Guide
When the Forecast Keeps Missing: Diagnosing a Repeated Miss Before the Next Board Meeting
How to separate a forecasting problem from a demand, conversion or definitions problem when a PE-backed company misses its number several periods running.
A repeated forecast miss is a measurement problem until the evidence says otherwise. When a company misses its number once, the useful question is what went wrong in the period. When it misses three periods running in the same direction, the forecast itself is the thing under test, because a prediction that is wrong the same way every time is not describing the business it claims to describe. Diagnosing that means separating four causes which produce identical-looking misses and which are fixed in completely different ways.
One miss is an event, three in a row is an instrument problem
A single period’s shortfall is attributed inside the normal operating rhythm, by reading the revenue line back to the inputs that produced it. That work belongs to the monthly commercial operating review and does not need a separate exercise.
Repetition changes the question. Scatter in both directions, of roughly equal size, is noise around a method that works. Error in the same direction every period is bias, and bias is a property of the method rather than of the quarter. Before any analysis, write out the last four to six periods in three columns: the forecast at period open, the forecast at mid-period, and the actual. If the mid-period forecast is consistently closer than the opening one, the company can see a period coming but cannot see it early enough to act. If both are equally wrong, nobody is seeing it at all.
That table also sets the stakes correctly. A sponsor reads a third miss differently from a first. The first is an operating result. The third is a question about whether anyone in the company knows what is going to happen next, which is a harder thing to recover and a slower one. In S&P Global’s 2026 survey, 71 percent of GPs and 53 percent of LPs put operational value creation ahead of other sources of return, so the forecast is not read as a reporting artifact. It is read as evidence of operating control.
Four causes that look the same on a summary slide
A forecasting problem means the number was never achievable on the evidence available when it was set. A plan built from the top down to meet a required outcome, or deal probabilities applied to a book that does not behave the way those probabilities assume, will miss indefinitely no matter how well the company sells.
A demand problem means the opportunities never arrived in the volume the plan assumed. The selling worked on what it was given, and there was not enough of it. Diagnosing that on its own terms is its own exercise, covered in diagnosing a demand drop.
A conversion problem means the opportunities arrived and then did not become revenue in the period, because they were lost, because they closed smaller than assumed, or because they closed later than the cycle length in the model.
A definitions problem means the forecast and the result are not measuring the same object, so the gap between them is partly or entirely an artifact of counting.
On a board slide all four read as revenue below plan by some amount, with a variance commentary that could have been written about any of them. They diverge completely at the point of action. Buying more demand fixes the second, does nothing for the first and third, and makes a capacity-constrained business worse.
The test that separates them, from data you already hold
The fastest separator is a cohort trace, and it uses records the company already has.
Take every opportunity that was inside the forecast at the open of each missed period. Follow each one to where it stands today, and sort the list into four outcomes: still open, lost, won at a smaller value, and won in a later period.
The distribution is the diagnosis. A cohort still sitting open, many months after it was forecast to close, says the cycle length in the model is shorter than the cycle length in the business, which is a forecasting problem wearing the costume of a sales problem. A cohort mostly won but won late says the same thing more loudly, because the work was real and the dating was wrong. A cohort mostly lost is a conversion problem, and the stage at which the losses cluster tells you whether it is qualification or late-stage competition. A cohort won at materially lower values is a pricing or mix problem, and the revenue question is downstream of a commercial decision made deal by deal.
Then count separately what entered the pipeline in each period against the volume the plan assumed would enter. If entry volume is short, the demand explanation is live regardless of what the cohort shows, and the two can be true at once.
Many services businesses discover at this point that they cannot run the trace, because opportunity records carry no reliable created date, no close date history, or no value at the time of forecast. That is not a failed diagnosis. It is the diagnosis. A company that cannot reconstruct what it believed three months ago has no instrument, and the first fix is to start keeping the record rather than to argue about the number.
The definitions problem, which is the one nobody looks for
This is the cause most often missed, because it does not feel like a revenue question.
In a services business, signed, booked, scheduled, delivered, invoiced and recognized are six different events, often months apart. A forecast built on one and compared against another will show a gap that nobody caused. Repeat and renewal work counted inside new business will flatter a period and starve the next. A deal entered at total contract value in one report and at first-year value in another will reconcile in neither. Across multiple sites, the same stage names routinely describe different states, so a pipeline total is a sum of things that are not alike.
The tell is simple. Ask two people for the same period’s number from two systems. If the answers differ and nobody can say within a day which is right, the definitions are the problem, or are at least part of it.
The fix is a written definition for each term, with one named owner, applied to the comparison periods as well as the current one. Restating only the future leaves the trend unreadable and hides whether anything has actually changed. Where the same terms drive what a seller is paid, the credit rules need settling at the same time, and that sits with sales quota allocation rather than with the forecast.
What can be answered before the board meeting, and what cannot
Three things are counting exercises and can be finished inside days. The cohort trace, because the records either exist or do not. The entry-volume comparison, for the same reason. The definitions reconciliation, which needs one person with access to both systems and the authority to decide which definition governs.
Three things cannot be finished that fast, and claiming otherwise is how a company ends up correcting itself twice. Whether a conversion decline is lead quality or selling needs stage-level evidence across enough deals to be more than anecdote, usually with win and loss conversations behind it. Whether a value decline is pricing or mix needs the same volume of cases. Whether the plan was achievable when it was set is not only an analysis, it is a negotiation about the assumption base, and it will not be settled by one person working quickly.
The practical consequence is that the next board meeting should receive an attribution and not a remedy. A plan presented on top of an unfinished diagnosis commits the company to spending in a direction it has not yet evidenced, and it will be remembered as the moment the number was promised again.
What to tell the sponsor about a miss you cannot yet explain
Four things, in this order, and none of them is an apology.
What the miss is, stated in the definition both sides have agreed, so the conversation is not relitigating the measurement. What has been ruled out, and on what evidence, which is where the cohort trace earns its place because it is specific and checkable. What is still open, named, with the date the answer arrives and the person who owns it. And what the company is doing in the meantime, limited to actions that are sound under every explanation still standing, with an explicit note of what is being held back and why. Declining to buy more demand before knowing it is a demand problem is a decision worth stating out loud, because the absence of visible activity after a miss reads as paralysis unless it is explained as restraint.
If the forecasting method is changing, say so and say what the next number rests on. A forecast produced the same way that has now been wrong three times will be discounted whatever it says.
A sponsor can price an unexplained miss that has a date attached to it. What costs far more is an explanation offered early, accepted, and then withdrawn, because the second correction is the point at which the number stops being believed and the attention shifts from the result to the people reporting it.
When the miss coincides with a departure at the top of the function, the diagnosis changes; see holding the revenue number when the commercial leader leaves.
Frequently asked questions
Why does a sales forecast keep missing in the same direction?
Consistent error in one direction is bias in the method rather than variance in the business. The usual sources are a cycle length in the model that is shorter than the real one, deal probabilities applied to a book that does not behave that way, a plan set top down to meet a required outcome, or a comparison between two different definitions of revenue. Scatter in both directions is noise and can be left alone. A one-directional pattern will continue until the method changes, no matter how the team performs.
How do you tell a demand problem from a conversion problem?
Count what entered the pipeline in each period against what the plan assumed would enter, then trace what happened to the opportunities that were already in the forecast. Short entry volume with normal outcomes on the cohort points to demand. Normal entry volume with the cohort clustering in losses points to conversion, and the stage where losses gather separates qualification from late-stage competition. Both can be true in the same period, which is why they are counted separately rather than inferred from the revenue line.
What should you do when the forecast and the actuals use different definitions?
Write one definition for each term, name an owner for each, and restate the comparison periods on the new basis as well as the current one. Restating only the current period leaves the trend unreadable and makes it impossible to tell whether anything improved. Until that is done, the size of the miss is not known, and no remedy aimed at the commercial team can be justified from it.
What should a private equity sponsor be told about a third consecutive forecast miss?
The miss in the agreed definition, what has been ruled out and on what evidence, what remains open with a date and an owner, and what is being done in the meantime that holds up under every explanation still in play. Say plainly if the forecasting method is changing and what the next number is built on. Naming an open question with a date is more defensible than an explanation that has to be withdrawn later.
Can a repeated forecast miss be diagnosed before the next board meeting?
Partly. The cohort trace, the entry-volume comparison and the definitions reconciliation are counting exercises and can be completed quickly from records the company already holds. Whether a conversion decline is lead quality or selling, whether a value decline is pricing or mix, and whether the original plan was achievable all need more evidence and more agreement than a short window allows. The board meeting should get the attribution that is finished rather than a plan resting on the part that is not.
Have a revenue problem the board is asking about? Start a conversation.