Pipeline reviews spend most of their time on the largest deals, because size is the only risk signal available by eye. Size is a poor proxy. A model can rank every open opportunity by the gap between what the rep expects and what similar deals have historically done, which produces a much shorter list and a much better one.
How does a model decide a deal is at risk?
It compares the deal to the historical behavior of the group it belongs to, then flags the gap.At ORM each opportunity is grouped by a machine learning model, and for each group we predict a curve for how long it will take to close. Those curves run from 1 to 80 weeks, with most of the expectation before week 12 and very few groups carrying expectation past 52 weeks. Risk is what shows up when a specific deal drifts off the curve its group is on. A deal sitting at week 20 in a group that peaks at week six is behind, regardless of what its close date says.
That framing matters because it separates risk from stage. A deal in late stage with a date three weeks out looks safe on a dashboard and can be the riskiest record in the quarter if its group closed in half that time.
Which signals actually predict a deal going sideways?
The rep changing the close date, and the absence of any change at all.The best single signal is a close date revision. A deal that slips from one quarter to the next is less likely to close even when it stays in commit, so the revision carries more information than the value in the field. The earliest signal is the lack of a signal: no stage movement, no date movement, no amount movement, no notes. From the seller's side the same thing looks like a buyer who stops returning email and stops picking up the phone.
| Signal | What it indicates | Strength |
|---|---|---|
| Close date pushed by the rep | Slippage risk and reduced close probability | Strongest available |
| Slip across a quarter boundary | Lower close probability even in commit | Very strong |
| No change in stage, close date, or amount | Deal has gone quiet | Strong and earliest |
| Deal past the timing curve for its group | Behind comparable deals | Strong |
| Amount revised down | Pricing pressure or scope reduction | Moderate |
| Stage advanced | Genuine progression | Moderate |
| Logged emails and meetings | Effort rather than progress | Weak |
What counts as meaningful activity?
A change in stage, close date, or amount. Nothing else.That definition is narrow on purpose. It restricts the signal to fields where a change reflects a decision about the deal rather than effort spent on it. Applied consistently, it also gives you an aging rule with teeth. ORM uses a twelve month rule for most customers, and typically more than 10 percent of a pipeline has not been touched in 12 months by that standard.
Those records are inventory, not pipeline, and they distort everything downstream. Coverage ratios calculated on a pipeline carrying a tenth of its value in dormant records overstate the real number materially. See pipeline coverage for how much that ratio can hide.
How short should the at-risk queue be?
Short enough that every deal on it gets worked in one review.The failure mode is a risk report that names four hundred opportunities. Nobody works four hundred deals, so the list gets scanned and ignored, and the model gets blamed for being noisy. Set the score threshold against the review capacity you actually have. Twenty deals worked properly beats two hundred deals acknowledged.
Rank by expected value at risk rather than by probability alone. A 40 percent risk score on a $200,000 deal deserves attention ahead of an 80 percent score on a $15,000 deal, and a queue sorted purely on probability buries the deals that decide the quarter.
What do you do with a flagged deal?
Assign one action with a date, then check whether the signal changed.Three actions cover most cases. Requalify, when the model is flagging a deal that never had a real buying process behind it. Escalate, when the deal is real and stalled at a level the rep cannot move. Close lost, when the record has been silent past the aging rule and nobody will say so out loud.
The last one is the hardest and the most valuable. Removing dead records does not lower your real forecast, because those deals were never going to close. It lowers the reported pipeline, which is a different thing, and it makes every ratio calculated from that pipeline honest again.
Track outcomes on the flagged cohort. Compare the close rate of flagged deals that were worked against flagged deals that were not, and you get a defensible measure of whether the queue earns its place in the cadence. Our guide on deal slippage covers the underlying pattern in more depth.
When in the quarter should this run?
Weekly, starting in week one.Of the pipeline carrying close dates inside the quarter on day one, roughly 20 percent actually closes in that quarter. That means 80 percent of the value sitting in the quarter on day one is not realized in it. A risk queue that starts in week eight is triaging a quarter that has already been decided.
Getting the forecast right in the last week of the quarter helps nobody, and the same applies here. The value of a model scored risk queue is the number of weeks it gives you to do something, which is why the first review should happen before the first deal has closed. The broader cadence sits in our sales forecasting best practices guide.
Frequently Asked Questions
How does AI identify at-risk deals?
It scores every open opportunity against the historical behavior of similar deals, then ranks the gap between what the rep expects and what deals like this one have done. The strongest single input is a close date change by the rep, and the earliest input is the absence of any change at all.
What is the best predictor that a deal will slip?
A rep changing the close date. Once a deal slips from one quarter to the next it is less likely to close at all, even when it sits in commit. The revision carries more information than the date itself.
How do you spot a deal going quiet?
Watch for the absence of meaningful activity, meaning no change in stage, close date, or amount. Logged emails and meetings do not count, because a rep can generate activity on a deal that has stopped moving. From the seller side, the equivalent is a buyer who stops returning calls and email.
How long should a deal sit before it leaves the working pipeline?
ORM applies a twelve month rule for most customers. Typically more than 10 percent of a pipeline has not been touched in 12 months, and those records inflate every coverage ratio calculated from them.
How many deals should be on the at-risk queue?
Few enough that every one gets worked in a single review. A list that names half the pipeline is a report, not a queue. Set the threshold so the queue fits the review time you have, then widen it only if the team clears the list consistently.
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