How to Run a Forecast Roll-Up Process

What is a forecast roll-up?

It is the path a number takes from a rep to the company forecast, plus the rules governing what each level may adjust.

By the time a forecast reaches a board deck it has passed through three or four people, and each one changed it. Most companies have the path and not the rules. Reps submit, managers adjust, a segment leader adjusts again, and the CRO applies a final haircut. Nobody can reconstruct how the starting number became the ending number, so when the quarter misses, the postmortem produces opinions instead of causes.

The fix is structural. Each level submits its own number, the adjustment it applied is recorded separately, and every version is preserved.

What should each level add?

Each level should add information the level below it does not have. A level that adds nothing should not exist in the roll-up. A manager who submits the sum of their reps is adding latency, not judgment, and their step can be automated.

What does the submission schedule look like?

Fixed days, fixed times, and a freeze that makes week over week comparison possible. Archiving every version is what makes the rest of this measurable. Without stored submissions you cannot calculate bias by level, and bias by level is the most valuable output of the whole process.

How should adjustments be handled?

Record every override as its own line, never folded into the number.

A manager submitting $4.2 million against reps totaling $4.6 million has made a $400,000 judgment call. That judgment is worth capturing because it can be evaluated. Over six quarters you will learn whether this manager's adjustments improve accuracy or simply move the number toward what leadership wants to hear.

Require a one-line reason on each adjustment. Two reasons cover most legitimate cases: a named deal the manager does not believe will close, or a known pattern in a rep's history. Adjustments without a reason are guesses, and guesses that stack across four levels produce a company number nobody can defend.

Then run the comparison that most teams avoid. Put the model forecast next to the human roll-up and reconcile at the deal level. The gap is always made of specific opportunities, and finding out which ones the model doubts is more useful than arguing about which forecast to report. That reconciliation is where sales forecasting stops being a submission ritual and starts being analysis.

How do you measure whether the roll-up works?

Track accuracy and bias separately, at every level, by week of quarter.

Accuracy is how close the number was. Bias is whether the error points the same direction every time. They are different problems with different fixes. A manager who is consistently 12 percent low is easy to correct with a coefficient. A manager who swings 20 percent in both directions is producing noise, and no correction helps.

Two measurements belong on the scorecard:

- Error by level, so you can see whether each step improved the number or degraded it. - Error by week of quarter, so you can see how early the forecast became reliable.

The second one is the one people skip. Careful manual forecasting of new and expansion business typically lands near 90 percent accuracy, and it takes a great deal of effort to produce and does not update as conditions change between refreshes. A forecast that reaches that accuracy only in the final weeks arrives after every decision it could have informed. Judge the process on what it knew at day 1 and day 30, which is what forecast accuracy measured across the quarter actually reveals.

What breaks a roll-up?

Three failures, all of them procedural.

Inconsistent definitions come first. When commit means "I will resign if this misses" to one manager and "probably" to another, the aggregated number means nothing. Publish written definitions for each category with the evidence required to use it, and audit against them quarterly.

Late submissions come second. A manager who submits after the deadline forces someone to estimate on their behalf, and that estimate becomes the number in the deck. Hold the deadline and report the previous submission when someone misses it.

Adjustments without records come third, and it is the most damaging because it is invisible. Every unlogged change removes your ability to improve the process, since improvement depends on knowing whose judgment helped. Getting these three right is more valuable than any modeling work, which is why sales forecasting best practices start with the operating discipline rather than the math.