Quick answer: good pipeline management means few stages, objectively defined exit criteria, an owner and a dated next action on every deal, and stage movement driven by evidence from conversations rather than optimism. AI can enforce most of this by updating stages from call outcomes.
Design the stages around buyer behaviour
If a stage cannot be verified by something the buyer did, it is a feeling. 'Interested' is a feeling. 'Demo scheduled with decision-maker' is an event. Use five to seven stages, each with a single exit criterion.
- New — captured, not yet contacted.
- Contacted — two-way conversation happened.
- Qualified — need, budget window and timeline confirmed on a call.
- Demo/site visit — scheduled and attended.
- Proposal — commercials shared.
- Negotiation — terms under discussion.
- Won / Lost — with a mandatory reason code.
Keep the pipeline clean without nagging
Pipeline rot is caused by manual updating. When post-call AI writes the summary, extracts the commitment and proposes the stage change, hygiene becomes the default instead of a weekly clean-up ritual.
- 1Every open deal must have a dated next action; no date means it is stalled.
- 2Auto-flag deals with no activity for longer than the stage's normal cycle time.
- 3Force a loss reason so lost-deal analysis is possible.
- 4Review stage conversion rates monthly and fix the worst-performing transition first.
Forecast from evidence
Weight deals by historical stage conversion, not rep confidence. Then sanity-check the top of the forecast against activity data: if the deals expected to close this month have no calls in the last ten days, the forecast is fiction.
Frequently asked questions
How many stages should a sales pipeline have?
Five to seven. Fewer than five hides where deals die; more than seven creates admin work and inconsistent usage across reps.
What is pipeline velocity?
Number of open deals multiplied by average deal value and win rate, divided by average sales cycle length. It shows how fast revenue moves through the funnel.
How does AI improve pipeline accuracy?
By deriving stage changes, next steps and risk signals directly from call transcripts, so the board reflects what actually happened on conversations rather than what reps remembered to log.