The Summaries tab tells you what a meeting was about. The Intelligence tab tells you what kind of meeting it was.
Everything here is computed directly from the transcript. No model is involved, which has two consequences worth knowing: it is free, and it works on every meeting you have ever recorded, including ones from long before any of this was configured.
Who spoke, and how much
Per speaker: how many turns, how many words, how many questions, the average length of a turn, the longest single uninterrupted stretch, and the share of the total conversation.
The share is the number everybody looks at, and the honest reading is entirely situational. Seventy per cent in a discovery call is a problem. Seventy per cent while you are presenting is the job you were asked to do.
The speaker detection is deliberately conservative. It keeps only name-shaped labels that recur through the conversation, because a pasted document full of "Note:" and "Summary:" lines will otherwise register as a meeting with forty participants. That means an occasional real speaker who said one thing is missed, which is the right trade.
How much of a conversation it was
Interactivity measures how often the speaker changed relative to how many turns there were.
High means genuine back-and-forth. Low means people took the floor and held it. Neither is good or bad on its own, and the useful move is comparing meetings of the same type against each other. A discovery call less interactive than your own average is one where somebody did not open up, and that is worth knowing before the next one.
The arc
The meeting is divided into five slices, showing who was speaking and how many questions were being asked at each stage.
Three shapes are worth recognising.
Questions front-loaded, then a long tail of one voice. Discovery that turned into a pitch, usually about eleven minutes in.
Balanced throughout. Generally a real working conversation.
One voice for four slices, then everybody at the end. A presentation with questions afterwards. Fine if that was the intent, and a failure if it was meant to be a discussion.
The fingerprint
Three densities, measured per thousand words: how many questions, how many commitments, how much hard data.
Together they produce a one-word character for the meeting. A conversation dense in commitments is a planning meeting whatever it was called on the invitation. One dense in numbers and light in questions is a report being delivered rather than a discussion being had.
This is most useful in aggregate. A team whose customer calls are consistently light on questions has a coaching problem, and it shows up here well before it shows up in the pipeline.
Question dynamics
Per person: how many questions they asked, and how many they fielded.
The ratio is a strong signal about who was actually driving. In most sales conversations you want your own side asking more than answering, and the recurring surprise is how often it is the other way round.
Key terms
The words that recurred, grouped into four categories: financial, risk, action, and topic.
Scan this before reading anything else on the page. Which categories dominate tells you what the meeting was really about, and it is not always what it was called. A kickoff dominated by risk terms is a different event from a kickoff dominated by action terms.
Conversation dynamics
Three per-speaker measures, and these are the ones with something to say about behaviour rather than content.
Cut off. The number of times your turn ended mid-thought and somebody else started talking.
Interrupted. The number of times you started while somebody else was mid-thought.
Acknowledgement rate. The share of your turns that open by recognising what was just said before responding. This is the closest available measure of listening, and it is the one most worth improving because it is the one most directly under your control.
Alongside these sits relitigated topics: subjects that kept coming back. A topic revisited four times in an hour was not resolved, whatever the summary says about it, and it is almost always the right thing to put at the top of the next agenda.
Three shapes of meeting
The analysis adapts to what it was given rather than drawing charts regardless.
Dialogue. Two or more identified speakers. Everything above applies.
Single stream. The recording did not separate speakers. Word and question counts still hold; anything per-speaker does not, and the tab says so rather than attributing everything to one person.
Too short. A no-show, a technical failure, or a paste with almost nothing in it. It tells you rather than producing a chart of nothing.
Using it across meetings rather than within one
Almost everything here is more useful compared than read.
Against your own average. Your interactivity on this discovery call versus your last ten. Your question ratio on the calls that converted versus the ones that did not.
Against the meeting type. A retrospective should look different from a demo. Where they look identical, one of them is being run as the other.
Against the other side. Who asked more questions, you or them, is the single most portable signal in this tab, and it applies to interviews, reviews and negotiations as much as to sales.
The tab shows one meeting at a time on purpose. The trends belong in Coach, and the aggregate belongs in Reports.
Five things it reveals that people miss
The pitch that started too early, visible in the arc as questions stopping around the second slice.
The meeting nobody needed, visible as high interactivity, no commitments and no data.
The person who was there and not there, visible as a real attendee with almost no turns.
The unresolved thing, visible in relitigated topics, and almost always the right first item next time.
The monologue you did not notice giving, visible as a longest-stretch figure considerably higher than you would guess.
The honest limits
These are heuristics computed over text. Captions drop punctuation, and an interruption is inferred from a turn that trails off rather than from measured audio overlap.
Read the relative patterns, not the absolute numbers. Your interruption count against your own average across twenty meetings means something real. Your interruption count in one meeting means very little, and treating it as precise is how people end up arguing with a chart.
Did this answer your question?
No, ask a person