Six Years of Recorded Calls Mined for Case Studies

I've recorded my calls since March 2020: strategy sessions with prospects, coaching calls and team meetings. The recording service held 597 of them by October 2026, about 347 hours. My site had 31 case studies, and dozens of the results I'd described to prospects on those calls had never been written up. The subscription was about to lapse, and the recordings could go with it.

Of the 597 calls, 123 came back with no transcript text at all. The early recordings label speakers Speaker 1 and Speaker 2, so the transcripts don't say which voice is mine. And I'd told some of the same stories on several calls, with the numbers shifting between tellings, so nothing could go on a page until it was checked against the exact line.

The first agent listed every call by paging through the service until a page came back short, then downloaded each one in full with a script that could pick up where it stopped: 597 calls, 474 with text, 166,143 sentences. The 123 without text kept their title, date and participants. Next, the calls were sorted into sales calls, team meetings and the rest, and my side was picked out on 167 of the 225 early calls by how a seller talks compared with a buyer. Then a group of agents read the 225 calls from 2020 and 2021 that had transcripts, plus 66 later ones, in slices and pulled every story, technique and belief into 2,145 rows, each carrying the exact words it came from. A checker matched all 4,351 of those quotes against the transcripts and found every one. Retellings of the same story were merged with each version cited, which left 148 distinct stories from the 2020 and 2021 calls alone. Those were ranked against 299 job posts I'd saved, so the studies would answer what buyers ask for. Failure stories and stories that belonged to a former employer were held out. Writers built each study from checked lines only, a separate critic read every page, and one study with no result on record was held back.

• Recording service API, full transcript download
• Resumable download script
• Call sorting and speaker identification
• AI agents mining stories in parallel slices
• Word-for-word quote checker
• Ranking against 299 saved job posts

21 new case studies went live less than nine hours after the work started, taking this site from 31 studies to 52. The pages rest on 464 source lines, and the checker matched every one against its transcript. All 597 calls now sit on my own storage. A spec of how I sell, built from 330,590 words of my side of 99 sales calls and the 1,724 questions I asked on them, set the voice for all 21.

The results I'd been telling prospects out loud are now on the site, in my words, with their numbers checked. The archive, the story inventory and the voice spec stay ready for proposals, articles and the next round of studies.

Every number has to trace to a line on a call before it goes on a page. Retold stories drift: one result here came with two different daily figures on different calls, and its page uses the figure the transcripts back.