Forecasting You Can Defend

Stage-weighted forecasts

The most common method multiplies each open deal's value by a probability attached to its stage and adds the results. A deal worth 100,000 in a stage with a 30% probability contributes 30,000. Summed across the pipeline, the weighted total is an expected value: across many deals, about that much should close.

Two conditions make it honest. The stage probabilities must come from your own history, the share of deals that entered each stage and later closed won, not from a template or a hope. And the stages must mean something, which is why Lesson 1.2's exit criteria come first. With inflated stages, every weighted number is inflated with them.

The weighted forecast's weakness is that it describes a portfolio, not any single deal. A pipeline with one huge deal at 50% will forecast half its value, while the real outcome will be all or nothing. That is fine for planning capacity across many deals and dangerous for promising a quarter that depends on a few. Matthew Dixon's 2022 research is a useful sanity check on baselines: across his dataset, average win rates were 26%, with high performers near 60%. If your stage probabilities imply much higher overall win rates than your history shows, they are wrong.

Committed views and scenarios

The second method ignores probabilities for most of the pipeline and asks a narrower question: what is already signed, or as good as signed, and when will it be collected? A committed view counts only deals in stages that mean signed and collectible, at full value on their expected dates. It understates the eventual outcome but it is rarely wrong, which is exactly what finance needs for cash planning.

Good forecasting presents both, as scenarios. A worst case built from committed deals only. A base case of committed deals plus the weighted remainder. A best case that adds plausible upside. The spread between them is itself information: a quarter whose base case depends mostly on uncommitted deals is a riskier quarter than one whose base case is mostly committed, even if the base numbers are equal.

The discipline that makes scenarios useful is never letting them blur. Label each number with how it was built, and do not add numbers built different ways into one total.

Momentum, and measuring accuracy

A stage baseline treats every deal in a stage as equal, but they are not. A deal where the buyer replied yesterday and met twice last week is more likely to close than one silent for a month in the same stage. A momentum adjustment raises or lowers a deal's probability according to observable engagement, and it must be applied to the stage baseline, not to the probability a seller typed in, because a seller's number often already reflects their sense of momentum and applying it twice compounds the optimism.

Whatever method you use, score it. Rob Hyndman and George Athanasopoulos's Forecasting: Principles and Practice defines the forecast error as the actual value minus the forecast. The mean absolute error (MAE) averages the absolute errors in the units of the data. The mean absolute percentage error (MAPE) expresses errors as percentages so they can be compared across scales, but it has known problems: it is undefined when an actual value is zero, extreme when actuals are near zero, and it penalizes over- and under-forecasts asymmetrically. Scaled errors such as MASE compare your forecast with a simple benchmark, and a value below one means you beat it.

For a sales team, a practical scheme is to record each month's or quarter's forecast before the period starts, then compare it with the actual. Track MAE in currency and the direction of the error. A forecast that is consistently high is biased, which is a different problem from one that is noisy.

Forecasting in Kavanah

Kavanah implements each of these views separately and labels them.

The weighted view. get_pipeline_summary returns, stage by stage, deal count, total amount, total revenue, and probability-weighted versions of both, plus the open ('capturable') pipeline and won totals. Its weighting uses each deal's probability, falling back to the stage's default probability when a deal has none, which is why Lesson 1.2 asked you to set stage defaults from history.

The momentum-adjusted forecast. The Deal Room's Forecast view, the get_deal_forecast tool and an optional weekly forecast email share one engine. A deal's probability is its stage's baseline multiplied by momentum divided by 50, capped below 100, where momentum (50 is neutral) is built from counterparty emails, calls, meetings and document views on the deal's timeline, time in stage, and a passed close date. It uses the stage default, never the deal's own probability, precisely to avoid compounding. A deal with no recorded activity is treated as neutral rather than penalized, and the report says how many deals that applies to. Revenue is each deal's recorded fee, or the deal size times an assumed fee percentage when none is recorded, and every line says which. The output gives expected revenue by closing week, the top five revenue risks with a recommended action, and the upside if deals advance a stage. A forecast hold excludes a deal deliberately, and automatic daily recalculation is opt-in. None of it uses AI credits.

Scenario charts. generate_forecast_charts renders email-ready best, base and worst scenarios. Given a list of committed stages, signed and collectible, worst is those deals only at full value on their close dates, base adds every other deal at its recorded probability, and best adds an uplift; without committed stages it uses momentum rules instead.

Kavanah does not score forecast accuracy for you. Save each period's forecast (the weekly email is a convenient record) and compute the error yourself at period end.