Where this lives. Sign in and open /reply-intelligence in the app.
What it does
Reply Intelligence is the classification log behind every reply in the Inbox: what category each thread landed in, why, and, when the workspace has an AI key active, an additional AI-read intent, summary, urgency, and suggested next step.
Before you start
- Opt-outs, bounces, and out-of-office replies are always decided deterministically from headers and explicit language. No AI key changes those calls.
- The AI intent, Summary, Urgency, and Suggested next step columns only populate when an AI key is active for the workspace.
- Manual overrides happen in the Inbox thread view, not on this page. See Inbox.
Set it up, step by step
- Open Reply Intelligence to see every classified reply thread: Thread, Category, AI intent, Source (which engine produced the verdict), Urgency, Summary, Suggested next step, Needs review, and when it was classified.
- Filter by Category, or search by subject, classification, AI intent, or summary.
- Start with threads marked "Needs review." These are the low-confidence calls waiting on a human decision.
- Click a row to open its detail drawer.
- To correct a wrong call, click "Open inbox" and change the Categorize dropdown on that thread. The change is recorded as a human decision, not the classifier's, and it is final.
- Use "Leads who replied" to jump straight to the leads behind every positive, interested, meeting-request, or info-request thread.
What you should see
- Category values include positive, negative, neutral, out-of-office, unsubscribe, and wrong-person, plus finer categories such as meeting request, interested, info request, and do-not-contact once a thread is filed by hand.
- Unsubscribe or do-not-contact language in a reply stops that lead's sequence and suppresses the address. A plain negative reply stops the sequence without suppressing anyone.
- An out-of-office reply pauses that lead's sequence rather than ending it. One that says the person has left the company is filed as wrong-person instead, since it usually names the real contact to chase.
- A positive reply, however it is worded, stops the automated sequence and hands the thread to a person.
Common problems
- A message that both expresses interest and asks to be removed is classified as an opt-out, not positive. Unsubscribe language always wins.
- A bounce that reads like a policy or reputation refusal, for example mentioning being blocked, marked as spam, or blocklisted, is never treated as a dead address. The recipient is not suppressed for it; it surfaces instead on Deliverability as a receiver refusing your sender.
- If Source shows the deterministic engine rather than AI, that is expected either when no AI key is configured, or when the message hit a rule, such as a bounce, opt-out, or out-of-office, that AI is never allowed to override.
Related
Is an out-of-office message a reply?
No. It is an automatic acknowledgement from a mail server, and counting it as a reply inflates every rate that follows. It is classified separately for exactly that reason.
What happens when a reply is classified as an unsubscribe?
The person stops receiving the sequence, and the address is suppressed so no other campaign reaches them. That is the behaviour to expect from any tool that takes consent seriously.
Read next
Common questions
Is an out-of-office message a reply?
No. It is an automatic acknowledgement from a mail server, and counting it as a reply inflates every rate that follows. It is classified separately for exactly that reason.
What happens when a reply is classified as an unsubscribe?
The person stops receiving the sequence, and the address is suppressed so no other campaign reaches them. That is the behaviour to expect from any tool that takes consent seriously.
Read next
Last updated September 19, 2026. Written by the team that operates the platform.