AI pipeline forecasting tools predict deal outcomes by analysing post-opportunity data — transcripts, stage velocity, deal age. But if your pipeline leaks at the handoff before opportunities are created, the forecast is built on an already-filtered dataset and misses the real loss.
The promise and the blind spot
Gong’s Revenue AI OS hit $500M ARR in May 2026, up 55% year over year, built on data from 3,000 revenue leaders. Revenue intelligence is a serious category now. The pitch is compelling: AI listens to every call, reads every email, and tells you which deals will close and which will stall.
But here’s the structural problem. These tools analyse data from the opportunity stage onward — transcripts of AE calls, stage progression, deal age, stakeholder engagement. That’s all post-opportunity. If your real pipeline loss happens at the handoff, before an opportunity is ever created, the forecast is built on a dataset that’s already been filtered. The deals that leaked out never generated a single transcript for the AI to learn from.
Where the quality leak hides
The data backs this up. AI-sourced meetings convert to opportunities at 28% in AI-only pods versus 47% with human SDRs. AE win rates on AI-sourced opportunities run 9 to 12 percentage points lower. The drop-off isn’t happening in late-stage deal management where the forecasting AI is watching. It’s happening on the first AE call — the moment a meeting booked by an AI SDR gets qualified or doesn’t.
That first call is the handoff. It’s where meeting quality meets sales judgement. And it’s almost never measured systematically. The conversation intelligence platform is recording it, sure — but the forecasting model is trained on later-stage patterns, not on the qualifying call that determines whether a meeting becomes an opportunity at all. Your pipeline coverage ratio might look healthy, but it’s counting opportunities that already survived a filter you can’t see.
How this tends to play out
A SaaS company ran Gong across their entire sales org. Forecast accuracy on late-stage deals was strong for opportunities past discovery. But when you trace pipeline back to source, AI-sourced meetings were converting to opportunities at a far lower rate, and of those, AE win rates were well below human-sourced deals. The forecasting AI reported a healthy pipeline. It couldn’t see the 76% of AI-sourced meetings that leaked out before opportunity creation — because those calls never produced the opportunity-stage data the model consumes.
The fix
Point your conversation intelligence at the first AE call on every AI-sourced meeting, not just later-stage deals. Tag those calls. Measure qualification rate, disqualification reasons, and the gap between AI-sourced and human-sourced meeting quality. That’s where the leak is visible. Your forecasting AI is only as good as the dataset it’s trained on — and right now, it’s trained on the deals that survived the handoff, not the ones that didn’t.