Your retargeting audiences are shrinking and your ad reports attribute fewer sales than before, even though your traffic holds steady. Many marketing managers see the same thing. The cause is often the decline of third-party cookies, the small files that an outside player places to follow a user from one site to another.
Marketing personalisation does not disappear as a result. It changes its raw material: your own data, the context of the visit, and AI models capable of drawing useful signals from them. Here is how an SME can approach it, step by step.
Why cross-site tracking is fading
Several browsers block third-party cookies by default. Some mobile systems ask the user’s consent before allowing advertising tracking across apps, and consent banners let everyone refuse. A share of your visitors therefore becomes invisible to tools that relied on this tracking.
Three uses suffer: advertising retargeting, which shows an ad to someone who has already visited your site, conversion measurement, and adapting pages to the visitor. You keep, however, everything your customers share with you directly and everything they do on your own channels. A digital marketing strategy built on your own data holds up better against these changes, because it depends on no intermediary.
Three raw materials that belong to you
- Declared data: what the person tells you directly, for example their interests in a form, the size of their company, or how often they want to hear from you.
- Data observed on your channels: pages an identified visitor views, purchases, clicks in your emails, customer service requests. Specialists call this first-party data: you collect it yourself, on your own media.
- Context: the page displayed, the section, the device, the time of day, the browser language. These signals identify no one and already allow a message to be adapted.
Context deserves more attention than it gets. An anonymous visitor who reads three product sheets from the same range tells you their need without giving you their name. You can offer them the buying guide for that range rather than a generic banner.
What AI does better than hand-written rules
Simple rules are enough to start: “if the contact bought this product, offer this accessory”. They quickly reach their limits as the catalogue or database grows. AI models then take over four tasks.
- Segment: group contacts by actual behaviour rather than by age or city.
- Predict: estimate, from history, the probability that a customer buys again, unsubscribes or responds to an offer.
- Recommend: choose the most relevant product, article or service for a given profile.
- Draft variants: propose several versions of an email subject line or a page block, which your team reviews.
The rest is up to you. You decide the positioning, the tone and what the brand must never say, then you validate each scenario before it goes live. AI computes a probability; it does not know whether a message will feel intrusive to your customers.
The five-step method
- Inventory your sources: website, online shop, CRM (customer relationship management software), till, email tool. For each, note what you collect and on what basis (consent, contract, commercial relationship).
- Bring these sources together in a CRM, around a common identifier. The email address or phone number most often plays this role.
- Choose two or three moments to personalise: the homepage of a known customer, the email after purchase, the follow-up on an unanswered quote.
- For each moment, write a default version and a few variants per segment. Entrust the review to someone who knows your customers.
- Measure with a control group, meaning a share of the audience that receives the default version. You will then know what personalisation really brings.
Track few indicators, but track them over time: click rate by segment, conversion rate against the control group, unsubscribe rate, share of contacts who have filled in their preferences. These four measures are enough to decide whether to continue, adjust or stop a scenario.
When personalisation backfires
- Showing that you know too much. A message that quotes a search made the day before is unsettling. Prefer a useful suggestion to proof of tracking.
- Collecting “just in case”. Every field without a planned use weighs on your forms and on your responsibility.
- Forgetting the default version. A visitor who refuses cookies must find a complete and clear site.
- Making unsubscribing painful. A visible link and a simple preference centre protect your reputation.
- Letting a model alone decide a price or an exclusion. Keep human validation on anything that touches fairness between customers.
Consent rules vary according to the country of your contacts. Have your information notices and retention periods validated by a lawyer or by the person responsible for data protection in your company.
What changes by sector
- E-commerce: recommendations rely on the basket and purchase history rather than on retargeting. The email after purchase becomes your first personalised channel.
- Hospitality: preferences declared at booking (purpose of stay, language, arrival time) feed messages sent before and after the stay.
- Professional training: the course already completed and the declared job guide the next sessions offered, without tracking the learner outside your platform.
Putting this approach in place with AISYSNEXT
AISYSNEXT supports SMEs across the whole chain: inventory of data sources, scenarios, content and measurement. Our digital marketing team works with our ERP and CRM teams, which avoids reducing personalisation to a tool setting. The AROVA CRM, integrated with the ERP, brings customer information together in one place.
Want to know where to start with your current data? Request a free quote: we start from your situation, not from a ready-made template.
Frequently asked questions
Can you personalise for a visitor who refuses cookies?
Yes, through context. The page viewed, the section and the device are enough to adapt content without identifying the person. You respect their choice and your site remains useful.
Do you need a large database to use AI?
Not to start. Simple rules and a few segments already allow you to test the approach. Predictive models become relevant once you have sufficient and clean history.
Does personalisation replace targeted advertising?
It complements it. Advertising keeps bringing in new visitors, often through contextual targeting or audiences you build yourself. Personalisation then makes the most of the contacts you already have.
