Reply Actions

Reply actions let the agent read the replies to a campaign and do something useful with each one.

You set them up per campaign, on the Reply actions tab of the composer.

What Happens to a Reply​

When someone replies to one of your campaign emails, the agent works through four questions in order. Any one of them can end the process.

1. Was this a reply to one of our campaign emails? If not, the agent does nothing at all.
2. Are we allowed to answer? Out-of-office messages, mailing-list traffic, and robot mail are filtered out here, before anything else runs.
3. Which class is it? The agent reads the reply and picks one of your classes.
4. What does that class do? Any side effects first, then one reply.

Turning It On​

Open a campaign, go to the Reply actions tab, and switch Enabled on.

You need at least one class before it will turn on. New campaigns start with four ready-made ones, which you can edit, delete, or add to.

Classes​

A class is a kind of reply. Each one has a name, a description, and a set of actions.

The four you start with:

Class	Description	What it does out of the box
Positive intent	The reply says yes, asks for the offer, or wants to talk.	Adds the tag warm-reply.
Question	The reply asks something about the product, the price or the offer.	Nothing yet — add what you want.
Wants to unsubscribe	The reply asks to stop receiving emails.	Tags the contact and unsubscribes them.
Auto-reply or out of office	An automatic reply that nobody actually wrote.	Nothing. Correct — answering one starts a loop.

Writing a good class​

This is the part that decides whether the whole feature works.

The name and the description are what the classifier reads. They are not labels for your own benefit — they are the instructions. Vague descriptions produce vague sorting.

• ❌ Bad: name "Interested", description "interested people"
• ✅ Good: name "Wants a demo", description "The reply asks to see the product, requests a call or a demo, or asks about availability for a meeting."

Some rules of thumb:

• Describe the reply, not the person. "The reply asks about pricing" beats "price-sensitive lead".
• Give examples of the words people actually use. "Says something like 'how much', 'what does it cost', 'send me pricing'."
• Keep classes clearly apart. If two classes overlap, the agent has to guess, and so would you.
• Cover the boring cases too. Most replies are not exciting. A class for "not right now" saves you from mis-sorting them.

What a Class Can Do​

Each class has two parts: side effects, which change the contact, and one reply, which decides what goes back.

Side effects — "Also do this"​

Any number of these, in any combination. They always run before the reply.

• Add tags — pick existing tags or create new ones.
• Remove tags — clear a tag the contact no longer deserves.
• Update contact fields — write a value into any contact field.

Tagging is the highest-value thing you can do here. A tag turns a reply into something you can filter, segment, and follow up on later.

The reply — one at most​

This is a single choice per class. That is deliberate: two replies to one message would be two answers to the same question.

Choice	What it does
Do not reply	Records the class, runs the side effects, sends nothing.
Send an email reply	Sends a template you write, in the same email thread.
Hand it to an AI agent	An AI agent reads the campaign and the reply, then answers.
Unsubscribe the contact	Opts them out. Sends nothing.

Send an email reply​

Write the reply once and it goes to everyone in that class.

• Subject (optional) — leave it empty to answer in the same thread with the original subject. That is usually what you want.
• Reply body — the message itself.

Hand it to an AI agent​

Pick one of your AI agents. It reads the campaign email and the reply, then writes an answer in the same thread.

Use this for classes where a canned answer will not do — questions, especially. The agent answers that one message; it does not take over the conversation permanently.

Unsubscribe the contact​

Two things happen: the contact is removed from the campaign's lists, and they are opted out of marketing email generally. No reply is sent — someone asking to be left alone should be left alone.

This is the one action worth setting up even if you use nothing else on this page. Someone who asks to be removed and keeps getting mail is the person most likely to report you as spam.

Safety Limits​

The agent will not get into an argument with a robot. Several limits apply, and none of them are configurable:

• One automated reply per incoming message, ever. A message that has been answered is never answered again.
• At most 2 replies to the same contact on the same campaign per day.
• A lifetime ceiling per contact per campaign, so a slow back-and-forth dies out instead of refilling every night.
• Automatic mail is never answered. Out-of-office notices, vacation responders, and mailing-list traffic are detected and skipped — and they are not counted as replies in your stats either.

Automated replies are treated as normal one-to-one email, so they carry no unsubscribe footer. That is correct: they answer a message the person just sent you. Opt-outs and do-not-contact rules still apply, exactly as they do everywhere else.

Replies Have to Reach You​

None of this works if replies never arrive.

This is a genuinely common problem. Your campaign goes out through the platform, but your mailbox lives somewhere else — Google, for example. A reply to hello@yourcompany.com goes to Gmail and the agent never sees it.

The check​

When you save a campaign with reply actions enabled, the workspace checks whether replies to your reply address would actually arrive. An address counts as reachable in any of three ways:

1. It is already a workspace address. Nothing to do.
2. Your domain is set up for inbound mail. The domain is verified for email, inbound is switched on, and its MX record points here and has been verified.
3. Mail has arrived at that address before. The proof is that it already happened.

The test dialog​

If none of the three apply, a dialog appears when you save:

Make sure replies reach the agent Replies to hello@yourcompany.com have to arrive in this workspace for reply actions to run. Forward that mailbox to this address:

It gives you a forwarding address and a Copy button. Here is how to pass it:

1. Copy the forwarding address.
2. Open your mail provider and add a forwarding rule from your reply mailbox to that address.
3. Send a test email to your reply address.
4. Click "I've sent a test". The dialog watches for it and confirms: "Got it. Replies will reach the agent."

If nothing arrives, it says "Nothing yet." You can Keep waiting or Save anyway — the check is advice, not a gate. Closing the dialog cancels the save, so your campaign stays open.

A Setup That Works​

If you want somewhere to start, this handles most campaigns:

Class	Side effects	Reply
Positive intent	Add tag warm-reply	Do not reply — you want to answer these yourself
Question	Add tag has-question	Hand it to an AI agent
Not right now	Add tag follow-up-later	Do not reply
Wants to unsubscribe	—	Unsubscribe the contact
Auto-reply or out of office	—	Do not reply

Tag everything, answer the questions, opt out the people who asked, and keep the good replies for a human.

---

Next Steps​

• Campaigns — where the Reply actions tab lives
• The Email Agent — how replies feed back into who gets mailed next
• Lists & Segments — how unsubscribes are recorded