How I use Nonverbia (+ Claude) to build a "clone" of myself that writes in my voice.
Before Nonverbia I tried a couple of other call transcription/analysis tools, but none of them really convinced me. This is the one I feel most comfortable with, and not for just one reason. It's a combination of things that, put together, made me build a real system on top of it, not just use it as an archive of transcripts.
Let's start with the features I use every day exactly as they are, with no automation of mine on top. The skill that prepares the follow-up email to the client right after the call is the one that saves me the most work. Normally that kind of email I only write the day after, with a clear head, but with Nonverbia I have it ready while I still have all the details of the conversation fresh in mind. And I really like the automatic association of the call to the company I was talking to. It lets me filter the history by client very comfortably, something that with other tools I had to do by hand, renaming files or keeping a separate sheet. This feature is more useful than it sounds, because I don't only use it to search an old call: I feed the filtered calls to my AI agents that work for specific companies, so every agent gets the real context of the conversations had with that company, not a generic summary.
Also the evaluation Nonverbia gives, both to me and to the client, on the single call, I find it realistic, not the usual inflated report card. It made me notice things about my own behavior that I would not have caught at the moment (dead time, moments where I talk too much instead of letting the client speak), and it's one of the few pieces of feedback I actually use to correct myself, not just read and archive.
On the technical side, the MCP integration was simple. I connected it as a connector and it worked without the usual complications I have to deal with on other services. And the interface is nicely done and functional, you can tell it's built to be used every day, not just to exist.
The part I built on top myself is the "clone". Every night an instance of Claude, running headless, ingests the new calls and extracts metrics text based, directly from the transcript. For this specific use I ignore the automatic Nonverbia score on purpose, because here I need the verbatim text of what I actually said, not a score (a different use case from the coaching one above, where the score is exactly what I read and find useful). In parallel I export my sent emails from Thunderbird. The two sources do different jobs: the content, the ideas, the examples, the numbers, always comes from the call, my own words, verbatim.
The voice, the rhythm, the little quirks, how I build sentences, comes from the emails. Claude then writes posts or drafts combining the two, and the accuracy of Nonverbia's transcription is what makes the whole rest reliable. I learned the hard way that if I let the model "invent" content just from style rules, the result sounds fake and sometimes puts theses in my mouth that I never said. Anchoring everything to the transcript keeps it faithful to what I actually said in the call, and that faithfulness, more than the transcription itself, is the reason building a system like this on top of it makes sense. Review collected by and hosted on G2.com.
Nothing, I haven't found any flaws so far. Review collected by and hosted on G2.com.