What pharmaceutical forecasting is, and why most forecasts miss
Pharmaceutical forecasting is the process of estimating how many patients will use a drug, for how long and at what net price, so that a company can plan supply, budgets, field force size and investor guidance. There are three main methods: trend-based models built on past sales, epidemiology-based models that start from disease prevalence, and patient-flow models that track patients through lines of therapy. The right one depends on how much sales history you have and how complex the treatment pathway is.
The track record is humbling. A McKinsey team published a review in Nature Reviews Drug Discovery in 2013 (Cha, Rifai and Sarraf) that looked at roughly 1,700 analyst forecasts for 260 drugs launched between 2002 and 2011. As Fierce Pharma reported, more than 60 percent of those forecasts were wrong by more than 40 percent of actual peak revenue, in one direction or the other, and accuracy was still poor years after launch.
Those were sell-side analysts with every incentive to be right. Internal brand forecasts face the same problems plus a few more: assumptions owned by nobody, data that disagrees between teams, and a spreadsheet only one person can update. This guide covers how to pick the right method, what a sound model contains, a worked example with real arithmetic, the inputs that changed in 2025 and 2026, and how to decide whether to build the forecast in house or bring in help.
Pharma forecasting methods compared: which model fits your brand
Trend-based
Projects future volume from historical prescriptions, shipments or dispenses, adjusted for known events such as a competitor launch or a price change. Fast to build and easy to explain. It works for established in-market brands with at least two years of stable data, and fails for launches, label expansions or any market where the treatment paradigm is shifting.
Epidemiology
Starts from the number of people with the disease, then narrows the funnel: diagnosed, treated, eligible for your line of therapy, and finally your share. This is the standard method before launch and for rare diseases, where there is no sales history to trend. Its weakness is that every funnel step is an assumption, and small errors multiply.
Patient flow
Models patients as they start, switch, discontinue and return across lines of therapy over time. It captures what the other two miss: how long patients actually stay on drug, and where they leak to competitors. It is the best fit for chronic and specialty therapies, oncology and crowded categories, but it needs real-world data on persistence and switching to calibrate.
Analog
Borrows the uptake curve of comparable past launches, often used to shape the first two or three years of an epidemiology model. Useful as a sense check, risky as the only method, because no two launches share the same access, competition and price.
How to choose a pharma forecasting model
Pick the method from the decision the forecast supports, not from the software you happen to own. A supply planner needs a monthly unit forecast for the next 18 months. A business development team valuing an asset needs a ten-year revenue range with clear scenarios. A launch team needs to know how many patients to expect in each quarter and where they will come from. Those are different models.
| Your situation | Recommended approach | Minimum data you need |
|---|---|---|
| Pre-launch, new mechanism | Epidemiology funnel with analog uptake curves | Published prevalence, diagnosis and treatment rates, 2 to 4 analog launches |
| Rare disease | Epidemiology plus patient finding, reviewed quarterly against identified patients | Prevalence estimates, claims-based patient counts, hub or specialty pharmacy enrollments |
| Recently launched specialty brand | Patient flow, calibrated monthly to actual starts and persistence | Specialty pharmacy dispense and status data, hub data, claims |
| Mature brand, stable market | Trend-based with event overlays | 2+ years of prescription or shipment data, known event calendar |
| Brand facing loss of exclusivity or a negotiated Medicare price | Trend or patient flow plus explicit price and erosion scenarios | Payer mix, gross-to-net history, analog erosion curves |
Two rules hold across all of them. First, keep volume and price as separate modules, because they move for different reasons and are owned by different teams. Second, build a version you can explain in one page. Evaluate, a forecasting software vendor, makes the same point in its guidance on good and bad models: an overly complex model becomes a barrier to decisions rather than an aid to them.
How to build a pharmaceutical forecasting model, step by step
Write down the decision and the horizon
Name who will use the forecast, for what decision, at what granularity (month, quarter, year) and over how long. This sets the structure of everything that follows.
Define the market
Agree the indication, line of therapy, geography and patient population in writing. Most forecast arguments are really disagreements about the market definition.
Size the patient funnel
Prevalence or incidence, diagnosis rate, treatment rate and eligibility for your product. Cite a source for every number and record its date.
Model share and uptake
Set peak share and the time to reach it, using analogs and primary research. Add the events that move share: competitor launches, label changes, guideline updates.
Convert patients to volume
Apply dosing, compliance and persistence to turn patients on therapy into units. Persistence is where patient-based models earn their keep.
Convert volume to net revenue
Apply list price, then gross-to-net deductions by channel and payer: rebates, chargebacks, 340B, copay support and, from 2026 onward for selected drugs, Medicare negotiated prices.
Pressure-test and run scenarios
Find the three or four assumptions that move the answer most, and show low, base and high cases for them instead of a single number.
Track actuals against the forecast
Compare every month, explain variances by assumption, and re-base on a set schedule. A forecast nobody reconciles to actuals stops being trusted within a quarter.
A worked example: a patient-based forecast for a specialty brand
Here is a simple patient-based forecast for a hypothetical second-line specialty therapy in the US. Every input is illustrative and chosen for round arithmetic, not taken from a real product. The point is to show how the funnel works and where the sensitivity sits.
| Step | Assumption | Patients |
|---|---|---|
| Prevalent patients | Starting population | 200,000 |
| Diagnosed | 70% | 140,000 |
| Drug-treated | 60% of diagnosed | 84,000 |
| Eligible for second line | 35% of treated | 29,400 |
| Peak share of second line | 15% | 4,410 patients started |
| Average time on therapy | 70% of the year (8.4 months) | 3,087 patient-years |
| Net price per patient-year | $45,000 after gross-to-net | $138.9 million net revenue |
Now change one input at a time:
- Persistence falls from 70% to 55% of the year. Patient-years drop to about 2,426 and net revenue to about $109.1 million, a fall of 21 percent. Nothing about the market or the share changed.
- Net price falls 20 percent to $36,000 because of deeper rebates. Net revenue drops to about $111.1 million, a fall of 20 percent.
- Both happen together. Net revenue lands near $87.3 million, 37 percent below the base case.
That last case is close to the 40 percent error band in the McKinsey study, and it comes from two assumptions that brand teams often treat as fixed. This is why we push clients to calibrate persistence from real dispense data, using a defined measure such as those compared in our guide to PDC vs MPR adherence, and to model net price explicitly, as covered in our gross-to-net guide for specialty pharma.
What changed for pharma forecasting in 2025 and 2026
Three shifts mean that forecasts built two or three years ago need their price and volume logic rebuilt, not just rolled forward.
Medicare negotiated prices. On 25 November 2025, CMS announced negotiated prices for 15 more Part D drugs, taking effect on 1 January 2027 and bringing the total to 25 when added to the first 10, whose prices took effect on 1 January 2026. CMS put the net savings at 44 percent, or about $12 billion, against the previous year's spending on those 15 drugs. If your brand, a competitor or a therapeutic alternative is on the list, both your net price and the competitive dynamics in Medicare change, and the forecast needs a separate Medicare module.
The Part D benefit redesign. The redesign that began in 2025 changed how much of the cost sits with patients, plans and manufacturers. That affects out-of-pocket cost, which in turn affects abandonment and persistence, two inputs that patient-flow models depend on.
Better patient-level data. Tokenized claims, specialty pharmacy status feeds and hub data now let mid-sized brands measure starts, time to therapy and discontinuation directly instead of guessing. Our pieces on NBRx vs NRx for new patient starts and therapy discontinuation reasons cover how to read those signals. The forecast is only as good as the data used to calibrate it, which is why the data layer is usually the first thing we fix.
Common pharma forecasting mistakes
Treating persistence as a constant
One persistence number for every cohort hides the biggest driver of revenue in specialty brands. Measure it by cohort and line of therapy from dispense data.
- Calibrate from real data
- Revisit quarterly
Forecasting gross, reporting net
A model that stops at gross sales cannot explain why finance's number is lower. Build gross-to-net in, by channel and payer.
One number, no range
A single point estimate invites false confidence. Show low, base and high cases tied to named assumptions.
- Scenario per key driver
Optimistic launch curves
Analog curves picked from blockbusters inflate early years. Choose analogs on access, competition and route of administration, not just disease.
A model only one person can run
If the forecaster leaves, the logic leaves with them. Document assumptions, sources and version history inside the model.
No variance tracking
Without a monthly comparison of actuals to forecast by assumption, the model never improves and stakeholders stop believing it.
Build in house, buy software or hire pharma forecasting consulting?
There are three ways to get a forecast, and most mid-sized companies end up with a mix.
| Option | Works well when | Watch out for |
|---|---|---|
| In-house forecaster with Excel | One or two brands, an experienced forecaster, stable markets | Key-person risk, manual data refresh, no audit trail of assumption changes |
| Forecasting software platform | Several brands, a central forecasting team, need for standard templates | Licence cost, and the software does not supply the data or the judgment |
| Specialist consultancy | Pre-launch models, asset valuation, a one-off rebuild, or no forecaster on staff | Models handed over that nobody internal can run or update |
| Analytics partner for the data layer | The method is sound but the inputs are late, manual or disputed | Make sure the partner understands pharma data, not just pipelines |
When not to hire anyone. If you have a single mature brand, two years of clean data and a forecaster who reconciles to actuals every month, keep doing what works. Paying a consultancy to rebuild a model that is already within a few percent of actuals is money better spent elsewhere.
When outside help pays off. Before a launch, when your forecast has missed by more than your tolerance two quarters running, when net price assumptions are changing because of Medicare negotiation, or when the monthly refresh takes your team a week of manual work. In that last case the problem is usually data, not modelling, and fixing it is part of our pharma commercial analytics consulting work.
Where Greenwolf fits
Related reading
For the metrics that sit next to a forecast during launch, read our guide to pharma launch analytics. If you are choosing a patient-level data source to calibrate the model, see Komodo Health vs IQVIA. For non-pharma businesses, our sales forecasting service applies the same discipline of separate volume and price drivers, scenarios and monthly variance tracking.
Questions about pharmaceutical forecasting
What are the main methods of pharmaceutical forecasting?
What is patient-based forecasting in pharma?
How accurate are pharmaceutical sales forecasts?
What data do you need for a pharma forecasting model?
How does Medicare drug price negotiation affect forecasts?
Should we use Excel or forecasting software?
How often should a pharma forecast be updated?
Is your forecast running on late or disputed data?
Tell us which brand, which data sources feed the model today and how long the monthly refresh takes. We will map it in a short digital audit and show what an automated, reconciled forecast input would look like.
Part of our Pharma Commercial and Patient Analytics hub. Start with our specialty pharmacy analytics.