Every forecast gets adjusted before it reaches the board. A rep submits, a manager trims, a VP trims again, and the CRO applies a final view. Each adjustment feels responsible in isolation. Stacked, they produce a number with no traceable relationship to any deal, and they make it impossible to tell whose judgment was worth anything.
An override policy does not ban judgment. It makes judgment visible, attributable, and scoreable.
What counts as a forecast override?
Any change to a submitted number that is not accompanied by a change to an underlying deal. Moving a deal out of commit because the buyer went dark is a deal change. Subtracting 12 percent from a rep's roll-up because that rep usually runs hot is an override.The distinction matters because the two behaviors need different treatment. Deal changes belong in the CRM and flow through automatically. Overrides belong in a separate, labeled line that sits next to the raw roll-up and gets scored on its own.
Silent overrides are the version that causes damage. When the adjustment is baked into the submitted total, rep-level accuracy tracking breaks, because the number you score was never the number the rep produced.
Who should be allowed to override, and at what level?
One role, one level, once per snapshot. Pick the level that has the most deal context and the least distance from the buyer, which in most B2B SaaS organizations is the frontline manager.| Pattern | What it produces | Verdict |
|---|---|---|
| Rep submits, manager adjusts, VP adjusts, CRO adjusts | Untraceable number, compounded conservatism | Retire it |
| Rep submits, one adjustment at manager level, logged | Traceable judgment, scoreable separately | Use this |
| Rep submits, no adjustment permitted anywhere | Clean data, no capture of manager knowledge | Workable but wasteful |
| Adjustment applied only at CRO level before board | Late, blunt, no deal context | Retire it |
What evidence should an override require?
A named deal and a written reason. Never a blanket percentage.A percentage haircut applied to a total is an average of last year's error dressed up as this quarter's insight. It cannot be checked after the fact, because there is no claim to check. When a manager says "I am moving this deal out of commit because the economic buyer has not engaged since the 3rd," that claim resolves within weeks, and you learn something either way.
Two override reasons deserve automatic acceptance because they track the strongest known signals. The first is a close date change made by the rep. That is the single best predictor of slippage, and a deal that moves from one quarter to the next is less likely to close even while sitting in commit. The second is the absence of any signal at all: no stage change, no amount change, no buyer reply. Meaningful activity means a change in stage, close date, or amount, and a committed deal with none of those in the last two weeks has earned a downgrade.
For the underlying patterns, see deal slippage.
How should overrides be recorded?
In a structured field with four attributes: deal ID, direction, dollar amount, and reason code. Free-text notes are not an audit trail, because you cannot aggregate them at quarter end.Keep the reason codes short and mutually exclusive. Buyer went quiet. Legal or procurement delay. Budget freeze. Champion departed. Competitive displacement risk. Amount likely to shrink. Six codes handle most of what actually happens, and a short list gets used while a long list gets ignored.
At quarter end, aggregate by reason code. If one code accounts for more than a third of your override dollars, it points at a process gap rather than a set of unlucky deals. Champion departures clustering in one segment means multi-threading is thin. Procurement delays clustering at quarter end means the close plan starts too late.
How do you tell whether overrides improve accuracy?
Run both numbers forward and score them separately for at least six quarters. Archive the pre-override roll-up and the post-override number at every snapshot, then compare each against actuals.Three outcomes are possible, and each has a clear action. If the override consistently beats the raw submission, the manager judgment is adding value and should be kept and formalized. If the override consistently loses to the raw submission, the habit is costing accuracy and should be retired. If the override wins on absolute error but carries a constant signed bias, the manager is applying a fixed correction, which a model can do better and more consistently.
That third case is more common than most leaders expect. A standing 10 percent haircut is not judgment. It is a hardcoded correction for a structural problem, usually stage weights that no longer match real conversion. Fix the weights and the haircut becomes unnecessary. Our note on weighted pipeline covers how to recalibrate them.
What target should the policy be measured against?
Set the bar at the accuracy a well-run manual process delivers, then ask whether your overrides are getting you there. On new and expansion business, accuracy around 90 percent is typical, though producing it takes heavy manual effort and the number stops being true as conditions shift. ORM targets 95 percent without manual adjustment, holding from day one to day ninety of the quarter and updating as the quarter progresses.The relevant question for an override policy is whether human adjustment is closing that gap or papering over it. A team applying large, frequent overrides is telling you the underlying forecast is broken. The overrides are a symptom. Fixing stage definitions, stripping stale pipeline, and rebuilding conversion assumptions on real history will shrink the size of the adjustment more reliably than getting better at adjusting.
When should the policy allow no override at all?
On the day-one forecast. The day-one number is the most valuable one you produce, because it is early enough to change the quarter. Getting the forecast right in the final week helps nobody, since the quarter has already happened by then.Leave the day-one snapshot raw. Score it raw. It is the cleanest read you will get on whether your process, your data, and your assumptions work without a human smoothing the output. Once you have four quarters of raw day-one accuracy, you will know exactly how much of your forecasting capability is process and how much is one experienced manager's instinct. For the definitions underneath all of this, start with forecast accuracy.
Frequently Asked Questions
What is a forecast override?
An override is any change a manager or leader makes to a submitted forecast without a corresponding change to the underlying deals. The most common form is the haircut, where a manager subtracts a percentage from a rep roll-up because experience says the roll-up runs hot.
Should managers be allowed to override the forecast at all?
Yes, at one level only, and always as a visible separate line rather than a silent edit to the roll-up. Judgment carries real information about deals the data cannot see. Hidden judgment carries the same information and destroys your ability to measure anyone's accuracy.
How many override levels should a company have?
One. Cascading haircuts at rep, manager, VP, and CRO level compound into a number nobody can trace to a deal, and they make rep-level accuracy scoring meaningless because the submitted number was never the reported number.
How do you know whether overrides are helping?
Score them. Track the accuracy of the pre-override roll-up and the post-override number separately across at least six quarters. If the override does not beat the raw submission, the habit is costing accuracy and should be retired.
What evidence should an override require?
A named deal and a stated reason, not a percentage applied to a total. Overrides that name deals are checkable after the fact. Overrides that apply a blanket percentage cannot be evaluated, and they usually encode last quarter's error rather than this quarter's risk.
See how ORM turns these insights into action
ORM builds custom revenue forecast models for B2B SaaS companies. Not dashboards. Prescriptive analytics that tell you what to do next.
Schedule a Demo