Once you're in — yes. Text or call anytime. Access itself is application-reviewed, not self-serve overnight.

Will my manager see what I tell Sara?

No. Sara reports to you. Nothing goes up the org chart unless you choose to share it.

Does she record my customer calls?

Never. No bot in the meeting. You reach out after — by text or call.

How is this different from Gong or Chorus?

Gong owns the recorded call — for your manager. Sara owns what happens after you hang up — for you.

What are the most common sales call transcript analysis metrics?

Talk-to-listen ratio, longest monologue, question count, and a sentiment or engagement score. All four are computable straight from a transcript with no human judgment involved, which is exactly why they're the default — and exactly why they can't tell you what actually happened on the call.

Why doesn't a good talk-ratio score guarantee a good call?

Because talk ratio measures how much you spoke, not whether what you said landed. A rep can hit a perfect 40/60 ratio while dodging the one question that mattered, or blow the ratio while closing a deal on the strength of one honest answer. The metric and the outcome are only loosely related.

What should a rep track instead of transcript scores?

Where their certainty wobbled, what they dodged instead of answering, and what surprised them — asked immediately, before the call becomes a summary. None of those show up in a scored transcript, because a transcript can only tell you what was said, not what the rep actually thought while saying it.

FIELD NOTES

Sales Call Transcript Analysis Metrics: The Numbers and What They Miss

Talk ratio, longest monologue, question count, sentiment score — four sales call transcript analysis metrics, and the one read none of them give you.

2026-08-21 · SARA — KEEL'S AI DEAL ASSISTANT · GETKEEL.IO

A printed call-transcript page on a dark desk, one line singled out by hard light with a small ember-orange tab clipped beside it.

The transcript comes back scored within a minute now. Talk ratio: 44%. Longest monologue: 90 seconds. Sentiment: positive.

None of those three numbers would have told you the buyer flinched at the renewal terms and you talked right past it.

The four metrics every tool scores

Run a sales call transcript through almost any analysis tool and you'll get some version of the same four numbers.

Talk-to-listen ratio. How much you spoke versus the buyer, as a percentage. Popular because it's trivial to compute — and only loosely related to whether the call went well.
Longest monologue. The single longest unbroken stretch you talked. A proxy for "did you lecture," which sometimes it catches and sometimes it flags a rep who was finally explaining pricing clearly.
Question count. How many questions you asked. Rewards volume, not whether any question actually opened something up.
Sentiment or engagement score. A model's guess at tone, scored from word choice and pacing. It reads confidence and evasion off the same cheerful words.

Every one of these is real, computable, and defensible on a dashboard. None of them was built to answer "what should this rep do differently next time."

What a clean score can hide

A rep can post a textbook 40/60 talk ratio while spending the whole call avoiding the buyer's actual objection.

A rep can blow every ratio on the sheet — talk too much, ask too few questions, run one four-minute monologue — and still close, because the monologue was the one honest answer the buyer needed to hear.

Gong scorecards turned this gap into an entire category of rep complaints: the number moves every week, the actual read on the deal barely changes at all.

Why the numbers persist anyway

They're cheap to produce and easy to defend in a review. A manager can point at 44% and 90 seconds without listening to a single minute of the call.

That's the appeal, and it's also the limit. A metric that requires no listening can only ever measure what a transcript makes visible — word choice, pace, turn-taking. It was never going to reach the part of the call that happened in the rep's head.

Tools should sharpen a rep's own judgment, not stand in as a substitute for it. A score can flag that a call ran long. It can't tell you whether the rep left that call more certain about the deal or less.

The read a transcript can't produce

Four questions, asked by the rep, immediately, do more than any scored metric:

Where did your certainty wobble?
What did you dodge instead of answering?
What surprised you?
What would you say differently next time?

A transcript can be searched for the words that surround these moments. It can't answer the questions themselves — only the rep who was on the call can, and only for about twenty minutes before the honest version fades into the story they'll tell in the pipeline review.

AI call analysis covers the fuller case for running this immediately instead of waiting for a weekly report.

Somewhere to run the real analysis

Sara's built for the read a transcript metric can't produce — talk through what actually happened the moment a call ends, honestly, before the wobble gets smoothed into a score. Nothing routed to a manager, nothing scored against a rubric. Founders Club is invite-reviewed — apply at getkeel.io/founders.

The score was measuring the wrong thing

Talk ratio, monologue length, question count, sentiment — every one of them describes what the transcript contains.

None of them describes what the rep understood by the end of the call, or what they'd do differently on the next one. That's a different question, and it's the one actually worth analyzing.

By the team at Keel. We're building Sara, an AI deal assistant for the moments that don't get recorded.

MORE IN THIS SERIES

AI Call Analysis: What's Actually Worth Analyzing After a Sales Call

Most "AI call analysis" is a transcript and a talk-ratio score. The read that actually changes your next call is a different question entirely.