The Forecast Wasn’t Wrong. It Was Never Independent.

When a demand number and a sales target are the same figure, nobody can tell you’re wrong until the inventory is already sitting on the shelf or already missing from it.

A commercial operations analyst at a mid-size biopharma company once described her own forecasting process, without much prompting, as “guessing with extra steps.” She wasn’t being self-deprecating. She was describing the honest mechanics of a job she’d inherited: build a demand number for a product the company had never sold before, using a method designed for a product it had been selling for years.

That distinction between a method that works and a method applied to the wrong situation turns out to matter more than almost anything else in early-launch forecasting. And it’s easy to miss, because the forecast doesn’t look broken. It looks like a spreadsheet, updated on schedule, producing a number every month. The problem isn’t visible in the process. It’s visible six months later, in the inventory.

The Visible Problem

Here is what this typically looks like from the inside. A company launches its first or second commercial product. Sales sets a volume target for the year, because that’s what sales teams do targets drive compensation, territory planning, and board conversations. Somewhere downstream, that same number gets picked up by whoever owns supply planning and dropped into a forecasting spreadsheet, often the same spreadsheet used for the company’s established products.

Nobody sits down and decides to use the sales target as the demand forecast. It happens by inheritance. The trend-based method that works fine for a stable product doesn’t ask “what number should we forecast” it asks “what’s the trend,” and for a five-month-old product, there is no trend yet, only noise. So the method quietly falls back on the only stable-looking number in the building: the target.

The consequences show up gradually and then all at once. Finished-goods inventory builds ahead of actual uptake. Someone eventually notices the gap between shipments and stock. A write-down conversation happens. Then, because the correction usually overshoots in the other direction, a stockout risk appears a quarter or two later and because contract manufacturing lead times commonly run ten to twelve weeks, there’s rarely enough runway to fix it without an expedited shipment.

Industry data suggests this pattern is closer to normal than exceptional. Research on launch-period accuracy has found that even <cite index=”6–1">only about half of product launches meet or exceed prior expectations</cite>, and one large analysis of forecasts made shortly before launch found that <cite index=”8–1">actual sales diverged from those predictions by a median of roughly 71%, with a majority overstating eventual results by more than 160%</cite>. This isn’t a story about one company’s analyst missing something obvious. It’s a structural feature of forecasting a market that doesn’t fully exist until the product is already in it.

Why Common Fixes Fall Short

The instinctive responses to this problem are usually reasonable and usually insufficient on their own.

Adding a new forecasting tool or software platform is the most common reflex. But a tool doesn’t resolve a data-independence problem; it just gives the same inherited assumptions a nicer interface. If the underlying number is still the sales target in disguise, a more sophisticated system will simply produce a more sophisticated version of the same wrong number.

Hiring a dedicated demand planner is the second instinct, and it’s not wrong exactly, it’s often premature. A company with one or two commercial products frequently doesn’t have a capacity problem. It has a method problem. Adding headcount before fixing the method risks institutionalizing a flawed process with more people defending it.

A third common response is to tighten the review cadence more meetings, more check-ins without changing what’s actually being reviewed. Reviewing a bad number more often doesn’t make it a better number. It just makes everyone more aware, more often, of a discrepancy nobody has explained.

And a fourth: treating the forecast as a single company-wide process to be redesigned top to bottom. For a smaller organization, this tends to be significant over-engineering relative to the actual problem, which is usually contained to one product at one stage of its life.

Each of these responses is attractive because it feels like action. None of them addresses the actual mechanism: a forecast with no independent check against what the market is doing.

Also Read:- The Operational Gap Hiding Behind “On Track” in Outsourced Clinical Development

The Hidden Operational Issue

The deeper pattern is less about forecasting technique and more about ownership and visibility. Three things tend to be true simultaneously, and none of them show up on a dashboard.

First, the forecast has no independent identity. When a demand number and a sales target are allowed to be the same figure, there is structurally no way to notice when they should diverge and for an early-launch product, they almost always should. A target reflects what the company wants to happen. A forecast should reflect what the data suggests is actually happening. Collapsing the two removes the only check that would catch a miss early.

Second, useful signal often already exists inside the organization and simply isn’t connected to the forecasting process. Shipment data, payer determinations, prescriber activity these get discussed anecdotally in meetings, referenced in passing, and then left out of the actual model. The gap usually isn’t a data-collection problem. It’s a data-structuring problem.

Third, the people most exposed to a forecast miss a manufacturing partner with long lead times, for instance often see the number only after it’s finalized, when there’s no longer room to raise a concern that could change the outcome. Visibility that arrives after a decision window has closed provides the appearance of alignment without the function of it.

This is a workflow finding, not a data-science one. And it’s easy to miss because everyone in the process is doing their job competently. The analyst is following the method she was given. Sales is setting a reasonable target. The gap lives in the space between roles, not inside any one of them.

A Better Way to Examine the Problem

For teams dealing with a similar pattern a newer product, a forecast that keeps surprising Supply or Finance a few questions tend to surface the issue faster than a full process audit.

1. Trace where the current forecast number actually comes from. Follow it back to its source, not its spreadsheet. If the number traces back to a sales target, an incentive plan, or a board commitment more directly than it traces back to shipment or usage data, that’s worth flagging before anything else. Ask internally: if sales changed its target tomorrow, would the supply forecast change with it and should it?

2. Inventory the signal you already have before looking for new data. Most organizations collect more relevant data than they use. Shipment data, coverage determinations, early prescriber behavior check what’s already being received and simply isn’t structured into the model. Ask: what data do we discuss in meetings but never build into the number?

3. Identify who is exposed to a forecast miss and when they currently see the number. Map every party affected by a demand miss manufacturing, finance, inventory against the point in the cycle when they actually see the forecast. If exposure and visibility don’t line up, that’s a sequencing problem worth fixing before anything else. Ask: is there anyone who only finds out the forecast changed after it’s too late to respond?

4. Separate the incentive number from the operating number, deliberately. A sales target and a supply-facing forecast can serve different purposes without being reconciled into a single figure. Making the difference between them visible, rather than resolved away, is often more useful than making them match. Ask: what would it cost us to let these two numbers legitimately differ?

5. Pilot before replacing. Running a new method alongside the old one, even briefly, turns a debate about method into a comparison of evidence. Ask: could we prove this is better before asking anyone to trust it?

What Changed When the Problem Was Viewed Differently

The most useful shift wasn’t a new technology or a new team member. It was reframing the gap between a sales target and a supply forecast from something to be explained away into something expected and informative. Once that gap was allowed to exist openly, the incentive to quietly let the forecast track the target disappeared.

The other shift was recognizing that the fix didn’t need to be built from scratch. The data that mattered most was already flowing into the organization; it simply hadn’t been connected to a decision. That reframing from “we need new inputs” to “we need a structure for the inputs we have” is often the more realistic starting point for a smaller organization than assuming a data or systems gap.

None of this fully resolved the underlying uncertainty of forecasting a product with limited history. Early-launch forecasting research generally shows that <cite index=”9–1">simple trend extrapolation still improves materially on static targets, even without more sophisticated modeling</cite> — which suggests the value here wasn’t in a more advanced technique, but in removing a distortion that had nothing to do with technique at all.

Questions Leaders Should Ask

• Is our demand forecast, for any product, functionally the same number as a sales or incentive target?
• Can we trace our current forecast back to its actual data source, or only back to a spreadsheet?
• What early signal do we already collect but don’t structure into a decision?
• Who is most exposed to a forecast miss, and when do they currently find out the forecast has changed?
• Are we about to add a tool or a hire to solve what might actually be a definition problem?
• If our sales target and our supply forecast diverged sharply next month, would anyone notice before a customer or manufacturing partner did?
• Have we tested a new method against the old one with real data before committing to switch?

A Closing Thought

The most expensive assumption in this pattern isn’t a bad data source or a missing headcount. It’s the quiet decision usually made by no one in particular to let a target and a forecast become the same number. Once that happens, the organization loses its earliest warning system, and the first sign of trouble becomes an inventory conversation with the board instead of a data point on a dashboard.

The full case study examines how this particular engagement traced that gap through interviews and shipment data, what the team found when they tested different early-signal candidates against real results, and what happened including where the rollout didn’t go as smoothly as planned when the two numbers were finally allowed to diverge.

Read the full case study: *Demand Planning That Supports Real Decisions*