21 August 2026
A sales rep leaving a call needs four things fast: the next step, the objection that surfaced, the number that moved, and a place in the CRM where all three land without retyping. A three-paragraph summary is the wrong shape for that job — a CRM field wants a value, not a narrative, and someone still has to decide what that value is.
Ask a rep what they need twenty minutes after a discovery call ends, before the next meeting starts, and the list is short. They need to know what happens next and by when. They need the objection that came up, worded the way the prospect actually said it, not softened into 'some hesitation around pricing.' They need the number — the quote, the seat count, the renewal date — exactly as stated, because a rep who reports 'around ninety' when the prospect said 'ninety-two, maybe ninety' has already lost a negotiation before it started. And they need all of it sitting on the deal record, not in a separate notes app they'll forget to open before the next call.
A CRM field is a slot with a type: a date, a currency amount, a picklist value, a short string. A meeting summary is three paragraphs of connected prose written to be read once and then archived. Pasting one into the other doesn't create structure — it creates a wall of text sitting in a field designed to hold '$42,000, closes March 14th,' and the next rep who opens the record still has to read the whole thing to find that number.
This is the actual failure mode behind 'our reps don't update the CRM.' It's not laziness. Rewriting a summary into six separate fields takes longer than the call itself felt worth, so the field stays blank and the deal history lives only in someone's memory until they leave the company.
Good extraction doesn't produce a summary at all — it produces a short list of typed values, each one tied to the moment in the call it came from. A next-step field with a date. An objection field tagged to a category — price, timing, competitor, authority — with the actual sentence attached. A number field holding the figure someone said, plus who said it. AVAY builds these from the same transcript it keeps for every call, so a field isn't a guess made after the fact; it's a pointer back to a timestamp a rep can click and hear for themselves.
The difference matters most on the number. 'Deal size: roughly $90k' is a different fact from 'prospect said ninety, rep quoted ninety-two, gap unresolved.' The second version is what a sales manager actually needs to see in a pipeline review, and it only exists if the extraction keeps both numbers and the disagreement between them, rather than collapsing everything into one tidy figure.
No transcript-reading model should write straight into a live deal record without a stop first, because language about commitment is genuinely ambiguous. 'Finance is looking at it' and 'finance signed off, we're just waiting on legal' are both true-sounding sentences that mean very different things to a forecast. A rep who was on the call knows which one it was; a field written automatically from the words alone has to guess.
So the honest shape of the workflow puts a review step between the transcript and the field: candidate values surface, a rep confirms or edits each one, and only the confirmed version gets written. That's slower than a fully automatic write, on purpose — it's the difference between a CRM that's occasionally wrong and one nobody trusts.
AVAY doesn't ship a fixed list of CRMs it talks to. It reaches a team's own systems through connectors the team attaches themselves, which means what gets written where depends on what's been connected, not on a hardcoded integration built ahead of time. That's a narrower promise than 'works with your CRM' out of the box, and it's the honest one — check AVAY's current connector list for what's actually wired up before assuming a specific system is covered.
The useful split is pull versus push. Pulling — searching everything said across every past call for a mention of a specific account — works the moment the meetings exist, no setup required. Pushing a confirmed field into an external record needs the connector in place first. A team evaluating this for sales calls should ask which direction they need before assuming the other one comes free.
Extraction fails in predictable ways. Two people talking over each other on the number — one says eighty-five, the other corrects to ninety, and the transcript captures both without knowing which one stuck. Soft objections that never get stated as objections, just a pause and a change of subject. Jargon specific to the deal — a product SKU, an internal approval tier — that the model has never seen before and transcribes as something close but wrong.
None of that is a reason to skip extraction; it's a reason to keep the review step rather than drop it. A rep scanning five flagged fields for thirty seconds catches almost all of it. A rep who never looks at the transcript again catches none of it.
| What ends up in the CRM field | What the rep has to do before it's usable | |
|---|---|---|
| Manual notes | Whatever the rep remembers to type, if they type anything | Nothing — but the field is often just empty |
| Auto-generated summary | A paragraph pasted into a notes box | Rewrite it into fields by hand anyway |
| Structured extraction | Named fields — objection, number, next step — tied to a timestamp | Confirm or correct each one, not rewrite it |
No — not for anything that affects forecast or deal stage. Language about commitment is ambiguous enough ('finance is looking at it' versus 'finance signed off') that a rep who was on the call needs to confirm the value before it's written, even when the extraction is fast and mostly right.
A summary is prose meant to be read once — three paragraphs describing what happened. Extraction is a short list of typed values, each tied to a timestamp in the call: a next-step date, an objection category with the exact sentence attached, a quoted number with who said it. Only the second one drops into a CRM field without a rewrite.
It depends on which connector your team has attached — AVAY doesn't hardcode a fixed list of CRMs it writes to, it reaches your systems through connectors you set up yourself. Searching across past calls for account mentions works immediately with no setup; writing a confirmed field into an external record needs the connector in place first.
Less reliable than on calls where someone says 'the price is a problem' in those words. A pause, a change of subject, or a soft 'let me check with the team' often signals an objection without naming one, and extraction built on transcript language alone will miss some of these. That's exactly the gap a thirty-second human scan of the flagged fields is meant to catch.
Only if the extraction keeps both numbers rather than picking one. A good field for a contested price records the prospect's figure, the rep's counter, and the fact that the two didn't match — that's more useful to a sales manager than a single tidied-up number that erases the disagreement.
A CRM field wants a value tied to a moment in the call, not a paragraph about the call — and wherever someone has to guess at intent, a person needs to confirm it before it's written.
AVAY is a video meeting platform that transcribes the call itself — no bot joins, because there is nothing to join. Start one at avay.ai, read how each part works in the documentation, or see what it costs.