What pipeline forecasting is
Pipeline forecasting is the process of predicting how much revenue will close in a given period, usually a quarter, based on the open opportunities sitting in a CRM. Every opportunity carries a stage, an amount and an expected close date. Forecasting takes that raw pipeline and turns it into a single number a CRO can put in front of a board.
That number matters beyond reporting. It drives hiring plans, marketing spend, and how much capacity a company commits to before revenue actually lands. A forecast that is wrong by a wide margin does not just embarrass the person who presented it. It misallocates budget for the whole company.
Why forecasting is harder than it sounds
The core problem is that a forecast is built from inputs supplied by the people whose compensation depends on how those inputs look. A rep entering stage and close-date data on their own deals has an incentive to make the picture look better, or in some cases worse, than reality.
The industry term for the second failure is sandbagging: a rep understates a deal's stage or probability to protect against a quota increase or to guarantee they beat the number later. The opposite failure, stage inflation, is a rep moving a deal forward before it has actually earned that stage. Both distort the same underlying data set that the forecast is built on.
Recent research shows how widespread the miss still is. In a survey of 400 CROs, CIOs and RevOps leaders at North American enterprises with 1,000 or more employees, conducted by Censuswide for Clari Labs between September and October 2025, 87% of enterprises missed their 2025 revenue targets despite record AI investment (Clari, 2026). That gap sits behind most of the forecasting methodology below, since each approach is, at bottom, an attempt to close it.
The four main approaches
Rep-committed forecasting. The rep or their manager assigns a subjective category to each deal: usually something like Pipeline, Best Case, Commit and Omitted. Salesforce's own documentation defines these categories plainly: Pipeline covers all open opportunities, Best Case includes the amount likely to close plus anything already in Commit, Commit covers the amount a rep is fairly sure will close, and Omitted excludes an opportunity from the forecast entirely (Salesforce, 2026). This is the oldest and still most common method. It is fast to run and easy to explain to a board. It is also only as accurate as the judgment, and the incentives, of the person making the call.
Weighted pipeline. Each open deal's amount is multiplied by a probability tied to its stage, for example 20% at discovery, 60% at proposal, 90% at verbal commit, and the results are summed. This removes some subjectivity but assumes every deal at a given stage behaves like the historical average, which stops being true the moment a sales process or buyer base changes.
Historical and time-in-stage modeling. Instead of a fixed probability per stage, this method looks at how long deals of a similar size and stage have historically taken to close or die, and weights the forecast by that pattern. It is more resistant to a single rep's optimism, but it needs a large enough deal history to be statistically meaningful, which is a real constraint for a team that has not closed many hundred deals.
AI and signal-based forecasting. Platforms such as Clari, Aviso and Salesforce's own Einstein forecasting layer feed the model on activity data beyond the CRM fields a rep types in: email response times, meeting frequency, multi-threading across a buying committee, and calendar activity near a close date. Gartner surveyed 227 chief sales officers between August and September 2025 and found that organizations giving sellers AI-enabled next best actions were 2.6 times more likely to achieve strong commercial growth than those that did not, and it forecasts that by 2027, 95% of sellers' research workflows will start with AI, up from under 20% in 2024 (Gartner, 2026). The forecasting layer benefits from the same underlying activity data these tools already capture for other purposes.
What to look at when evaluating a forecasting approach
The methodology matters less than the discipline behind it. A few checks apply regardless of which method a team runs: does the model reconcile the rep-submitted number against a system-generated one, and is the gap between them tracked over time rather than argued about live in a forecast call? Are stage-to-category mappings documented and consistent across teams, or does each manager define "Commit" differently? Does the model account for multi-quarter slippage, meaning deals that keep moving one quarter to the right, rather than treating each quarter as a fresh start? Is there an audit trail showing when a deal's stage or category last changed, so a late jump from Pipeline straight to Commit the week before quarter-close is visible rather than hidden?
What commonly goes wrong
Most forecasting failures are not a math problem. They are a data hygiene problem wearing a math costume. A weighted pipeline model is only as good as stage discipline; if reps advance deals to unlock a higher probability weighting rather than because the deal actually progressed, the model inherits that bias with a false sense of precision attached.
The second common failure is treating the forecast call as the only checkpoint. When a team's most rigorous review of pipeline health happens once a week or once a month, small distortions compound before anyone catches them. Teams that revisit and recalibrate forecast inputs more frequently tend to catch stage inflation and slippage earlier, before it shows up as a quarter-end miss.
The third is confusing a more sophisticated model with a more accurate one. An AI-based forecast trained on activity signals still depends on the CRM data underneath it being entered correctly. No model corrects for an opportunity that was never logged.
Sources: Clari · Gartner · Salesforce