What you get
A summary is a paraphrase, and a paraphrase cannot be acted on.
No rep has ever changed their forecast because a tool said the prospect raised pricing concerns. They are right not to — they were there, the tool was not, and a paraphrase gives them nothing to check it against.
What changes a decision is the sentence.
Nine outputs
Each one a quote, each one pressable.
An objection is stored with the rep's answer to it as a second quote, separately timestamped, and a verdict on whether it landed — resolved, partly resolved, left alone, or deflected. That second quote is the whole of coaching; knowing an objection was raised is worth very little.
Commitments carry which side made them and how firmly. Competitor mentions are classified by how they came up, so somebody naming a tool they use for something unrelated is not counted as competitive pressure. Pricing signals are typed — a budget stated, a budget asked for, a discount requested, sticker shock, pushback on value.
- They said14:22
“We've been burned on migrations before and I don't have another six months.”
The rep answered14:51“Totally fair. Let me show you what a staged cutover looks like — you'd be live on the first workspace in three weeks.”
Resolved - They said31:07
“Your pricing is well above what we budgeted for this.”
The rep answered31:19“I hear you. Where did the budget come from?”
Deflected
- Sentiment is an arc, not a number —
Minute by minute, so "it went fine" and "it went fine until the pricing came up" are different answers.
- Topics carry how long each was dwelt on —
Which is usually more revealing than which topics came up at all.
- Money is normalised in the instruction itself —
A spoken amount in local idiom becomes a figure with a currency, preserving whether it was monthly or yearly.
- A value the system does not recognise is dropped, not coerced —
Rather than being forced into the nearest category — so the categories stay worth reading.
The measurements
Three of them involve no model at all.
Share of the talking, the longest unbroken stretch and whose it was, and how often the turn changed — computed arithmetically from the timings. Identical whether the words came from audio or from a transcript file, and not subject to anybody's opinion.
Somebody who is neither the rep nor the prospect counts toward the total but neither share, so a call with a third party honestly shows shares that do not add to one rather than quietly inflating somebody's figure.
- Turn-taking is capped so calls compare —
A short call and a long one can be read against each other.
- Trackers watch for your own phrases —
Words and phrases you define, run against every new transcript, with the surrounding context kept so a reviewer sees the mention without opening the call. Single words match as words; a phrase carries its own context and matches loosely.
›Do action items become tasks?
Not today. They are listed on the call with their party, an owner hint and a due hint — and creating a task from one is a manual step. Worth knowing if you were planning to run follow-through from them.
›Are the counts of questions asked available?
No. The field exists and is never filled, so treat question counts as absent rather than as zero.
›What happens if the transcript is empty?
The call finishes with a note saying there was nothing to extract from, rather than sitting in processing forever.
›What about a very long call?
There is a ceiling. A call long enough to exceed it fails permanently rather than being processed in pieces, and re-running fails the same way.