Customers

AI personalization that respects the GDPR

AI personalization lets each customer receive messages and offers that are more useful to them. This guide explains how to do it within the GDPR and UK data protection law, with three simple questions before every campaign, the data worth using and the tools you can choose.

What AI personalization is and why the GDPR applies

AI personalization fits within the GDPR as long as you know which data you use, on what legal basis, how you explain it to customers and which tools process it. That matters to any small business that wants to stop sending the same newsletter to everyone, such as an online pet supplies shop in Nottingham whose marketing lead is asking an AI assistant to draft a different email for each customer based on what they bought.

AI personalization means adapting what each customer sees, such as an email, a product suggestion or the order of a catalogue, based on data about that person or their situation. AI makes it possible at scale, with wording and suggestions that change for every recipient. When the tailoring reaches each individual and updates with their recent behaviour, it is often called hyper-personalization, and that is where data protection law asks for the most care.

As soon as personalization uses data about identifiable people, the GDPR applies, or in the UK the UK GDPR alongside the rules on electronic marketing known as PECR. A purchase history linked to a customer already counts. This guide gives a general overview and is no substitute for advice from a lawyer or your data protection officer.

Four levels of personalization, from least to most data

The more personalization relies on personal data, the more obligations it brings. In practice, start with the lowest level that meets your goal.

  • Context

    Content adapts to the situation at hand, such as browser language, country or the page being viewed. A shop shows delivery times for the visitor's country. This usually works with data that is neither stored nor linked to a specific person.

  • Stated preferences

    The customer tells you what they want in a form, such as topics, budget or how often to hear from you. It is the most transparent level, because people know what they shared and why.

  • Behaviour

    You use what the customer does, such as pages viewed, past purchases or emails opened. If tracking relies on non-essential cookies or similar technologies, it needs prior consent.

  • Profiling

    Data is combined to assess or predict aspects of a person, such as how likely they are to buy or what interests them. The GDPR calls this profiling and has specific rules for it, with more transparency and, in some cases, a data protection impact assessment.

Three questions before you send a personalized campaign

Data protection law takes account of what people can reasonably expect from their relationship with you. These three questions turn that idea into practice. Ask them with a real message in front of you, not the general idea of the campaign.

  • Would they be surprised to learn where the message comes from?

    "We're writing because you bought dog food a month ago" surprises nobody. "We noticed you browse our site late at night" does. If the honest answer is that the customer would find it odd, drop down a level.

  • Can you explain it in one line at the bottom of the message?

    If you cannot write next to the recommendation what it is based on without it sounding wrong, the problem lies in the personalization and not in the wording. That line is also good transparency practice.

  • Is the AI inferring something the customer never told you?

    From purchases or browsing, a model can infer health, family circumstances or financial situation. Health, religion and sexual orientation are special category data with much stricter rules. Exclude such inferences from your instructions and your segments.

Legal bases, consent and automated decisions

Every use of personal data needs a legal basis. For personalization, the usual ones are contract, legitimate interests and consent. Suggesting products based on past purchases can rest on legitimate interests, provided you have weighed your interests against the person's rights and recorded that assessment. People can always object to direct marketing, including any profiling tied to it.

Marketing emails have rules of their own. They generally need consent, with a common exception for existing customers and products similar to those they bought, and every message must offer a simple way to unsubscribe. Website tracking falls under separate cookie rules, which require prior consent for anything not strictly necessary. Refusing should be as easy as accepting, boxes cannot be pre-ticked and consent can be withdrawn at any time.

Article 22 of the GDPR gives people the right not to be subject to a decision based solely on automated processing, including profiling, that produces legal effects or similarly significant effects on them. Automatically turning down a credit application is a typical example. A product recommendation normally falls well short of that threshold. Excluding certain customers from important terms because of their profile could come closer.

Such decisions are only allowed in specific cases, for example when they are necessary for a contract, authorized by law or based on explicit consent. Even then, the person can ask for human intervention and contest the decision. For a small business, the practical answer is usually that AI suggests and a person decides whenever the outcome really matters to the customer.

How to set up AI personalization, step by step

A sequence that works for an online shop, an accounting firm or a sales team. Each step leaves a written note, which you will be glad of if a customer or a regulator asks.

  1. 01

    Define the goal

    Write down what you want to achieve and for whom, for example reminding customers when to reorder a product they buy often. A specific goal tells you which data you actually need.

  2. 02

    Minimize the data

    Data minimization is the principle that you should use only the data needed for the purpose. List what you hold and keep only what the goal needs. For a reorder reminder, the product and the purchase date are enough. Set how long you keep the data, too.

  3. 03

    Choose and record the legal basis

    For each use, decide whether it rests on contract, legitimate interests or consent, and write it down. Check your cookie banner and sign-up forms so that consent is collected properly.

  4. 04

    Pick tools with proper agreements

    Check that the AI provider signs a data processing agreement, where it processes the data and whether it uses it to train its models. If data leaves the UK or the European Economic Area, the transfer needs appropriate safeguards. Where you can, give the AI pseudonymized data, meaning without names or email addresses, and add them back afterwards.

  5. 05

    Ask the three questions and review

    Before each campaign, put several real messages through the three questions. Afterwards, review results, unsubscribes, complaints and whether the AI produces unsuitable or discriminatory suggestions.

Example

Reorder reminders at an online pet shop

Back to the Nottingham shop from the start, a typical case with a small team and a customer base that buys pet food regularly. The goal is to remind each customer when their bag is probably running low and suggest accessories that suit their animal.

The marketing lead uses only two kinds of data. Purchase history, with product, bag size and date. And the species and size of the animal, which customers give in an optional form that explains how it will be used. She leaves out browsing data from anyone who declined cookies and forbids, in the AI instructions, any inference about the animal's health or the customer's finances. Her adviser confirms that reminders about similar products to existing customers fit the usual exception, with an unsubscribe link in every email.

An automated workflow prepares the week's reorder list. The AI receives only the product, the date and the type of animal, without names or email addresses, and drafts the text. The workflow then adds the customer's name and a footer line, "we're writing because you bought this food a few weeks ago". During the first month, a person checks a sample before each send.

To measure it, she compares reorders from customers who receive the reminder with a group that only gets the general newsletter. She also tracks unsubscribes, complaints and any access or objection requests. If unsubscribes rise, she reviews tone and frequency before adding any more data.

One campaign, with its legal basis checked

Pick one campaign, write its goal in a sentence and list the data it uses. Ask the three questions of three real messages and check the legal basis with your adviser. That gives you a first personalization you can defend.

Obinova is an AI agency based in Barcelona and Paris. Our AI audit and consulting starts from your real processes. We review with your team which customer data you use and where personalization adds value, then prioritize use cases with you. A mission begins with a launch kick-off, followed by follow-ups several times a year. Legal sign-off stays with your legal adviser.

Our training prepares your team to use ChatGPT, Claude, Gemini or Microsoft Copilot with good habits, such as knowing which data to keep out, how to pseudonymize and how to check what the AI suggests before it reaches a customer. It is delivered on a dedicated video platform or in person at your company. If you decide to automate, we build the workflow in n8n and document it with your team.

Frequently asked questions

Is AI personalization legal under the GDPR?

Yes, as long as you follow the rules. You need a legal basis for each use of data, clear information for customers, only the data you need and providers with proper safeguards. Tracking with non-essential cookies also requires prior consent. The UK GDPR follows the same logic. For questions about your specific situation, ask your legal adviser.

Do I need consent to send personalized recommendations?

It depends on the channel and the data. Non-essential tracking cookies need consent. Marketing emails usually do too, with a common exception for existing customers and similar products, always with an unsubscribe option. Suggestions based on past purchases can rely on legitimate interests if you have weighed and recorded them, and people can object at any time.

Personalization vs profiling: what is the difference?

Personalization means adapting content to a person or their context. Profiling is a specific kind of personalization that combines data to assess or predict something about someone, such as how likely they are to buy. Showing delivery times for their country is not profiling. Scoring each customer on their behaviour is, and it comes with more obligations.

What is hyper-personalization, and how does it differ from personalization?

Classic personalization works with segments, such as every customer who bought dog food. Hyper-personalization uses AI to tailor each message to one specific person, from their history and recent behaviour. That usually puts it in the behaviour and profiling levels, which carry the most GDPR obligations, so use it only where the goal justifies the extra data.

Can I use ChatGPT to personalize customer emails?

You can, with care. Use an approved business account and check the provider's terms on data use, the data processing agreement and where data is processed. Enter only what is needed, such as the customer's area of interest, without names or sensitive data, and read every draft before it is sent.

Do I need a DPIA for AI personalization?

It depends on the risk. A data protection impact assessment is required when processing is likely to be high risk, for example systematic and extensive profiling or combining many data sources. Regulators such as the ICO publish lists of processing that needs one. A reminder based on purchases is usually far from that threshold, but record your reasoning and check it with your adviser.

Personalize with well-managed data

Book a free 30-minute video call. We will look at the customer data you use, which campaign you could personalize first and the tools you have, and tell you where to start safely.

Free 30-minute diagnosis →
Free diagnosisWhatsApp

WhatsApp · Reply within 2 hours during business hours