Why stage-weighted forecasts drift
Stage-weighted forecasting assumes stage is a proxy for probability and that stage is current. Both assumptions decay quietly. A deal enters a late stage on the strength of one good call and stays there while the champion goes silent, because nothing in the system disagrees with a field a human set.
- ✦Stage is set by a person with an incentive and is rarely revised downwards.
- ✦The date on the record is the date someone last touched the record, not the date something happened.
- ✦Aggregate accuracy can look acceptable while the composition is wrong in both directions.
Evidence the model should be reading
Everything needed to challenge a stage is already in the workspace, provided conversations, campaign response and payments live on the same records as the pipeline. The forecast becomes useful when it reads those rather than only the stage field.
- ✦Deal aging by stage — time in stage against your own historical distribution.
- ✦Engagement direction — whether the number of people engaged is growing or narrowing.
- ✦Conversation evidence — what the last call actually established, from the transcript rather than a note.
- ✦Commercial motion — whether a quote exists, whether it was opened, whether pricing has moved.
- ✦Silence — the strongest and most consistently ignored signal in any pipeline.
Revenue analyticsEvidence from calls
Make the assumptions visible
A forecast anyone can interrogate beats a more accurate one nobody understands, because the point of the meeting is to decide what to do. Ivorycom's approach is to surface what changed, which deals carry the most risk, and which assumptions the number depends on — so the conversation is about the deals rather than about the spreadsheet.
- ✦What moved since the last review, and what moved it.
- ✦Which opportunities the number is most sensitive to.
- ✦Where the model and the owner disagree, stated rather than averaged away.
- ✦Which deals are aging past their historical stage duration.
Forecasting is a data-quality problem first
No model survives duplicate accounts, stale ownership and quotes that live outside the record. Because quoting and payment sit on the opportunity, acceptance moves the forecast in the same action rather than waiting for someone to update a field — which removes one of the largest sources of routine forecast error.
- ✦One golden record per customer, so the same deal is not counted twice.
- ✦Quote acceptance updates the opportunity, the forecast and the finance view together.
- ✦Provenance on fields, so a disputed number can be traced instead of argued.
Quotes that update the forecast
What AI should and should not do here
Forecasting is the area where an overconfident model does the most damage, because the output goes to a board. The useful contribution is surfacing evidence and contradiction — this deal is aging unusually, this champion has gone quiet, this stage does not match the conversation. The unhelpful contribution is a single confident number with no reasoning attached.
- ✦Surface contradictions between the stage and the evidence.
- ✦Attach the reasoning to any adjustment it proposes.
- ✦Leave the call with the person who owns the number.
Frequently asked questions
What is revenue forecasting software?
Software that assembles a forward revenue projection from pipeline data. The useful kind also exposes what the projection depends on — deal aging, engagement, conversation evidence and commercial motion — so the number can be challenged rather than merely reported.
Why are stage-weighted forecasts unreliable?
They assume stage reflects probability and that stage is current. Stages are set by people with incentives and are rarely revised downwards, so a deal can sit at a high weighting while nothing has actually happened for weeks.
What evidence improves a forecast most?
Time in stage against your own historical distribution, whether engagement is broadening or narrowing, what the last conversation actually established, whether a quote exists and has moved, and how long the account has been silent.
Does the AI produce the forecast?
It surfaces evidence, contradictions and risk with the reasoning attached. The commitment stays with the person who owns the number, which is the correct division of responsibility for a figure that goes to a board.
How accurate is it?
No accuracy figure is published here. Forecast quality depends on your data completeness and sales process far more than on the model, and a vendor percentage would not describe your pipeline.
What is the fastest way to improve forecast accuracy?
Usually data quality rather than modelling: remove duplicate accounts, make quote acceptance update the opportunity automatically, and get conversations onto the record so aging and engagement can be measured at all.
Ivorycom