Biosimilar Forecasting in the UK: A Method That Survives Contact With Actuals

Biosimilar forecasting has a credibility problem. Ask three people in the same company for next year's number on a molecule facing loss of exclusivity and you will usually get three answers — one from finance's spreadsheet, one from the brand team's optimism, and one from whoever last spoke to a procurement pharmacist. All three will be wrong, and nobody will find out how wrong until it is too late to matter.

That isn't because biosimilars are unforecastable. It's because most forecasts are built as a revenue line with assumptions buried inside it. The UK market punishes that approach faster than almost any other, because the NHS is an unusually decisive buyer: national commissioning guidance has for years pushed rapid adoption of best-value biological medicines, and regional tenders can move the majority of a molecule's hospital volume in a single decision.^[1]

Here is the method we build into pharmaceutical forecasting platforms, and why each step exists.

1. Forecast the molecule market first, your share second

The single most common structural mistake is forecasting your product's sales directly. Volume moves for two different reasons — the whole molecule market growing or shrinking, and share moving between originator and biosimilar competitors — and if your model blends them, you can never diagnose a miss.

Model the total molecule market in therapy units (days of therapy, standard units, or patients — anything but packs and pounds). Mature biologic markets in the UK typically grow modestly and predictably; the violent movement is almost always in share. Splitting the two means that when actuals land, you can see *which* assumption was wrong instead of arguing about the total.

2. Anchor uptake curves to comparable UK launches, not global averages

Biosimilar uptake follows an S-curve, but the steepness is a local property. The UK's historical pattern is distinctive: for molecules where the NHS ran coordinated switching programmes, biosimilar share has reached high double digits within a year — far faster than most European averages suggest.^[1] A curve borrowed from a global forecast pack will be too slow.

The practical approach: build a small library of UK reference launches per setting (hospital-tendered biologics behave differently from primary-care molecules), and pick the anchor by mechanism — is volume moved by a tender award, by a switching programme, or by organic clinician choice? A tender-driven molecule doesn't have a smooth curve at all; it has steps, timed to contract cycles.

3. Make price erosion a separate, explicit assumption

Volume share and price tell different stories. UK biosimilar list prices erode steeply after loss of exclusivity, and net prices (after tender discounts) erode further still. If value is forecast directly, price erosion silently eats your volume story. Keep them apart: units × share × net price, each visible, each challengeable in its own right.

4. Wire in actuals and score the forecast monthly

This is the step that separates a forecast from a wish. Every month, load the latest market data — IQVIA sell-out where you have it, supplemented by public NHS dispensing data such as the hospital medicines dataset and primary-care prescribing published through platforms like OpenPrescribing^[2] — and compare it against what the forecast said.

Two numbers matter: variance (how far off was this month?) and bias (are we *systematically* high or low?). Variance is noise until it is a trend; bias is a broken assumption wearing a disguise. A forecast that is 4% high every month for five months doesn't need a better month — it needs its uptake curve corrected.

5. Automate the run, govern the assumptions

None of the above survives if the forecast lives in one analyst's workbook. The refresh has to be automated — new actuals in, model rerun, variance reported, without a fortnight of manual assembly — and the assumptions (uptake anchors, growth caps, tender outcomes, price points) have to live in one governed place where they are visible, versioned and owned by the team rather than by a formula in cell AX412.

That is the difference between a forecast the board glances at and one the business actually plans against: not a cleverer model, but a system that admits when it is wrong and makes correcting it cheap. We've built this as a working platform — see the automated forecasting case study for what it looks like in practice.

A good biosimilar forecast is not a prediction. It is a set of explicit, priced assumptions with a monthly appointment to be proven wrong — and corrected.

If your team is forecasting a UK portfolio through spreadsheets and stitching the actuals together by hand, that is a solvable problem in weeks, not quarters. Book a discovery call and bring the molecule you find hardest to call.