Conversation-Intelligence Deal Diagnosis (the Dealbot)
Run an AI agent over every call, email and Slack to find the real reason you win, lose, and stall
- Difficulty
- Moderate
- Time to result
- ~days to results
- Steps
- 4
- Confidence
- 90%
Dump all recorded sales interactions (call transcripts, emails, Slack) into an agent and have it diagnose deals more objectively than the humans who ran them. It surfaces the true loss reason (often different from what the rep reported), flags in-flight risks in real time, and identifies systematic weaknesses in the sales process that you then fix in weekly sprints — treating GTM gaps as bugs.
Origin
Built by Jeanne DeWitt Grosser's team at Vercel on top of Gong transcripts, starting as a post-quarter 'Lostbot' loss review and evolving into a real-time 'Dealbot'.
Core principles
- 01AI reading every interaction sees patterns humans miss and isn't biased toward the rep's self-serving explanation
- 02The reported loss reason is frequently wrong — 'lost on price' often means you never reached the economic buyer or failed to demonstrate value
- 03Treat systematic sales weaknesses as bugs in your GTM process that shouldn't exist
- 04In a high-velocity release environment, agents can diagnose enablement gaps faster than humans can be trained
How to run it
- 1
Start with a retrospective loss review
Feed the agent your top losses of the quarter sorted by deal size and have it read every call, email and Slack message to determine why each was really lost.
Pro tip Sort by deal size so you learn most from the biggest misses first.
- 2
Look past the rep's stated reason
Compare the AE's explanation to the agent's. Vercel's biggest quarterly loss was reported as 'lost on price'; the agent found they never reached the economic buyer and the buyer visibly didn't accept the ROI/TCO math — the real cause was an inability to demonstrate value.
Pro tip When you keep losing on 'price', suspect an unquantified value story and go codify your ROI/TCO math for the whole team.
- 3
Move it into real time
Run the agent live and push insights into each deal's Slack channel: 'you're this far in and haven't talked to an economic buyer' or 'that economic-buyer call didn't sound like it went well, here's how to follow up.'
- 4
Run weekly GTM bug-fix sprints
After each release, run the agent across recent calls to find where the team handles objections badly or gets stuck, then huddle weekly and fix those gaps — add objection-handling content, update the discovery guide, change the demo — like closing engineering bugs.
Pro tip Frame the gaps explicitly as 'bugs in your go-to-market process' so the team treats fixing them as non-optional.
In the wild
The AE recorded the quarter's largest loss as lost on price. Running the agent over every Slack, email and Gong call showed they never really engaged the economic buyer and the buyer didn't accept the ROI math.
→ Reclassified the true loss reason as failure to demonstrate value; prompted Grosser to build out and codify Vercel's value-quantification for the whole GTM team.
A Vercel airline customer transcribed every support call. Rather than only auto-answering calls to cut cost, a C-level exec wanted to know why people call so fewer would need to next week — an analysis AI can do across all transcripts far faster than a human tagging a CRM.
→ Shifted the goal from deflecting calls to eliminating their root causes.
Common mistakes
Trusting the rep's stated win/loss reason
Reps rationalize — 'lost on price' masked a failure to reach the economic buyer and prove value; without reading the raw interactions you fix the wrong thing.
Treating recurring objection-handling failures as normal
If you don't frame systematic sales gaps as bugs and sprint on them, they persist release after release and quietly cost you deals.
Is it for you?
Best for
Sales leaders with recorded-conversation data (Gong-style) and a high release velocity who want objective win/loss diagnosis and continuous enablement
Not ideal for
Teams without recorded interaction data or the discipline to act on the findings in a regular cadence
From the transcript
“we take all of our gong transcripts and we dump them um into an agent called the dealbot”
“actually you lost because you never really got in touch with economic buyer”
“the reason we lost was an inability to demonstrate value”
“They're bugs in your go to market process, so you should not have them”
“that dealbot, the Lostbot version was like two days”
From the episode
What world-class GTM looks like in 2026
Jeanne DeWitt Grosser (Vercel, Stripe, Google)