I recorded a discovery call in March and listened to it three times. The deal had gone quiet after the proposal, and my memory was confident about why: price. The recording disagreed. At minute eleven the prospect asked what happens in the first thirty days, and I answered with a list of features and no dates. She said "okay" and changed the subject. We never came back to it.
That four-second pause was the whole deal. I have been teaching since 1998 and I still needed the tape to see it.
The stall is almost never the objection you remember
Memory is a bad witness. What you remember from a call is the last five minutes: the friendly goodbye, the vague "let me talk to my partner," the number you said out loud. The buyer made the decision much earlier, and usually without telling you.
Last quarter I pulled twenty-two recorded calls from my own pipeline. Fourteen had stalled at the proposal stage. When I read the transcripts instead of remembering them, eleven of those fourteen had the same shape: the buyer described a specific operational mess in the first ten minutes, I asked two or three follow-up questions, then I moved to solution. I never asked what happens if the mess stays. I never asked who else feels it. The buyer spent the rest of the call being polite.
That is not a closing problem. It is a twenty-minute-old diagnosis problem, and it shows up in the transcript as a stretch of conversation where the buyer's answers get shorter and mine get longer.
What the ten-minute autopsy actually pulls
An autopsy is not a vibe check. You are pulling four specific things out of every call and putting them in writing.
One: the buyer's own words for the problem. Not your paraphrase. The actual sentence, copied.
Two: the first question the buyer asked that you did not fully answer. There is almost always one, and it usually arrives right before the energy drops.
Three: the next step, described twice. Once as the buyer said it, once as you wrote it in your follow-up email. If those two sentences do not match, the deal is not stalled. It never started.
Four: the number of topics opened and never closed. Count them. Ten open loops on a call reads as enthusiasm to you and as chaos to them.
You do not need an ops manager or a sales coach to do this. You need a recorder and something that turns audio into text. Fathom and Granola both record and summarize without asking you to configure anything, Fireflies works if you live in Zoom, and Otter is the cheap fallback. Then paste the transcript into Claude or ChatGPT with a saved prompt asking for those four items and nothing else. The pass takes about ten minutes per call once the prompt is written. That prompt is the entire AI sales call review, and almost nobody builds it.
Build a sales call scorecard once and stop improvising
A review that ends with "that felt okay" is worthless. Feelings are what got you here.
So write six questions and score every call from one to five on each line. Mine looks roughly like this:
- Did I restate their problem in their own words before mentioning anything I do?
- Did I ask what happens if this does not get fixed?
- Did I name a number, a date, or a scope before they asked for one?
- Did I leave with a specific next step, a specific person, and a specific date?
- Did I talk less than sixty percent of the time?
- Would a stranger reading this transcript know what they were buying?
Put it in a Google Sheet, one row per call, with the date and the outcome. After twenty rows you have a founder-led sales audit you can actually read. You will see the same two or three lines dragging the average down, and those are the two or three behaviors worth fixing this quarter.
This is what people mean by AI sales coaching for founders, and it is worth being honest about what it is. It is not coaching. It is evidence. The software does not flinch when it tells you that you spent nine minutes talking about your process.
You are the worst person to grade your own call
Two reasons. The first is that you were there. You know what you meant, so you will hear what you meant when you remember the call. The transcript does not have access to your intentions. It only has your sentences.
The second is that the call was emotionally expensive. You were nervous, you were charming, you recovered from a weird moment at minute six. Nobody wants to walk back into that.
The recording does not care, which is the point. Discovery call analysis AI is useful here precisely because it has no stake in how the call felt. It will hand you the flat sentence where the buyer stopped asking questions, and it will not soften it. That is the moment the deal died, and it is almost never the moment you remembered.
The free session at https://workshop.mastermindshq.business walks through this on a real transcript, including the unflattering parts.
One call is a story. Twenty calls are a system
The single call tells you what happened. The pattern tells you what to change. Those are different jobs.
After I reviewed twenty-two calls, the finding was not "Joe needs to close better." The finding was that I was attracting people whose problem I had not priced, so the proposal stage was doing work the discovery call should have done. I rewrote the first three minutes of my calls, added two qualification questions to my intake form, and moved the number and the timeline to page one of the proposal instead of page three. Proposals stopped needing two follow-ups to get an answer.
That is the compounding part. Fixing one behavior a month, based on twenty transcripts, moves more revenue than any script you could buy. It also takes less time than you think, because most of the work is reading, not deciding.
More on how this fits into the rest of the business lives at https://www.mastermindshq.business/mastermindvsl.
The ten-minute Friday habit
Here is the version I actually run. Friday afternoon I pull the week's discovery calls, usually three to five. I run each transcript through the saved prompt, score the six lines, and write one sentence in a running doc: what I would do differently in the first ten minutes.
That is it. No dashboards, no new hire, no twelve-week program. One person, ten minutes a call, one sentence of output.
Do it for a month and you will know exactly which sentence in your discovery call costs you the most money. It will not be the price.
If this resonates and you want to build this kind of infrastructure in your own business, the Mastermind is where we do the work live. You can learn more at mastermindshq.business.