AI in clienteling: how next-best-message works, and where privacy sets limits

KEY TAKEAWAYS Summary by the editors

1. Next-best-message is a model that ranks clients and possible messages, such as a new arrival, a restock or a trade-in offer, by their likely relevance, and hands the associate a shortlist with a draft text.
2. Under GDPR Article 21, clients may object at any time to direct marketing, including profiling related to it, and their data must then no longer be used for that purpose.
3. GDPR Article 22 restricts decisions based solely on automated processing that have legal or similarly significant effects, which is why a human should decide on offers such as credit, exclusions or individual pricing.
4. Under the EU AI Act, Article 50 transparency obligations for AI systems that interact directly with people apply from 2 August 2026, which matters for client-facing chat assistants.
5. Trade-in and resale offers work best in clienteling when they are based on items the client actually bought and when the client has agreed to that kind of outreach.

AI in clienteling mostly means next-best-message: a model ranks which clients a store associate should contact, which product, service or trade-in offer is most relevant to each, and drafts a message the associate can edit and send. The value lies in saving time and improving relevance. The limits are set by data quality, by clients' rights to object to profiling for marketing, and by rules on automated decisions and AI transparency in the EU and Switzerland.

What is next-best-message in clienteling?

Next-best-message (or next-best-action) is an established technique from banking and telecoms, adapted to fashion retail. Instead of a marketing team planning one campaign for thousands of clients, a model scores many possible actions for each client and picks the most promising one. In clienteling the output is not sent automatically. It appears in the associate's app as a task: contact this client, mention this item, suggest this appointment.

Typical message types in fashion include new arrivals in a client's preferred brands or sizes, a restock of an item the client asked about, care and alteration reminders, invitations to events, and increasingly trade-in or resale offers for items bought in earlier seasons.

How does AI decide who to contact and what to say?

Most systems combine three layers. A selection layer estimates which clients are likely to respond now, using recency of purchases, browsing, past replies and seasonality. A recommendation layer matches products or offers to each client, using purchase history, sizes, style attributes and stock. A generation layer, often a large language model, drafts the message in the associate's tone and the client's language.

The quality of each layer depends on the one before it. A fluent draft built on a wrong size or an item that sold out yesterday does more harm than no message at all, which is why stock and product data feeds matter as much as the model.

Read also

Client data and consent in clienteling: a GDPR and Swiss nFADP checklist

How can trade-in and resale offers fit into clienteling?

Brands with take-back or resale programmes can use clienteling to propose a trade-in at the right time, for example when a client buys a new coat and owns an older one from the same line. This works best when the offer relies on verified purchase history rather than guesses, when the associate can explain condition criteria and the value of a credit, and when the client has agreed to receive such offers. A trade-in suggestion that arrives from a personal advisor can feel more like service than promotion, but it is still direct marketing in legal terms.

What are the privacy limits under GDPR and Swiss law?

Several rules shape what AI clienteling can do in Europe and Switzerland:

• Right to object (GDPR Article 21): where data are processed for direct marketing, the person may object at any time, including to profiling related to that marketing, and the data must then no longer be processed for those purposes.
• Legitimate interest (GDPR Recital 47): direct marketing may be based on legitimate interest, but only where clients can reasonably expect the processing given the context in which data were collected.
• Automated decisions (GDPR Article 22): people have the right not to be subject to decisions based solely on automated processing, including profiling, that produce legal or similarly significant effects. Excluding clients from offers or setting individual prices automatically deserves careful review.
• Swiss nFADP: the revised Swiss data protection law, in force since 1 September 2023, introduced profiling as a legal concept and requires privacy by design and by default.
• EU AI Act Article 50: from 2 August 2026, AI systems intended to interact directly with people must inform them that they are interacting with AI, unless this is obvious from the context.

In practice, an AI that suggests to an associate whom to message is less sensitive than an AI that messages clients directly. The more autonomous the system, the more transparency, documentation and human oversight it needs.

Where does AI clienteling go wrong?

Common problems are not exotic. Models trained on past campaigns favour clients who already buy often, so the same people are contacted repeatedly while lapsed clients are ignored. Drafted messages repeat product copy and lose the associate's voice. Recommendations break when product attributes are incomplete or stock data lags. And if associates cannot see why a client was suggested, they stop trusting the list. Contact frequency caps, explanations next to each suggestion and a simple way to dismiss a task help keep the tool credible.

Read also

How do you measure clienteling ROI across store, chat and online?

What should a sensible set-up look like?

1. Record consent and channel preferences at the point of capture and make them visible in the associate app.
2. Limit the model to data clients would reasonably expect a store to use, and document the purpose.
3. Show the reason for each suggestion and let associates edit or reject drafts.
4. Set contact frequency limits per client across all channels.
5. Measure outcomes against a control group before scaling.

Frequently asked questions

What is next-best-message in retail?

It is a model that selects, for each client, the most relevant next communication, such as a new arrival, a restock or an event invitation. In clienteling the suggestion is usually passed to a store associate, who edits and sends it.

Do I need consent to use AI for clienteling messages?

Not always. Under GDPR, direct marketing can rely on legitimate interest if clients can reasonably expect it, but electronic marketing rules and the right to object still apply. Sensitive data and high-risk profiling under Swiss law require express consent.

Can AI send clienteling messages automatically?

Technically yes, but fully automated sending changes the nature of clienteling and raises transparency duties. From 2 August 2026, AI systems interacting directly with people in the EU must disclose that they are AI unless obvious.

How do trade-in offers fit into clienteling?

Associates can suggest a trade-in or resale credit when a client buys a related new item, based on verified purchase history. Such offers count as direct marketing and should respect consent and frequency limits.

GuideThe complete guide to AI in fashion e-commerce, marketing and retailRead the complete guide