Spotting deals that will slip before the forecast call

Forecast calls go wrong in the same ways: deals that have gone quiet, close dates pushed for the third time, a champion who left. A weekly AI review reads every open deal for those signals and gives the sales leader a ranked list with the evidence, before the call instead of after the quarter.

01the problem and who owns it

Sales leaders inspect pipeline by asking reps how deals are going, and reps are optimists by profession. By the time a deal visibly slips, the quarter is gone. Leaders with large teams cannot read the activity history on every opportunity, so inspection focuses on the biggest deals and misses the middle.

The head of sales owns the forecast and answers for it to the CEO and board. RevOps owns the data. What is missing is a consistent, evidence-based read of every deal, done the same way every week.

02what the AI does, step by step

1. Snapshot the pipelineThe run copies every open opportunity with its amount, stage, close date, owner, and history of changes, so week-over-week movement is measurable rather than remembered.
2. Compute hard signals with codeDeterministic checks find close dates pushed more than once, stages older than your typical duration, no logged activity in a set number of days, and amount changes. These rules come from your own historical deals.
3. Read the soft signals with a modelFor flagged and large deals, a model reads recent emails, call summaries, and notes for signs like a new decision maker, budget language, a competitor mention, or a buyer who stopped replying.
4. Score and explainEach deal gets a risk level and two or three plain reasons with links to the evidence. The explanation matters more than the score: it is what the manager will ask the rep about.
5. Rank and group the outputThe leader gets a list grouped by team and sorted by amount at risk, plus deals that improved since last week so the review is not only bad news.
6. Capture what happenedAfter the call, managers mark each flag as confirmed, explained, or wrong. At quarter end, flags are compared with actual outcomes to tune the rules.

03systems it connects to

• CRM. Opportunity data and field history from Salesforce or HubSpot.
• Email and calendar. Activity data, read with permissions limited to customer-facing threads.
• Call summaries. From your meeting recorder or conversation intelligence tool.
• Reporting. A dashboard or document in the tool your leaders already use for the forecast call.

04human checkpoints

• Rule calibration. RevOps and the head of sales agree on thresholds from historical data before the first run.
• Manager conversation. Every flag is discussed with the rep. The review informs judgment and never changes a forecast category on its own.
• Quarterly backtest. Leaders review which flags were right and which were noise, and adjust.

05what to measure

• Forecast accuracy. Committed number versus actual bookings, compared with prior quarters.
• Flag precision. Share of high-risk deals that did slip or close lost.
• Missed slips. Deals that slipped without ever being flagged, reviewed one by one.
• Time to action. Days from first flag to a documented next step on the deal.

06risks and guardrails

• Quiet is not always bad. A deal in legal review can look silent. Let reps annotate known pauses so the model stops flagging them.
• Gaming the inputs. If reps learn that logged activity lowers the score, they will log activity. Weight buyer responses and stakeholder breadth over raw counts.
• Surveillance concerns. Reading rep email feels invasive if it is not explained. Be clear about what is read, why, and who sees it, and keep access limited to customer-facing activity.

07build vs buy

Revenue platforms such as Clari, Gong forecasting, and Salesforce pipeline inspection already do much of this for teams on a standard sales motion. If that describes you, evaluate them first.

Custom review makes sense when your cycle is unusual, such as long public sector deals or channel sales, when the useful signals sit in systems those platforms do not read, or when you want the rules and their history fully under your own control.

08related playbooks

Browse every sales playbook or the full library.