See How the Call Actually Went
Paste a transcript with speaker labels and you get the numbers: your talk to listen ratio, how long your longest uninterrupted stretch ran, how many times the call changed hands, and how many questions each side asked.
Pick which speaker is you and the coaching notes follow from there. The transcript is parsed and counted on your device, which matters when the call is a real customer conversation.
Talk Ratio Is the Number Nobody Checks
Every rep has an opinion about how much they talk on a call. Almost nobody has a measurement, because getting one used to mean someone sitting in and keeping score.
It is a word count and a division, which is why it is worth doing. Your words over the total words, per call, and suddenly the thing everyone has a feeling about has a number attached. The number is not a verdict on its own, but it is a fact, and it is checkable by hand if you disagree with it.
What makes it useful is the comparison over time rather than any single reading. One call at seventy percent might be a demo where talking was the job. Five discovery calls in a row at seventy percent is a pattern.
The Longest Stretch Matters More Than the Ratio
If you only look at one number here, look at this one.
A call can sit at a respectable fifty percent and still have gone badly, because the fifty percent arrived as one six hundred word monologue in the middle while the other side waited. Averages hide that completely. The longest uninterrupted stretch does not.
This counts consecutive turns rather than single ones, which is how a monologue actually appears in a transcript: one long answer broken across several paragraphs by whoever made the transcript. Treating those as separate turns would make a four minute speech look like four short ones.
Past roughly two hundred words a listener stops following and starts waiting for a gap. That is the first place to cut, and it is far more actionable than nudging a ratio.
Questions, and Why the Count Is Low
Question count tends to correlate with the thing you actually wanted from a discovery call, which is finding out what the other side needs.
Both explicit questions and the ones that lost their question mark are counted, because machine transcripts drop punctuation constantly. “How many people does that affect” is a question whether or not a transcript bothered to mark it.
The number is usually lower than people expect, and it is usually lowest on exactly the calls where the talk ratio was highest. Those two readings tend to move together, and seeing them side by side is more persuasive than either alone.
What This Deliberately Does Not Do
It does not find objections, commitments or next steps.
Those require reading for meaning, not counting. A pattern matcher looking for objections would flag “that sounds expensive” and miss “I need to think about the timing”, which is the one that actually killed the deal. Producing a confident list of objections that misses the real one is worse than producing no list, because people trust the list.
So this half is the measurable half, and it is honest about being that. For the reading-for-meaning half, the meeting analyzer works from a recording and produces a transcript, action items and chapters, and you can ask it about objections directly.
It also counts words rather than seconds unless your transcript carries timings. A fast talker and a slow one with the same word count get the same ratio, even though the slower one held the floor longer. If seconds matter to you, use a transcript with timestamps.
It Needs Speaker Labels
This is the one requirement, and it is worth saying plainly: a transcript with no speakers cannot be split into a ratio.
What works is any line that names a speaker and then a colon. Leading timestamps are fine and are stripped, a time in brackets after the name is fine, VTT voice spans are read, and a line that continues the previous speaker’s turn attaches to them rather than becoming its own turn.
What does not work is a wall of text, however good it is. Most meeting tools label speakers by default, and if yours labels them “Speaker 1” and “Speaker 2” that is enough, since you pick which one is you.
Getting a Transcript to Analyze
If you have the recording but no transcript, the transcript generator produces one with speaker labels, which is the format this needs. For a call recorded on a phone, the phone call summarizer handles the recording and the transcript together.
If the transcript you have is cluttered with timestamps and filler, the transcript cleaner strips that out, and the transcript formatter can put speaker labels into a consistent shape first.
