Summary
What the visitor wanted, what the agent did, and how it ended — written for someone who was not there.
How it works
From a finished chat to a working record
This runs on its own after every conversation. There is nothing for you or the visitor to trigger.
- 01
Conversation ends
The visitor closes the chat or hangs up the call. Nothing is asked of them.
- 02
A reviewer reads it
A reviewer goes through the entire transcript, start to finish, including the parts nobody would scroll back to.
- 03
Fields get filled
Score, tags, action items, knowledge gaps — and any custom fields you defined — are filled from the transcript.
- 04
Actions can run
The reviewer can save a lead or ticket to your datastore, or run a follow-up in a connected app.
What comes back
Built-in fields — plus any you add
Defaults keep conversations comparable. Custom fields let you capture what your process actually needs — then actions can land that data where work happens.
Outcome
Whether the conversation completed, was abandoned, needs a follow-up, or should be escalated to a person.
Sentiment
How the visitor actually felt across the conversation, not just how politely it ended.
Goal score
A 0 to 100 read on how well the conversation got the visitor what they came for.
Key points
The handful of things that actually mattered, pulled out of a long back-and-forth.
Next steps
Anything promised or implied, each with a priority so your team knows what to pick up first.
Unanswered questions
What the agent was asked but could not answer. This is the list that tells you what to add to your knowledge base.
Tags
Short labels that make conversations groupable, so themes show up without anyone tagging by hand.
Who does the reviewing
Start with the default, replace it when you outgrow it
You should not have to configure anything to get value on day one — and you should not be stuck with our fields on day ninety.
Built in
The default reviewer
Change nothing and it still works. Every conversation comes back with a summary, an outcome, sentiment, a score, key points, next steps, unanswered questions, and tags.
- No setup, no prompt writing
- Sensible for support, sales, and feedback alike
- Good enough to run in production on day one
Your own
A reviewer you define
When the default fields stop matching your business, build your own reviewer: tell it what to look for, add custom fields, and decide which tools it can use.
- Write the instructions in plain language
- Add custom fields — score, tags, action items, knowledge gaps, or your own
- Allow datastore writes and connected-app actions after the review
After the review
Not just a record — real follow-through
Filling fields is step one. The same reviewer can write to your datastore and run actions in the apps you already connected.
Save to a datastore
Create or update leads, support tickets, feedback, bookings, or any custom table you built — so the outcome lands where your team already works.
Act in connected apps
Draft a follow-up email, notify a channel, or sync a CRM record using the same app connections your other agents already use.
Custom fields you choose
Beyond the defaults, define the exact shape of the record: deal size, competitor mentioned, refund requested, NPS-style score, or whatever your process needs.
Never blocks the visitor
Extraction and actions run after the conversation closes. A slow follow-up never freezes the chat or the voice call.
Why it matters
What teams do with these records
A pile of transcripts is a cost. A pile of structured records is a feedback loop.
Find the gaps in your knowledge
The unanswered-questions list is a ranked to-do for your knowledge base, written by your own customers instead of guessed at in a meeting.
Work the follow-ups that matter
Instead of scanning transcripts for who needs a callback, your team opens a list of next steps that is already prioritised.
Watch sentiment move
One unhappy conversation is noise. A month of sentiment scores after a pricing change is a signal you can act on.
Prove the agent is working
Goal scores and outcomes give you a straight answer to how well the agent is doing, without reading a single log.
Setup
Three steps, no code
Open your agent
Go to the agent you already built and find its post-conversation section.
Pick a reviewer
Keep the built-in default, or choose a reviewer you have written yourself.
Let it run
From the next conversation onward, every one that ends comes back as a record. Nothing else to wire up.
Related feature
Voice-to-voice agent
See how visitors hold a spoken conversation with your website agent, and how those calls become records too.
Explore
Use cases
Support, lead capture, feedback, bookings, and product lookup — and what a finished conversation looks like in each.
FAQ
Common questions
Does this slow down the conversation?
No. The review only starts once the conversation is over, so the visitor never waits on it. If a review ever fails, the conversation itself is completely unaffected.
Do I have to set anything up?
No. There is a built-in reviewer working from the start, so conversations come back with a summary, outcome, sentiment, score, key points, next steps, unanswered questions, and tags without any configuration.
Can I change what gets extracted?
Yes. You can write your own reviewer, describe in plain language what to look for, and choose the exact fields you want back — including score, tags, action items, knowledge gaps, and any custom fields. Useful when you care about things like budget, a competitor being mentioned, or a refund being requested.
Can it do more than fill in a record?
Yes. After reading the transcript, the reviewer can write entries to your datastore and run actions in your connected apps — for example saving a lead and drafting a follow-up email — without making the visitor wait.
Does it work for voice calls too?
Yes. Voice conversations are transcribed while they happen, so a call is reviewed exactly the same way a text chat is. Long calls are handled without dropping the parts where intent and outcome usually live.
Will it invent things that were not said?
The reviewer is held to the transcript. If the conversation gives it no basis for a field, it leaves it out rather than guessing, and it takes the conservative reading when a conversation is too thin to judge.
Can different agents review differently?
Yes. The reviewer is chosen per agent, so your support agent and your sales agent can extract completely different things from their conversations.