Scrolling back through a long conversation to work out "what did this customer want and where did we leave it?" takes minutes. The section reduces that to one button: it reads the whole conversation, produces a two- or three-sentence summary, names the customer type and suggests an approach for the agent. This guide covers what the button does, what each field means, and what to watch out for.
AI profile
⚠️ The analysis never runs on its own, and when it runs it sends the person's messaging to the AI provider. That is why the button is drawn only for people with contact edit permission, and why every press spends your portfolio's AI credit.
In this article you will find:
What it is for
Running the analysis
What each field tells you
The Conversation records list
When a warning appears instead of the button
Limits and known issues
Privacy and credits
Common mistakes
What it is for
The section draws on two separate sources and shows them one under the other:
The analysis — Produced when you press the button. The whole conversation is read and a summary, a customer type, a suggested approach and topics of interest are derived.
Automation records — The AI conversation step writes what it collected in each run here. This part accumulates even if you never press anything.
The practical benefit shows up at handover: the agent taking over reads three lines instead of fifty messages and continues in the right tone.
The same section also appears in the contact panel in the inbox; both show the same data.
Running the analysis
Open the person's card (or the contact panel in the inbox).
Scroll to the AI profile section.
Press Analyse with AI.
After a short wait the summary, customer type, approach and topics appear, along with "Analysis complete."
If an analysis was run before, the button reads Analyse again. Re-analysing overwrites the old summary: you cannot keep two analyses side by side. It is there to refresh the summary as the conversation moves on.
If there are no messages to analyse, nothing runs and an error appears. You cannot analyse someone who arrived from an ad and never wrote.
On a contact who has not been analysed, the section is not blank; a sentence explains what will happen. Without permission it simply says "Not analysed yet."
What each field tells you
Summary — What the person wants and where the conversation stands, in two or three sentences.
Customer type — A short label such as "price-focused", "undecided" or "ready to buy". It is independent of your own tags and is never written into the contact's tag field.
Approach — A one-sentence suggestion for the agent: what tone to use and what to lead with.
Topics of interest — Subject names laid out as badges; at most eight.
Last analysed — The date and time of the analysis. That line tells you how current the summary you are reading is.
The model is restricted to what appears in the conversation; an empty field (customer type, for example) means it could not be derived, and that field is not drawn at all.
The values are written in the account's language. Even if the customer wrote in English, a Turkish panel gets a Turkish summary: the profile is shown to your team, not to the customer.
The Conversation records list
A collapsible list called Conversation records sits below the section; the number beside the heading says how many records there are. Tap it to open.
Each row holds:
Date and flow name — The name of the automation that left the record.
Note — The summary the AI left in that run. If the model left no summary, what the person wrote stands in for it.
Collected — What was written to which field in that run, as field: value badges. If nothing was collected it says "Nothing collected".
This list shows records left by the AI conversation step; fields typed in by hand or written by the Collect data step do not appear here. The other ways to see custom field values: Setting up and using custom fields.
Records are not cleared by an analysis: if the person triggers the same flow again, the list grows. At most 30 records are kept, newest first.
When a warning appears instead of the button
A yellow warning where the button should be means AI is unavailable at that moment. There are three possible reasons:
AI is switched off in your plan. A plan upgrade or a portfolio setting is needed.
No key. No AI key is defined in the portfolio.
Out of credit. The portfolio's AI credit is used up.
The warning text says which one it is. The Conversation records list keeps showing in that case: that part needs no analysis.
Under heavy use the provider can refuse a request temporarily; in that case try again after a while.
Limits and known issues
The analysis reads the newest 60 messages of the conversation; older ones are not included. On very long conversations the summary describes the recent period.
Only the first 400 characters of each message are read; very long messages are trimmed.
The last 10 automation records are placed beside the conversation, so what the person asked for previously is taken into account.
The summary is at most 1000 characters, the customer type 80, the approach 300; topics are limited to 8.
The Conversation records list keeps at most 30 records; older ones drop off.
The analysis does not refresh itself: as new messages arrive the summary ages, and you have to press the button to update it.
Re-analysing changes the old summary irreversibly.
The output is not saved as a tag or a field and cannot be used as a condition in automations.
With no messages at all, the analysis will not run.
Privacy and credits
The analysis sends the person's messaging to the AI provider. For that reason:
The button is pressed by hand; the panel never analyses anyone on its own.
The button is drawn only for roles that can see and edit the contact; the messaging of someone outside your scope cannot be analysed.
Every analysis spends your portfolio's AI credit. Analysing hundreds of contacts one after another burns through it quickly.
For conversations carrying sensitive data (health, payment details), decide for yourself whether to run the analysis.
A worked example
At an estate agency the shift handover works like this:
The evening agent opens the cards of the three people who had long conversations during the day.
On each of them they press Analyse with AI.
The first summary says "looking for a three-bedroom, budget is firm, wants a viewing at the weekend"; the customer type is ready to buy and the approach is "offer a viewing".
The agent offers a viewing without reading fifty messages.
On the second person the customer type comes back as price-focused, so the agent leads with the price list.
On the third, the Conversation records list shows what was written to the budget field a week earlier, and the same question is not asked twice.
Common mistakes
Expecting the analysis to run on its own. The button has to be pressed.
Assuming the summary is current. Check the Last analysed line; messages after that date are not in the summary.
Treating the customer type as a tag. It is never written to the tag field and cannot be used in a filter. If you want to filter by it, add a tag yourself.
Thinking re-analysing keeps the old summary. It overwrites it.
Assuming the whole conversation is read. The newest 60 messages are.
Trying to analyse someone who never wrote. Without messages the analysis does not run.