Every business has a list it never looks at: the people who came twice and then stopped. The couple who booked a table for an anniversary last spring. The client who finished a course of treatments and never rebooked. The tenant who asked about a villa, stayed once, and disappeared. They did not complain and they did not leave a review. They simply went quiet.
Bringing them back is the least glamorous work in a small business, and the first thing that gets dropped. It means sitting down with a list of names, remembering what each person took and when, and writing a message that does not read like a mailshot. That is exactly the kind of patient, repetitive, personal work an AI agent can carry, as long as it is taught restraint.
What the win-back and retention role does
In the SVI agent builder, win-back and retention is a ready role. You switch it on next to the other customer-facing agents: bookings, first-line support, reviews. It is not a mailing tool with a send button; it is an agent with rules, and the rules are what make it usable:
- It works from your own history. Who came, what they took, how long ago, in which channel they wrote.
- It writes to one person at a time, about what that person actually bought, not about everything you sell.
- It handles the reply. Win-back is not a broadcast. The point is the conversation that starts when someone answers, and the agent carries that conversation to a booking request.
- It stops when it should. A no ends the sequence for that person, and so does silence.
The last point is the one most campaigns get wrong, and the one that decides whether your name stays welcome in a customer's messenger.
Who counts as a customer who went quiet
There is no universal number of days after which a customer is lost. A hair color grows out in weeks; a dental check-up comes round twice a year; a family that rents the same villa books once a season; a restaurant regular may eat with you every Friday or twice a year, and both are normal. A rule copied from someone else's industry will either nag people who were never gone or reach the rest long after they stopped caring.
So the useful definition is relative: a customer is quiet when they have missed their own rhythm. You describe that rhythm to the agent in plain words, the way you would explain it to a new administrator:
- Services with a natural interval. "Color clients usually come back within six to eight weeks. After ten, write."
- Seasonal businesses. "Guests who stayed last high season should hear from us before booking opens, not in the empty months."
- Occasion businesses. "If someone booked for a birthday, the moment to write is a few weeks before the next one."
- One-time visitors. "Anyone who came once and never returned gets one message, not a sequence."
Those sentences are the whole configuration. There is no segment builder to learn and no code to write, and you can change the wording later in the same way.
Tact: what a good win-back message sounds like
A win-back message is read by someone who owes you nothing. It has one job: to be easy to answer. Everything that makes it louder makes it easier to ignore. What works, in any language:
- It names something real. The treatment they had, the table they liked, the apartment they stayed in. Specific beats friendly.
- It asks instead of announcing. A question invites a reply; a banner invites a swipe.
- It offers a step that is easy to take. A time that is open this week, a short answer to a question they once asked.
- It is short. Two or three sentences in the channel where that person already writes to you.
- It does not promise anything you have not approved. No invented discount, no gift you never agreed to.
That last line matters more than it sounds. Most win-back campaigns reach for a discount because a discount is easy to write. An agent that is not allowed to improvise one has to do the harder and better thing: remind the person why they liked you.
How often to reach out
Frequency is where goodwill is spent. The pattern we set by default is deliberately slow:
- One message, at the moment that fits the customer's own cycle. Not on a campaign calendar, and not because the month was slow.
- No answer means no follow-up for a long while. Silence is an answer, and pressing it is how a business turns into spam.
- A second touch only with a new reason. A new season, a service they asked about that now exists, the date they usually book. Not the same message again with a bigger exclamation mark.
An agent is very good at holding this discipline, because it never has a bad week that makes it want to message everyone at once. It is also very good at the bookkeeping behind it: who was already contacted, who said no, who asked to be left alone.
What the agent never does
The same rules that govern every SVI agent apply here, and in win-back they matter twice over, because the agent writes first:
- It answers only from your own data. It does not invent prices or open slots and it does not grant discounts you have not approved.
- It keeps customers apart. It never reveals one customer's data to another.
- A booking is a request until a person confirms it. If a quiet customer comes back and wants Saturday, that request waits for your yes.
- It does not pretend to be a human. It can be warm and natural, and it does not deny being an AI assistant.
- When unsure, it hands over. A complaint, an unusual request or a confused reply goes to a person with a short summary.
Where this role already works
Win-back is rarely a department of its own; it sits inside a team. In the ready company configurations it comes switched on where returning customers carry the business: in the AI receptionist for spas, salons, hotels and clinics, in the restaurant and delivery team you can install from ready companies, and in logistics, where a client who stopped shipping with you is worth a call.
That matters practically. The agent that writes to a quiet guest and the agent that takes the booking are in the same team, reading the same data, so the conversation does not break at the handover. The reviews agent is there too, which is why a reply to an old complaint and a win-back message never arrive on the same day by accident.
Where the replies land
Everything that comes back arrives in one place: SVI CRM, as a request ready to confirm, or in your own CRM if you prefer to keep your list where it is. Customers answer in the channel they know, which is usually WhatsApp, Telegram, Instagram or email, and the agent follows them there rather than asking them to move.
If you want to feel the tone before you decide, talk to a live one. Andaman Lotus Spa is a fictional demo company running the real agents of the reception team. Tell it you were there months ago and never came back, and see how it handles you.
Test it on your own customers before you pay
Win-back is the one role where a mistake is visible to the people you least want to annoy, so it is checked before anyone receives anything. The launch goes in three steps: installation, on us; test mode, also on us, where the agents draft the real messages to your real quiet customers and nobody receives them, while you read every draft; and launch, where payment starts.
In test mode you are reading for the things only you can judge: is this the right moment for this person, does the tone sound like your business, is this someone you would rather call yourself. What you do not like, you change as a sentence of instruction, and you look again. Afterwards, the numbers worth watching are narrow and honest: how many quiet customers replied, how many of those replies became requests, how many asked to be left alone.
How to start
Open the agent builder and look at the customer roles: win-back and retention sits next to bookings, support and reviews, and a team of three or four of them is a realistic start for a small business. Or skip the settings and begin with a conversation with Mai: describe your business and how often your customers normally come back, and she will assemble the team and start the test mode with you.