An erosion curve has three parameters, the entry date, the depth it settles at and the speed it gets there, and only the speed reliably transfers from your last event. Grabowski and Vernon (1992) documented brand prices holding or rising after generic entry while volume collapsed, so a supply plan derived from the revenue plan will be wrong in one direction or the other. The decision worth planning is whether to run one more campaign, because the minimum campaign frequently exceeds the entire remaining sellable demand once expiry dating is applied. After entry the cost of a shortfall falls and the cost of excess rises, which moves the target well below the service level the product carried under exclusivity. Discontinuation notice periods and tender commitments make the tail a contractual minimum rather than a forecast.
The demand plan for the product that loses exclusivity in fourteen months still shows a gentle upward trend. The statistical model built it from four years of history, every month of it under patent, and it has extrapolated the only pattern anyone ever showed it. Meanwhile the site has a campaign booked in nine weeks that will produce something close to eleven months of cover at the current run rate, and the active ingredient for the campaign after that has to be ordered in March.
Somebody has to decide whether that second campaign happens. The decision is worth more money than any refinement to the statistical model will ever be, because it cannot be reversed and the model can.
The forecast cannot see the cliff and should not be asked to
Every method in the standard library extrapolates structure it has already observed. Exponential smoothing, the state space models underneath it, and anything trained on the item's own history are all doing the same thing in different notation. A loss of exclusivity is a structural break with an approximate date and no precedent in the series, which puts it outside what any of them can produce. Fitting a longer history makes the problem worse, since more pre-cliff data means more confidence in the pre-cliff trend.
The workable arrangement is to keep two series. The statistical forecast produces the counterfactual, meaning what the product would have done had exclusivity continued, and the erosion assumption is applied on top as a multiplicative overlay with a named owner and a documented basis. The assumption ledger discipline covered in a separate post applies directly here, and the override governance in another is what keeps the overlay from quietly becoming the forecast.
The practical benefit of that separation shows up when the entry date moves, which it will. You shift the overlay by three months and the baseline is untouched. When the two are baked into one manually adjusted number, moving the date means rebuilding the whole series and nobody can reconstruct what changed.
Erosion is three parameters and only one of them transfers
The entry date is the first parameter and the least knowable. Patent expiry is the earliest legal date in the simple case, and the real case involves paragraph IV certifications, settlements whose entry dates are confidential, at-risk launches, and supplementary protection certificates in Europe that expire on different dates in different member states. Under the Hatch-Waxman Act of 1984 the first eligible generic filer can hold 180 days of exclusivity, so the first half year may have one competitor and the seventh month may have five. Treat the date as a distribution across several months and build the plan for at least the earliest credible date and the median one.
The second parameter is depth, meaning the share the brand settles at once the market has finished reorganising. Depth is largely a function of how many entrants arrive, and that number is unknown when the plan is made. Grabowski and Vernon (1992) in the Journal of Law and Economics, and the US Federal Trade Commission's 2002 study of generic entry, both found generic prices falling further as the number of entrants rose, with the steep part arriving once several were established. A depth assumption is therefore a competitor count assumption wearing different clothes, and it should be written down that way.
The third parameter is speed, usually expressed as a half-life in months, and this is the one that carries over from your previous events. Speed is set by the substitution mechanics of the channel rather than by the molecule. Where most volume moves through a channel with automatic pharmacy substitution, the half-life is short, and it will be short again next time. Where volume moves through hospital tenders that renew annually, erosion is a staircase whose steps are the tender dates, and the useful planning parameter is the tender calendar rather than a smooth curve.
Biologics behave differently for a reason that is structural. The Biologics Price Competition and Innovation Act of 2009 created a separate interchangeability designation in the United States, and without it a biosimilar cannot be substituted at the pharmacy counter, so each switch requires a prescriber decision. Erosion that depends on thousands of individual prescribing changes runs on a slower clock than erosion that happens automatically at dispensing.
Price and volume come apart after entry
Grabowski and Vernon (1992) and Frank and Salkever (1997) both documented the pattern usually called the generic paradox. After generic entry a brand's price frequently holds or rises even as its volume share collapses, because the buyers who remain are the least price sensitive part of the market and the brand stops competing for the rest.
That has a direct planning consequence. The revenue plan and the supply plan need separate erosion curves, one on units and one on realised net price, and deriving either from the other at a constant price will produce a material error.
Take a product at 100 units and 100 of value. In the first case the brand defends formulary position with discounts, and units settle at 40 while net price settles at 55, so value lands at 22. A planner who takes the finance number and divides by last year's price plans 22 units against real demand of 40 and misses nearly half of it. In the second case the brand holds price on a loyal residual, units settle at 20 and net price at 105, so value lands at 21 and the same derivation plans 21 units against a real 20. The second answer happens to be close and the arithmetic that produced it was wrong both times.
Gross-to-net movements make this sharper, because rebates tend to escalate at exactly the point where the brand is fighting to hold contracted positions. The unit series and the value series diverge most in the months where the campaign decision is being argued.
The last campaign is the decision worth planning
Here is where the arithmetic earns its place. Take a solid oral dose product made in campaigns on a shared line. The validated minimum campaign is eight batches, a batch yields 400,000 tablets, and there are 30 tablets in a pack, so the smallest campaign anyone can run produces about 106,000 packs. Baseline demand is 12,000 packs a month, which makes that campaign roughly nine months of pre-cliff cover.
Shelf life is 36 months from manufacture, and the wholesaler contract requires 18 months of remaining dating on receipt, so anything made today has an 18 month window in which it can actually be sold. Now apply an erosion curve with a three month half-life falling to a residual share of 15%, and integrate demand across those 18 months. Month one is around 9,900 packs, month three around 6,900, month six around 4,350, month twelve around 2,440, and month eighteen around 1,960. The whole 18 month sellable window sums to roughly 71,000 packs.
The smallest campaign anyone can run is 106,000. It exceeds the entire remaining sellable demand by about 35,000 packs before you count a single unit of the stock already sitting in the network from the previous campaign.
The question stops being how much to make and becomes whether to make anything at all. The levers that change the answer are the minimum campaign size and the dating requirement, both of which are technical and contractual matters that take years to move, so the conversation has to start well before the plan does. Splitting a campaign across several markets is the usual escape, and it only works where the packs are actually interchangeable.
The cost of being wrong changes direction at entry
Under exclusivity, running out is expensive and holding a few extra months of a high margin product is cheap, so the target service level sits very high and nobody argues about it. After entry both terms move, and they move in opposite directions.
The cost of a shortfall falls, because the margin on each pack has dropped and the customer you fail to serve was leaving anyway. The cost of excess rises, because unsold packs now expire rather than sell late, and pharmaceutical disposal carries a controlled destruction cost on top of the write-off. Feed both into the standard critical ratio and the target moves from something like the ninetieth percentile of the demand distribution to something well below the median. A separate post covers how that ratio is constructed.
Most organisations never make the adjustment. The service target on the product is a field in a master data record that was set years ago and nobody has authority to lower it, so the plan keeps protecting a high service level on a product where high service is now the expensive mistake. Changing that field is a five minute job and it is worth more than a quarter of forecasting work.
Expiry dates arrive country by country and stock is not fungible
Exclusivity does not end everywhere at once. Supplementary protection certificates in Europe are granted nationally and expire on dates that differ between member states, and Regulation (EU) 2019/933 added a manufacturing waiver that lets generic makers produce inside the EU during the certificate term for export, and stockpile during the final six months for day one entry. From a brand planner's seat, that regulation is a statement about the shape of the curve. Competitors can have finished, released stock waiting on the day protection ends, which compresses the early part of erosion into weeks.
Surplus in a still-protected market usually cannot be moved into one that has gone generic, because the pack, the leaflet and the authorisation number differ by country, a boundary a separate post sets out in full. What does move is parallel trade. Once a national price falls, product starts flowing toward higher priced neighbours, and the erosion overlay for those markets needs a term for imported volume that has never appeared in their own history.
The tail is contractual before it is commercial
Stopping is not a decision the commercial team can take alone. Section 506C of the US Federal Food, Drug, and Cosmetic Act, as amended by FDASIA in 2012 and the CARES Act in 2020, requires manufacturers of certain drugs to notify the FDA at least six months before a permanent discontinuance or an interruption likely to disrupt supply. In the European Union, Article 23a of Directive 2001/83/EC requires notification to the national competent authority, with two months' notice for ceasing to market a product. Tenders already awarded carry committed volumes with penalties, and some national contracts run past the point where the product is worth making.
So the tail enters the plan as a minimum quantity with dates attached rather than as a forecast. The last build has to cover that minimum with adequate remaining dating at every delivery point in it, and that constraint sometimes forces a campaign the erosion curve says is uneconomic. Finding this out after the campaign wheel has been fixed for the year is a common and avoidable failure.
Where this stops
The entry date is genuinely unknowable, settlements are confidential, and no amount of planning rigour changes that. What rigour does is sort the decisions by whether they differ across the scenario range. Most of them do not, and the ones that do concentrate almost entirely in the campaign and long lead ingredient commitments, which is where the argument should go.
Your own history is a small sample. Three previous events give you three curves, which supports a range for the half-life and does not support fitting anything. Any curve that arrives with four decimal places has been over-specified somewhere upstream.
And there is a limit no forecasting improvement reaches. When the minimum campaign exceeds the entire remaining sellable demand, as it did in the arithmetic above, the answer does not change if the erosion assumption is off by twenty percent in either direction. The binding constraint is a validated batch size, and moving it means a technical change with regulatory work behind it, started years before the cliff by someone who was looking that far ahead.
Rebase your last three loss of exclusivity events to month zero at first competitor entry, plot monthly units as a share of the twelve month pre-entry average, and read the half-life off each of the three curves.
More on this in industry on the DemandLabs platform.
