How to Coach a Sales Team From Its Own Call Recordings

An advisor at a retirement planning firm finishes a first appointment at four in the afternoon. When they open their laptop the next morning there is a short page waiting for them. How the conversation went against the firm's own sales method. Which questions they asked and which they skipped. One thing to do differently in the next meeting.

Nobody typed up the call. Nobody sat and listened to the recording.

That page is an agent reading the transcript of their own meeting. It is one of ten agents running at that firm, and of all of them it is the one I get asked about most often, because almost every company with a sales team is now recording its calls and almost none of them have anything reading them.

The build is not hard. What decides whether a rep reads that page or closes it has very little to do with the model.

Score against their standard, not a generic one

That firm had a sales method before we arrived, and a coach who teaches it. The method is what the agent grades against: the questions it says to ask in a first meeting, how an objection is meant to be handled, whether a next step was agreed before the call ended.

A generic call quality score is worth nothing to a rep, because it does not match what their manager asks them about on Monday. It also invites an argument about the rubric instead of a conversation about the call, and that argument is unwinnable, because the rep is right.

So the rubric is the work, and writing it is not an AI task. Sit with whoever already debriefs calls and get one page out of them: what a good first meeting contains, in order, in their words. Four or five items, each one observable in a transcript. If an item cannot be checked by reading, "built rapport" being the usual offender, either define it as something a person actually says or take it out of the rubric.

Then make every line on the page cite the call. Each item points at the part of the conversation it came from: the timestamp and the sentence.

The first reason is that a rep will argue with a score and cannot argue with their own sentence. Feedback that quotes the transcript arrives as a fact about the meeting, and the conversation moves on to what to do next time, which is the only part worth anyone's hour.

The second reason is that transcripts are less reliable than they look. We read back 687 of our own and found the speaker labels unreliable in roughly a third of them, with whole exchanges collapsing under one name. For a summary of what was decided, that barely matters. For credited feedback about a person it matters completely, because the worst version of this build tells a rep they talked too much, quoting the prospect. Citations make that checkable in seconds instead of arguable for a week.

Decide who sees it, in what order, before it ships

This is the choice that decides whether the build survives its first quarter, and it is not a technical one.

Start here: the first audience is the person who was on the call. Their own page, the way a rep reads their own numbers. The coach sees it as part of the debrief they were already going to have. A view across many of a person's meetings is a different product, built separately and introduced openly. At the retirement planning firm that is a second agent. At a national wealth management firm with several hundred advisors, the coaching agent sits in the advisor portal we built inside the firm's own cloud, with a per-advisor view that looks across meeting history, and the pilot on it started in September.

Get the order backwards, so the scores land on a manager's desk first, and you will not get better meetings. You will get people managing the recording.

The same decision has a second half. The agent drafts and a person decides. A coaching page is not a performance rating, and it cannot quietly become one in month four. Name that gate in the scope, where every review gate belongs, rather than negotiating it after the first awkward conversation.

Pick a number you can read before January

The temptation with this build is to claim it moved the closing ratio. I have been offered that number for the retirement planning firm and I am not publishing it. It was a guess offered in passing, the firm computes the ratio once a year, and human sales coaching was running the whole time alongside the software.

That is a practical problem as much as an honesty one. A measure that reports once a year cannot tell you in March whether the thing you launched in February is working, and it cannot tell you which half of the work moved it.

So pick a leading measure the system can move and a person can read monthly. The share of meetings that got a scorecard at all. The share of calls where a next step was set, which the agent can count because it reads every call rather than the three a manager had time for. How many reps opened their page this week. What proportion of pages the coach edits before sending, which is the fastest signal you have on whether the rubric is right. Record the before on whichever one you pick, because that number expires the day you launch.

Keep the annual figure. Just do not sit still waiting for it.

Start with ten calls

Take ten recent recordings, the page your coach already teaches from, and one afternoon. Score the ten by hand, with the coach in the room, each score quoting the line it came from.

You will learn three things before anyone writes code. Whether your rubric separates a good call from a bad one. Whether your recordings are reliable enough to quote back to the person who was on them. And whether the coach would send that page to a rep exactly as it reads.

If the answer to the third is no, what you need is another week on the rubric, not a month of engineering. It is the cheapest week in the project and it is the one everybody skips.