AI how-to guides Customers and sales
How to personalise customer experience with AI
Build a single, consented view of each customer, let AI rank products, content and messages from their behaviour and context in the moment, and serve that decision to every channel from one place. Prove the lift with holdout tests before scaling. People set the rules, the brand limits and what must never be personalised.
What AI does
- Reads behaviour, purchase history and context in real time to infer what a customer is trying to do.
- Ranks products, articles and offers for each person instead of for a broad segment.
- Adapts copy, order and timing of messages within the limits you set.
- Serves the same decision to web, app, email and support so channels do not contradict each other.
- Runs experiments and reports which personalisation choices actually change outcomes.
What stays with people
- Deciding what may be personalised, for whom, and which data may be used.
- Setting brand, fairness and pricing rules that the model cannot override.
- Reviewing test results and choosing what to roll out.
- Handling complaints and requests from customers about how their data is used.
How to put it in place
-
Start from identity and consent
Join customer records across channels into one profile and attach consent and suppression flags to it. Anything without a lawful basis stays out of the model from day one.
-
Pick one journey with a clear outcome
Choose a moment where relevance matters and the result is measurable, such as product discovery, onboarding emails or the support home page. Define the metric you want to move and the baseline you have today.
-
Encode the rules before the model
Write down brand guardrails, fairness checks, frequency limits and exclusions, and put them in the decision layer so they apply to every recommendation. Decide what happens when there is too little data about a customer.
-
Test against a holdout
Keep a control group that gets the current experience and compare it with the personalised one. Look at the business metric and at signals of trust, such as unsubscribes and complaints, not only at clicks.
-
Extend to other channels from one service
Once a journey shows lift, let other channels consume the same decisioning service with the same rules. Monitor drift, fairness and opt-outs, and retune with the business owner.
What you need
- Customer and behavioural data you are allowed to use, with consent recorded.
- A product, content or offer catalogue with clean, consistent attributes.
- Integration points in the channels: website, app, email platform and support tools.
- An experimentation setup to run holdouts and measure lift.
- A business owner for each journey and someone responsible for data protection.
When it is not worth it
- Small catalogues or few customers, where simple rules give the same result.
- When customer data is fragmented and there is no consent record: fix that first.
- Channels you cannot change quickly enough to act on the decision.
Risks and how to control them
- Using data without a lawful basis or beyond what customers expect: build on consent, minimise data, and assess profiling under data protection rules.
- Manipulative or unfair treatment, such as exploiting vulnerabilities or excluding groups: set explicit fairness checks and exclusions, and be careful with practices the AI Act restricts.
- Creepy experiences that damage trust: limit sensitive inferences and let customers see and change their preferences.
- Optimising a click metric at the expense of the business: measure against a holdout on outcomes that matter and watch trust signals.
How Thinkia does it
We lay the foundation first: customer identity, consent flags and suppression rules. Ranking logic, fairness checks and brand guardrails are encoded before any live traffic, and A/B and uplift measurement prove the result against a holdout without breaking customer trust.
The personalisation layer runs on Synapse and works without third-party cookies or non-consented data. Web, app, email and care channels then consume the same decisioning service with shared governance, and we prioritise journeys where the lift can be measured.
The Thinkia solution behind this guide
- AI personalisation engineEvery customer gets a different experience. Built for them.
- Marketing personalization at scalePersonalization at scale, governed by consent
Thinkia products involved
Frequently asked questions
What is AI personalisation?
It means deciding, for each customer and moment, which product, content or message to show, based on their behaviour and context rather than on the average of a broad segment.
Is it legal under GDPR?
It can be, with a lawful basis, transparency, data minimisation and respect for people's rights. Profiling that has significant effects on people carries extra obligations, so involve your data protection officer early. This is not legal advice.
Do we need a customer data platform first?
You need a reliable, consented customer profile. Whether that comes from a customer data platform or another source matters less than having identity, consent and history in one place the decision layer can read.
How do we know personalisation is working?
Keep a holdout group that gets the generic experience and compare business outcomes between the two. Without a control group, any improvement could be coming from somewhere else.
Can we personalise without third-party cookies?
Yes. First-party behaviour, purchase history and context within the session are usually the strongest signals, and they come with a clearer consent trail.