SolutionsRunning the team
The forecast is a story until somebody can check it.
Commit day. The number on the board is the sum of forty judgement calls made by people whose compensation depends on those judgements. You are not being asked whether the number is right. You are being asked whether you can defend it, and the honest answer is usually that you cannot.
01What is really happening
A close date is a guess wearing a uniform.
Ask where a forecast comes from and the chain terminates in someone’s feeling about a deal. Every layer above that adds precision without adding information: a percentage, a category, a roll-up. The number gets more confident as it travels upward and no more true.
- Commit, best case, and pipeline are labels, not measurements.
- Sandbagging and happy ears cancel out only by luck.
- Nobody can point at the evidence for any single date.
02What Adrata reads
Won, lost, and stalled are three different fates.
Deal timing is modelled as competing risks rather than a coin flip, because a deal that goes quiet has not lost. It has entered a different state with a different clock. The intervals come from conformal prediction, which produces ranges with a stated coverage guarantee instead of a single confident date.
- A range with a coverage guarantee, not a percentage with no provenance.
- Stalling modelled explicitly, rather than hidden inside "open".
- Which specific deals are carrying the uncertainty in the number.
03The move
Bring a number somebody can argue with.
The output is a forecast whose every component can be opened: the deal, the evidence, the interval, and what would have to be true for it to land. Where the data is too thin to support an interval, that is reported rather than smoothed over, which is the difference between a forecast and a story.
- The deals that actually move the number, ranked.
- What would have to change for a deal to land in the quarter.
- The parts of the forecast the system will not vouch for.
The model behind it
You should be able to argue with the recommendation.
A move you cannot interrogate is a move you will ignore the first time it is wrong. These are the models doing the work in this scenario.
Calibrated timing
When will this actually close, and how sure can you be?
Deal timing is modelled with competing risks: won, lost, and stalled are different fates, not one binary. The intervals come from conformal prediction, which produces ranges with a stated coverage guarantee rather than a single confident date.
Causal uplift, not correlation
Did the action cause the outcome, or did the good deals just get more attention?
Treatment effects are estimated with propensity weighting and doubly-robust estimators, adjusting for deal size, buyer-group size, whether the deal was intro-sourced, whether it was an existing customer, and known competitive pressure. Estimates that have not been adjusted for confounders are quarantined so they can never be presented as a cause.
The right to say nothing
Is there enough evidence here to justify a recommendation at all?
Learned components stay inert until they beat the simple heuristic on held-out data, and every model has a floor below which it abstains rather than guessing. On a new workspace the honest answer is often "not yet", and the system is built to say so instead of manufacturing confidence.
Where it runs
The surfaces this scenario uses.
Who this is for