days.

Those are market-wide figures across established products. A product in its first months, with no published criteria and no appeal history, does not do better than the average. It does worse, and it does worse at the exact moment when every early data point is being used to calibrate the rest of the launch.

Bridge supply hides the leak for two quarters

Here is the part that makes launch different from everything that follows. Most launches put free product in front of the coverage gap on purpose. Bridge and quick-start programs exist so a patient is not left waiting while a payer decides, and that is the right thing to do for the patient. It also means that for the first two quarters, demand looks real and almost none of it is paid.

So the early signal is clean in the wrong direction. Units are moving. Prescribers are writing. Nothing in the demand data says that authorization is failing, because the patient got the drug anyway. The failure only becomes visible when the free supply tapers and paid claims have to carry the volume, and that is precisely when the first real rebate, chargeback and government pricing actuals start landing.

By month twelve the contracting terms are signed, the channel is built, the copay program rules are set, the hub vendor is selected and the pricing corridor is established. Those are Phase 1 decisions and they are expensive to reverse. The data that would have informed them shows up in Phase 2.

What Phase 1 actually decides

Phase 1 Pre-Launch and Launch Readiness is usually described as preparation. It is better understood as commitment. Almost every structural driver of Gross-to-Net for the life of the product gets fixed in that phase, before a single unit is dispensed.

1. The pricing corridor. List price sets the ceiling on every statutory discount that follows, and it is the one number that cannot be quietly revised later.
2. Contract architecture. Which payers, which tiers, what rebate structure and what volume commitments. These terms outlive the people who negotiated them.
3. Channel design. How many distributors, which specialty pharmacies, what fee structure. This determines where visibility ends and where it survives.
4. Affordability program rules. Eligibility, caps and duration, which decide whether the program supports patients or subsidizes benefit design.
5. What gets measured. The least reversible decision of all, because a signal nobody instrumented at launch cannot be recovered retroactively.

What to instrument before approval day

If the sequencing problem is real, the fix is not a better forecast. It is moving the evidence earlier. Four things are worth building before the product ships, because none of them can be reconstructed afterward.

1. A first-fill funnel, not a shipment report. Attempts, rejections, reasons, appeals, reversals and time to fill. If the only launch metric is units shipped, the leak is invisible by construction.
2. A paid versus supported split from day one. Report bridge and quick-start volume separately from paid claims in every launch review. Blending them produces a number that is comforting and wrong.
3. Time to therapy as a tracked measure. Twelve days is a national average. Your own number is knowable and almost nobody measures it during a launch.
4. A Gross-to-Net accrual built to be revised. Assume the first four quarters are estimates and design the true-up governance before you need it, rather than defending a number that was never more than a placeholder.

The part that carries over

The enterprise’s role here is narrow and worth stating precisely. It develops, manufactures, prices, contracts, distributes and supports. It does not diagnose, prescribe, dispense or treat. Every lever above sits on the supply side of a decision a clinician and a patient make together, which is exactly why removing friction is the only honest way to influence the outcome.

At launch, that sentence is doing more work than anywhere else in the lifecycle, because so little has survived long enough to be measured yet. A launch does not fail in the forecast. It fails in the ten places a patient can leave, during the one period when nobody can see them leaving.

Figures cited are the published findings of the named sources. The operating patterns described reflect my own experience across launches and are offered as commentary, not as legal, regulatory, clinical or pricing advice. The views expressed are my own and are not made on behalf of any current or former employer. All trademarks belong to their respective owners.