AI and follow-up

How much of customer follow-up should you hand to an AI?

Hand an AI the part of customer follow-up that takes rigor: replying fast, following up, sending the recap of a meeting, keeping the tracking file up to date. A Muzau agent handles that part beside your team, who keep the relationship, the meetings and the decisions. You choose, action by action, what goes out on its own and what waits for your approval. An agent can also start in trial mode: everything it prepares then waits for your review.

Why handing over everything, or nothing, does not work

Customer follow-up mixes two kinds of work. The gestures that repeat: replying to a request that came in the evening, following up a quote on the right day, sending the recap of a meeting, noting where each deal stands. And what takes a person: understanding a need, agreeing a condition, reassuring an unhappy customer.

Leaving everything to the team means leaving the repetitive gestures to whatever time is left after the day's meeting. Handing everything to an AI means leaving it decisions that are not its to make. So the real question is: which gestures go out on their own, which wait for approval, and who takes over when the AI hands off.

How the agent goes about it

Input

Incoming requests, your meeting notes, the emails to follow up

It does

Replies, follows up, drafts recaps, and passes you what is beyond it

It delivers

Messages that go out on their own or wait for your approval, depending on your setting

Choosing, action by action, what the AI does on its own

On Muzau's Autonomy page, a single choice sets everything: "I review everything", where every action waits for your green light, or "They do it all on their own". You can also set each type of action separately: replying to an email, sending a follow-up, booking a meeting, updating a record, attaching a document, acting in a connected tool like Google Sheets, HubSpot or Pipedrive. The setting is made separately for the agents that talk to customers and for those that serve the team.

Starting in trial mode: everything goes through your review

A specific agent can be put in "trial mode": everything it prepares lands in "To approve", even if the rest of the business works on autonomy. If you say when creating it that nothing goes out without you, it starts in trial mode, like the default email follow-up agent. You can also set exceptions: "it replies on its own, but does not book meetings without me".

When follow-ups and recaps can go out on their own

A Muzau follow-up agent created from the Follow-ups page starts in trial mode: every follow-up waits for your approval. After 10 follow-ups in a row sent without edits, Muzau offers to let them go out on their own. The decision is yours, and "Not yet" resets the count to zero. For meeting recaps, the agent's page makes the same offer after 10 recaps in a row sent without edits. A recap with no certain recipient, with an attachment, or that cites an amount absent from the note and the Brain always stays in "To approve".

What comes back to the team when the AI hands off

When a request is beyond the Muzau agent, because someone needs to call back, decide, or check information it does not have, it passes it to you: a card in "To do" says who wrote, what needs doing and why. Muzau notifies you by email, then reminds you if it is still waiting after four hours, and in Slack or Microsoft Teams too if you connect them. To write to the customer in its place, see how to take over from an AI. Even on autonomy, an email reply the agent is unsure about waits in "To approve". The details are on the page can an AI that writes to your customers get it wrong.

Sharing customer follow-up between the team and an AI: the method

With or without a tool, the split is decided gesture by gesture, not in bulk.

• List the gestures of follow-up: acknowledging a request, answering a common question, following up a quote, sending a recap, updating the file, booking a meeting.
• For each one, two questions: does it repeat almost identically? Can a mistake be fixed with one email? Two yeses: it can be handed over.
• Keep with the team whatever commits the business: an off-list price, a discount, an unusual lead time, a complaint.
• Write these limits in short sentences: "never a discount", "never a delivery date without approval".
• Start with review. A type of message that goes out several times in a row without edits can then go out on its own; one you often correct stays in review.
• Name who takes over when the AI hands off: a request passed on that nobody opens is still a request without a reply.
• Now and then, check at random what went out on its own.

An example of a split, once the first messages have been reviewed: the acknowledgment and the quote follow-up go out on their own, the reply to a price question waits for approval, a complaint goes straight to someone on the team.

For what customer follow-up itself covers, see what customer follow-up is, and how to keep it up. And to know what the customer reads when the AI writes to them, see how to tell them it is an AI.

What it does not do

• It is the agent that judges that a request is beyond it or that it is unsure of its answer: that AI judgment can be wrong, which is why trial mode makes sense at the start.
• In the site chat, nothing is reviewed before it goes out: the visitor is waiting at the bubble, and only a reply held back by the automatic check does not go out and is passed to you.
• With no setting from you on the Autonomy page, agents act on their own, within the limits of their safeguards, except those that started in trial mode: choose "I review everything" or trial mode to review before sending.

Frequently asked questions

When can you let an AI follow up customers without review?

When it has proved itself on your own messages. A Muzau follow-up agent created from the Follow-ups page starts in trial mode, every follow-up waits for your approval; after 10 follow-ups in a row sent without edits, Muzau offers to let them go out on their own, and "Not yet" resets the count to zero.

Can an AI do customer follow-up in place of a salesperson?

No, not the part that makes the sale: the meeting, the negotiation and the decisions stay with the team. A Muzau agent handles the written part beside them: it replies, follows up, drafts recaps, and hands off through a "To do" card when someone needs to call back, decide or check information it does not have.

Can you review everything an AI writes to customers?

By email and by SMS, yes: with Muzau, the "I review everything" choice on the Autonomy page holds every action until your green light, and a specific agent can also be put in trial mode. In the site chat, the reply goes to the visitor without review, since they are waiting at the bubble.

Can you let an AI reply on its own, but not book meetings?

Yes. On Muzau's Autonomy page, each type of action is set separately: replying to an email can go out on its own while booking a meeting waits for your approval. An exception set on a specific agent does the same, for example "it replies on its own, but does not book meetings without me".

Which repetitive follow-up tasks should you automate first?

Those that repeat almost identically and where a mistake can be fixed with one email: acknowledging a request, following up a quote left unanswered, the recap of a meeting. Keep them in review at first. With Muzau, a follow-up agent created from the Follow-ups page starts that way in trial mode: every follow-up waits for your approval.