In July I started grading my own sales calls with AI. About a dozen calls later, the scores ran from 4.5 to 16 out of 16. The low ones all broke in the same place, and it was rarely the pitch.
How I set it up
Fathom records and transcribes the call. When its speaker labels get mixed up, I use the Gemini transcript instead. Then Claude reads the transcript and scores it.
The rubric came first. I took four calls from my business partner and had Claude pull out what he does on every one. That gave eight criteria, each scored 0, 1 or 2, for a total out of 16. His best call scored 16 out of 16, and it became the benchmark every other call is measured against.
Asking before recording
Everyone on these calls knew they were being recorded. I learned how to ask the hard way. On an early call I brought up recording at the start, live, and the first six minutes went to an awkward back and forth about it. Now consent goes in an email before the call, with a short consent form. By the time we talk, it’s settled.
What the scores showed
The pitch held up on almost every call. Discovery and the close did not, especially with tough buyers. Three habits kept costing me points.
- Price before value. I gave the number before the buyer had said what a result was worth to them.
- Discounting on the first call. Nobody had asked for a lower price. I offered one anyway.
- One offer, several prices. The same entry offer was quoted at different prices on different calls.
What failed
Claude scored one call with a real-estate broker too high. On paper the call looked fine. Polite, on topic, a clear offer.
I was on that call. The broker said no to running ads, and I pushed the same objection four more times after that. I told Claude what happened, and it rescored the call at 6 out of 16.
The AI reads words. It doesn’t feel the room unless you tell it what the room felt like. So I read every score against my own memory of the call, and the prompt now asks Claude to quote the transcript line behind each score so I can check it.
What changed in how I sell
The scores pushed a bigger change than I expected. I used to sell studio time by the hour. Now I sell done-for-you production at a higher price, and that only works if I stop doing what the scores flagged.
So I stopped underestimating my time. I stopped discounting before anyone asks. I stopped being shy about the pitch. And I ask a question I used to skip.
You run a business. What are your numbers?
If I don’t know what a client is worth to them, I can’t price anything.
The prompt
Copy it, paste a transcript under it, and add your own benchmark call.
You are scoring a recorded sales call. Use the transcript below.
Score each criterion from 0 to 2.
0 means missing. 1 means partly done. 2 means done well.
1. Frame. Did the seller set the agenda and the time at the start?
2. Discovery. Did the seller learn the real goal before pitching?
3. Metrics. Did the seller ask what a new client or a result is worth?
4. Decision. Who else decides, and who else is the buyer talking to?
5. Objections. Was each objection answered once, then left alone?
6. Path. Did the seller recommend one option that fits what the buyer said?
7. Close. Did the call end with a dated next step?
8. Hygiene. Talk time, screen sharing, no discount offered unprompted.
For each criterion, give the score and quote the line from the transcript that justifies it.
Then give the total out of 16, the one thing that cost the most points, and one question the seller should have asked.
Benchmark. This call scored 16 out of 16. Use it as the reference.
[paste the benchmark transcript or summary]What I do now
- Consent first. A pre-call email and a consent form, never a live ask.
- A benchmark call. Every score is measured against one real call that earned 16.
- My read on top. I check every score against what I felt on the call.
- Quotes behind every score. If Claude can’t point to the line, the score doesn’t count.
