When personalisation looks smarter than it really is
We’ve never known more about our customers.
We know what they bought, what they browsed, what they clicked, what they abandoned, how often they purchase, which categories they prefer, whether they opened the app yesterday, visited the website this morning, or ignored our last six emails.
And now we have AI layered over all that data, helping us analyse behaviours, identify patterns, predict what somebody might do next, and generate the message most likely to persuade them.
It sounds like the perfect recipe for relevance.
But at RARE Conference, where I first presented The Illusion of Relevance, I posed a slightly uncomfortable question: Is AI helping us build better customer relationships, or simply helping us automate the appearance of one?
Because knowing what somebody did is not the same as understanding why they did it.
And when we forget that distinction, personalisation can look remarkably intelligent from the marketer’s side while feeling completely irrelevant to the customer.
Worse still, it can lead us to spend more money, build more automation, and optimise more campaigns around an assumption that was wrong in the first place.
That’s the illusion of relevance.
Behaviour shows the what. Intent explains the why.
We have more behavioural data than we have ever had before. The problem is that behaviour still needs interpretation.
Take the abandoned basket, one of ecommerce’s favourite behavioural signals.
Someone puts a product into their basket and leaves without purchasing. We observe that behaviour and immediately infer intent: they wanted the product, but something stopped them buying it.
So, we send the reminder. Then perhaps another reminder. Maybe we introduce urgency. Eventually, we might offer a discount.
Perfectly logical.
Except the basket could have stalled because of price, delivery timing, unexpected shipping costs, uncertainty, distraction, or comparison shopping.
They may have put the item there simply to calculate the final cost. They may be waiting for payday. They may already have bought it elsewhere.
Or, as I suggested at RARE, perhaps their child started drawing on the wall with yoghurt.
Same behaviour. Entirely different context.
And that matters because the right marketing response depends on the reason behind the behaviour.
If shipping costs created the friction, a discount on the product may not solve the problem. If the customer became uncertain about the purchase, urgency could make matters worse. If they were simply interrupted, they may need nothing more than a useful reminder.
The data tells us what happened.
It doesn’t necessarily tell us why.
And when we build automation around the wrong interpretation, we don’t just risk sending an irrelevant email. We risk institutionalising that wrong assumption across thousands of customers.
We do this throughout the customer journey
Abandoned baskets make the problem easy to see, but marketers make these assumptions everywhere.
A browse is interpreted as product interest. But somebody browsing a category could be researching, comparing, daydreaming, buying a present, or simply curious.
A lapsed customer is interpreted as disengaged or disloyal. But perhaps they moved house, changed jobs, had a baby, stopped needing the product, or simply haven’t needed to buy again yet.
A redeemed discount is treated as evidence that the promotion drove the conversion. But sometimes the customer had already decided to buy and happily took the discount we put in front of them.
These aren’t arguments for ignoring behavioural data. Behavioural signals are enormously valuable. They’re arguments for recognising their limitations.
Data records behaviour. Marketers still have to interpret motivation. And AI doesn’t remove that responsibility.
AI makes the inference faster. That doesn’t make it right.
AI can identify patterns across millions of customer interactions that no human could possibly analyse manually. It can help with timing, segmentation, recommendations, journey logic, content, and next-best-action decisions.
That capability is extraordinarily useful. But look at what the machine usually has to work with.
This customer clicked. This customer purchased. This customer browsed. This customer stopped engaging. This customer returned. People who behaved like this customer previously tended to do X next.
From those signals, AI can make increasingly sophisticated predictions. What it cannot necessarily see is the context surrounding them.
And because the output looks intelligent — precise segments, confident recommendations, personalised content — it becomes very easy for us to mistake prediction for understanding.
That creates a commercial risk as well as a customer-experience one. If the underlying assumption is wrong, AI can simply help us scale the wrong strategy faster.
We can spend more media budget retargeting the wrong person, increase message frequency when the issue was uncertainty, offer discounts unnecessarily, or move customers into journeys that don’t reflect what they actually need.
Efficiency is useful. Efficiently doing the wrong thing is not.
Personalisation can recognise you without understanding you
For years, we’ve tended to describe personalisation as though it sits on a straight line: the more customer data we have, the more personalised the experience becomes, and therefore the more relevant it becomes.
But those things aren’t synonymous.
A brand can recognise me perfectly and still misunderstand me completely. It can know I recently bought something and decide I must therefore want another one. It can notice I’ve browsed a category several times and decide I’m deeply interested in it. It can see I clicked three emails about a particular topic and conclude that I should receive more of the same.
Every one of those conclusions might be statistically sensible. None necessarily explains what was happening inside my head.
And this is an important distinction. Recognition is not understanding.
The question therefore shouldn’t simply be, “Can we personalise this?” It should be, “Will this personalisation improve the customer’s experience and the commercial outcome?”
Because those two things should not be in conflict.
Does the personalisation do anything useful?
Sometimes what we call personalisation doesn’t actually help the customer at all. It simply demonstrates how much data we have collected about them.
So, before adding another personalised element, recommendation, AI-generated message, or journey branch, ask what it actually does.
Does it reduce friction? Does it improve timing? Does it make the decision easier? Does it increase confidence? Does it remove an objection? Or does it merely prove that the brand has data?
That last question is worth spending some time with.
Showing somebody the product they looked at yesterday may be personalised. Helping them resolve the uncertainty that prevented them buying it may be relevant.
There’s an important difference. And the latter is far more likely to produce a meaningful commercial result.
The brief sets the ceiling
The temptation with AI is to begin with execution.
“Write a reactivation email.”
“Create five subject lines.”
“Build a lifecycle sequence.”
“Give me some personalisation ideas.”
AI can do all of those things, quickly and competently.
But there are bigger questions sitting behind each task.
Why did those customers go quiet? Which emotional trigger matters here? How does this customer actually make the decision? Which personalisation ideas are worth doing?
Ask AI to write an email promoting 20% off and it will happily produce something tidy, coherent, and reasonably persuasive.
What it doesn’t know is whether 20% off is the right strategy in the first place.
Perhaps the customer needs reassurance rather than urgency. Perhaps the problem is delivery. Perhaps discounting damages perceived value. Perhaps the campaign is solving entirely the wrong problem.
A vague prompt produces vague marketing. A generic brief produces generic content.
And a shallow view of the customer produces shallow messaging, however polished the execution becomes.
Persuasion needs understanding too
The same applies when we start asking AI to help us make marketing more persuasive.
Urgency, scarcity, social proof, authority, reassurance, AI can apply all of them. But knowing the techniques isn’t the skill. Knowing which one fits this customer, this decision, and this moment is the skill.
Urgency can motivate somebody who is procrastinating. Apply it to somebody who is uncertain, and it may simply create pressure.
Scarcity can make an opportunity feel more valuable. Manufactured scarcity can damage trust.
Social proof can provide reassurance. Sometimes what the customer really needs is straightforward information.
Behavioural science isn’t a collection of cognitive biases we sprinkle over a campaign like seasoning. It requires an understanding of how people make decisions.
AI can support that thinking. It shouldn’t replace it.
Sometimes the most relevant response is restraint
There’s another problem with marketing technology that AI could amplify: our assumption that every signal requires an action. Something happens, so something must be sent.
A browse triggers an email. An abandoned basket triggers a sequence. A period of inactivity triggers reactivation. A purchase triggers recommendations.
Automation is very good at action. Judgement sometimes tells us to do nothing.
Perhaps the signal is too weak. Perhaps we need another behaviour before changing the journey. Perhaps the customer should be allowed to re-enter when their circumstances change. Perhaps they don’t need another message from us today.
Some of the most intelligent personalisation may therefore be completely invisible to the customer.
Not sending an unnecessary email.
Not offering a discount when one isn’t required.
Not treating a single click as a declaration of intent.
Not forcing somebody through a journey simply because the automation says they have reached message three.
That may not make for an impressive personalisation demo. But it can reduce waste, protect margin, and create a much better customer experience.
The advantage is moving upstream
For years, email and CRM teams have been rewarded for output: how many campaigns were produced, how many automations were launched, how quickly they went live, and how much revenue was attributed to them.
AI is making that output easier. Which means output itself becomes less of a differentiator.
The real advantage moves upstream: understanding the customer, interpreting the data, defining the actual commercial problem, identifying the behavioural barrier, choosing the right message, protecting the brand voice, designing meaningful tests, and deciding whether what has been produced is actually good enough to send.
This is why I believe AI increases rather than diminishes the value of judgement.The more capable the tool becomes, the easier it is to produce something that looks finished. And the more something looks finished, the easier it is to stop questioning the thinking underneath it.
That is the real danger. Not terrible AI-generated marketing, but average AI-generated marketing that looks perfectly acceptable.
Multiply that across thousands of brands using similar tools, asking similar questions, and accepting similar outputs, and we don’t necessarily create better customer relationships.
We create a sea of competent sameness.
Five questions before you hand an interaction to AI
None of this is an argument against AI or personalisation. I use AI every day. The productivity gains are real, and its ability to support marketers will continue to improve. But the next phase has to move beyond asking how much work AI can do for us.
Before handing a customer interaction to AI, ask:
-
Is this useful to the customer, or simply efficient for us?
-
Does it improve the decision experience?
-
Does it reflect intent, or only behaviour?
-
Does it protect trust, tone, and brand voice?
-
Would we be happy if the customer knew how this interaction was created?
Those questions won’t stop us using AI. They should help us use it better.
Because the opportunity isn’t to use AI to produce more personalisation, more journeys, and more messages faster. It is to use AI to support better thinking.
To challenge assumptions. Explore alternative explanations. Find gaps in our journeys. Develop stronger hypotheses. Design better tests. And, ultimately, create experiences that are more deserving of the customer’s attention.
We have more data about our customers than ever before. We have more technology with which to act on it. But knowing more things about someone doesn’t automatically mean we understand them better.
AI can help us make increasingly intelligent guesses. Our job is to remember they are still guesses.
Otherwise, all that sophisticated technology may simply make the illusion of relevance more convincing, and more expensive.
