Personalisation has become one of those marketing practices that few people question anymore. We collect customer data, analyse behaviour, identify interests, and use what we discover to make our communications more relevant. The more we know, the better we can personalise. And better personalisation should mean better customer experiences, shouldn’t it?
Not necessarily.
There’s an important distinction that often gets overlooked in discussions about personalisation: the difference between what a customer has chosen to tell us and what we’ve worked out about them.
Because while both might give us valuable information, they don’t necessarily give us the same right to use it.
And understanding that difference could be one of the most important decisions marketers make about their personalisation strategies.
The Case Study That Still Makes Me Think
Years ago, when I was teaching personalisation, data mining, and customer behaviour, I used to discuss the now-famous case of US retailer Target.
In 2012, Charles Duhigg wrote a fascinating article in The New York Times Magazine about how Target had developed a predictive model capable of identifying customers who were likely to be pregnant, based on changes in their shopping behaviour.
The story became famous because of an incident involving a father who complained after his teenage daughter received pregnancy-related offers from Target. He was furious that the retailer appeared to be encouraging his daughter to become pregnant. According to the account, he subsequently discovered that she was already pregnant.
Target’s marketing had apparently revealed something about his daughter that he didn’t know himself.
The sophistication of the analysis was impressive. But it was the customer’s experience of the resulting marketing that made the story so memorable.
Target reportedly went on to make its baby-related promotions less conspicuous by mixing them with offers for unrelated products.
In other words, the challenge wasn’t simply identifying the right customer. It was communicating in a way that didn’t make that customer uncomfortable.
More than a decade later, I think there’s still a lesson here that marketers haven’t fully absorbed.
Knowing something about a customer doesn’t automatically mean they want you to demonstrate that knowledge.
Declared Data and Inferred Data Aren’t Psychologically Equivalent
Imagine two customers.
The first signs up to a retailer’s baby club, enters her expected due date, and asks to receive advice and offers throughout her pregnancy.
The second makes a series of purchases that lead the retailer to conclude she’s probably pregnant.
From a marketing perspective, both customers might appear suitable for similar communications. They may even have almost identical needs.
But psychologically, the circumstances are very different.
The first customer has voluntarily shared information for a particular purpose. She understands why the retailer knows she’s pregnant and what sort of communications she can expect.
The second hasn’t necessarily shared that information at all. Her purchases have simply allowed the retailer to make an inference about her personal circumstances.
And that inference might be correct. It might also be wrong.
Even when it’s correct, she may have no expectation that the retailer is analysing her behaviour in this way, let alone using its conclusions to communicate with her.
This is where relevance can become intrusive.
The issue isn’t necessarily the product being promoted. It’s the unexpected revelation of how much the brand appears to know.
Why Personalisation Can Trigger Discomfort Instead of Appreciation
Behavioural science helps explain why two apparently identical offers can produce very different reactions.
One important factor is our sense of autonomy. We like to feel that we have control over our decisions, including what information we disclose about ourselves and when.
When a brand uses something we’ve deliberately shared, its communication can feel like a useful response to our needs.
When it acts on something we’ve never disclosed, it can create an uncomfortable sense that our privacy or control has been compromised.
There’s also the issue of expectations. We develop mental models of how relationships work, including our relationships with brands. If I buy a foundation from a beauty retailer, I might reasonably expect recommendations for complementary products or a replenishment reminder.
What I might not expect is for that retailer to draw conclusions about my age, personal circumstances, health, or emotional state and start communicating on that basis.
The brand might believe it’s demonstrating how well it understands me. I might wonder why it’s making assumptions about my life. And once that discomfort occurs, the experience can change how I feel about the brand itself.
Personalisation that was intended to strengthen a customer relationship can end up undermining trust.
The Problem With Getting It Right
We often discuss inaccurate personalisation as though it’s the main problem. Someone receives an irrelevant recommendation, a poorly timed reminder, or an offer based on an incorrect assumption.
Those mistakes can certainly be frustrating.
But there’s another possibility that deserves more attention: personalisation can be entirely accurate and still be inappropriate.
Think about the different things a brand might infer from customer behaviour.
A sudden change in purchasing habits might suggest financial difficulties. Browsing activity might indicate a medical concern. Changes in household spending could suggest a relationship breakdown. Repeated visits to certain product categories might indicate pregnancy, bereavement, or another major life event.
In some circumstances, these insights could help a business provide a genuinely better experience.
But there are also circumstances where explicitly acknowledging them could feel intrusive, upsetting, or even harmful.
Consider a customer who has been browsing products associated with pregnancy. A retailer might see an opportunity to send a timely promotional email. But it doesn’t necessarily know whether that customer is pregnant, trying to conceive, shopping for someone else, or dealing with a pregnancy loss.
That’s a lot of uncertainty to build a personalised campaign around.
And even if the inference is accurate, the customer may not want that subject appearing in their inbox, particularly if other people have access to their email or devices.
The mistake is assuming that predictive accuracy automatically creates permission to act.
It doesn’t.
When Helpful Becomes Uncomfortably Personal
This doesn’t mean behavioural personalisation is inherently bad.
Far from it. Some of the most effective customer experiences are built around anticipating needs rather than waiting for customers to articulate them.
A retailer recognising that someone regularly purchases a consumable product and offering a convenient replenishment reminder can be helpful.
An email programme identifying that a customer tends to browse certain categories and making those products easier to find can reduce effort.
A brand recognising that a customer hasn’t purchased for some time and offering an easy way to update their preferences can improve the relationship.
The distinction is in the nature of the inference, the sensitivity of the information, and how the personalisation is experienced.
There’s a difference between saying, “Here are some more products you might like,” and effectively saying, “We think we know what’s happening in your personal life.”
The former helps the customer make a decision. The latter risks making the customer feel observed.
And sometimes the best personalisation is almost invisible. The journey becomes easier, the content more useful, and the choices more appropriate, without the brand feeling the need to announce everything it has inferred.
This isn’t about hiding invasive profiling behind less obvious messaging. It’s about questioning whether the inference is appropriate to make or use in the first place, and whether a less intrusive approach could deliver the same benefit.
There’s Also a Data Protection Question
This distinction isn’t just psychological.
Under UK GDPR, inferred information can fall within the rules governing special category data when organisations deliberately infer or act upon characteristics such as health status, religious beliefs, or sexual orientation.
The Information Commissioner’s Office makes it clear that these rules can apply even when the inference isn’t accurate.
That matters because marketers sometimes regard inferred information as less sensitive than information explicitly provided by a customer. In reality, the opposite may be true from the customer’s perspective, and the law can impose additional requirements on how it is processed.
But regulatory compliance should be the starting point, not the complete test of whether personalisation is appropriate.
Something can be technically permissible and still damage the relationship you’re trying to build.
Five Questions Worth Asking Before Acting on an Inference
Before using inferred customer information to trigger or personalise a communication, I’d encourage marketers to consider five questions.
1. Did the customer tell us this, or did we work it out?
This helps establish whether the customer is likely to understand why they’re receiving the communication. It also encourages us to distinguish expressed preferences from assumptions.
2. How confident are we, and what happens if we’re wrong?
Not every behavioural pattern means what we think it means. The consequences of an inaccurate inference about a preferred clothing style are very different from those of an inaccurate inference about a sensitive life event.
3. Would the customer reasonably expect us to use this information?
A useful test is to imagine explaining the personalisation openly to the customer. Would they regard it as helpful and reasonable, or would the explanation itself make them uncomfortable?
4. Does using this insight genuinely benefit the customer?
Not every opportunity to improve targeting improves the experience. Sometimes the benefit sits almost entirely with the business. That doesn’t automatically make the activity inappropriate, but it should prompt closer scrutiny.
5. Could we achieve the same outcome without making the inference personal?
This is often the most overlooked question. Could we offer useful content without referencing a sensitive circumstance? Could we let the customer select their interests? Could we provide a choice rather than assuming we already know the answer?
And, of course, before any of these questions, we need to establish whether we’re legally entitled to process and use the information in the intended way.
These aren’t arguments against personalisation. They’re arguments for doing it more thoughtfully.
The Best Personalisation Respects What the Customer Hasn’t Said
I’ve always advocated for helpful email marketing. The objective isn’t simply to use more data, build more segments, or make every message feel individually tailored.
It’s to make the customer’s experience better.
Sometimes that means recognising behaviour and responding to it. Sometimes it means asking customers what they want rather than assuming. And sometimes it means deliberately choosing not to use information, even when we believe we’ve correctly interpreted it.
The Target case remains memorable because it demonstrated just how much a business could discover from ordinary purchasing behaviour. But its more enduring lesson is about what happens when that knowledge enters the customer relationship.
We spend a great deal of time asking how we can make personalisation more relevant.
Perhaps we should spend more time asking whether the customer would welcome that relevance in the first place.
Because good personalisation isn’t about proving how much you know about your customers. It’s about using what you know to make their lives easier, without making them wish you knew less.
