Personalised marketing.

What counts as personalisation

Personalisation is adjusting the offer, content and service to a specific person based on what you know about them. Putting a name into an email barely counts as personalisation: it is politeness, not working with data. Real personalisation begins where two different people are shown different things - and different for a reason.

The point is to shorten the customer’s path to what they need. A person who has already bought cat food from you should not have to browse the whole catalogue again. A person reading an article about implants for the third time is probably waiting not for a general brochure but for a cost estimate.

The flip side is obvious: to adjust something, you need to know something. And how you collect and use that knowledge determines whether the person perceives personalisation as care or as surveillance.

Digital identity: what knowledge about a customer is made of

Everything that describes a person in the digital environment is usually called their digital identity. It is made up of several layers.

• Personal data - name, age, gender, city, contact details. The simplest layer and the most sensitive from a legal point of view.
• Behavioural data - what a person viewed, what they put in the basket and did not buy, how often they visit, from which device, at what time of day. Usually the most useful layer: it speaks of intent rather than of a profile form.
• Interaction history - purchases, support requests, returns, participation in promotions. What has already happened between you and the customer.
• Social data - public activity in communities, subscriptions, reactions. The layer to be treated with the most caution.

Individually, each layer offers little. Value appears when data from different sources is linked into a single profile and you see not “a visitor from such-and-such IP” and “order number so-and-so” separately, but one person with a history.

How neural networks help with personalisation

A neural network is an algorithm that learns from large volumes of data and finds patterns in it that were not defined in advance. In the task of personalisation it does three things.

It finds non-obvious groups. Classic segmentation divides customers by clear attributes: city, age, order value. An algorithm can discover that there is a stable group united not by age but by the combination “visits from a phone late in the evening and always studies the delivery section”. Nobody would have singled out such a group by hand.

It predicts the next action. Based on the behaviour of similar customers, the model estimates what a person is most likely to do next: buy, leave, come back in a month. This allows you to intervene in time, rather than after the customer is already lost.

It selects recommendations. The most widespread application. Products in an online store, selections in a service, the next piece in a blog. The wider the range, the more noticeable the effect.

In practice it looks like this: a bank understands from transaction history that a savings product is relevant to the customer rather than a loan; an online store shows accompanying products for an order already placed; an online service rearranges its home page around the interests of a specific user.

What personalisation gives a business

First - more precise targeting and less wasted budget. When you show ads only to those the offer suits, money stops going on empty impressions. On how to analyse the effectiveness of your investment, there is an article on calculating ROI.

Second - higher conversion. A relevant offer is simply accepted more often: the person does not have to wade through what does not suit them.

Third - better retention. A customer who does not have to explain their preferences all over again each time is less likely to leave for a competitor. We looked at the motives that keep people with a brand in our article on consumer psychology.

Note that none of these three points contains specific percentages. The uplift depends heavily on the niche, the size of the catalogue and the quality of the data, and any universal figures here would be invented. The effect needs to be measured on your own campaigns by comparison with a control group.

Where the line lies

Personalisation easily turns into intrusiveness. There are several signs that make it clear the line has been crossed.

The person does not understand how you know what you are showing them. If the offer relies on data they did not knowingly give you, the reaction will be negative, even if formally everything is legal.

The offer touches on sensitive topics - health, debts, family circumstances. Here it is better to say too little than to guess exactly right.

The customer has no way to switch it off. The option to unsubscribe, disable recommendations or ask for data to be deleted must be explicit and working.

Plus the legal side: collecting and processing personal data requires consent, and storing it requires compliance with the law. We covered the practical side of protecting a customer database in our article on data protection.

Where a small company should start

Not with neural networks. First - with a single place where customer data is stored, and with the discipline of filling it in. While the purchase history sits in one spreadsheet, support requests in a messenger, and the source of an enquiry is recorded nowhere, there is nothing to personalise.

Next, choose one scenario with a clear benefit. Usually this is an email triggered by an action: a person viewed a product and did not buy, a person bought and after a calculated period needs a refill, a person has not visited for three months. Such scenarios do not require complex models and produce a noticeable result.

Once the first scenario has proved itself, add the next one, and always compare against a control group that received nothing. Without such a comparison you will not be able to tell the effect of personalisation from seasonality.

And only when enough data has accumulated and manual rules stop coping does it make sense to look towards recommendation algorithms.

The technical foundation of personalisation is a website that can recognise a returning visitor and pass data to the records system. Building such a link-up is a development task, and it is worth solving before you start choosing algorithms.