Scoring sales calls with AI, against your own scorecard

A sales-coaching expert had a method that worked: a scorecard of 12 criteria in 4 sections, proven by scoring calls by hand. The question was whether software could apply the same method to hundreds of calls, and whether a manager would believe the result.

The client’s method, not ours

The product does not invent a definition of a good call. It scores against the client’s own scorecard, criterion by criterion. Two methods run side by side: a fast rules-based one, and AI for the judgment the rules cannot make.

Every score shows its evidence

A score on its own invites an argument. A score with the sentence from the call that earned it invites a coaching conversation. Every criterion comes with quoted evidence: 1,187 quotes in the first round alone.

Clean data before clever scoring

Two problems had nothing to do with AI quality, and both would have misled a manager:

• Calls credited to the wrong person. An AI classifier moved 60 calls to the salesperson who actually made them.
• Training calls counted as sales calls. Coaching and training recordings were dragging the team’s average down. Excluding them moved it from 47% to 72.2%. That is a data fix, not a change in performance, and we said so.

Fast enough to use

In our own first cloud version, scoring a call took 120–180 seconds. We found the cause and fixed it within the week: 1.6 seconds per call, about 75 times faster.