“Hi Gledis, we picked these just for you.”
There is something almost touching about seeing my own name above a row of products selected by a machine that has probably observed more of my behaviour than several members of my extended family.
Recently I was researching modular sofas in Milan. This was a perfectly ordinary activity. I needed information about sofas. The internet, however, is not particularly good at temporary interests. Spend enough time looking at upholstery and a recommendation system can reasonably conclude that something profound has happened. I am no longer someone who needed to research a sofa. I have entered my Sofa Era.
This is technically personalization.
The system has identified me, recorded what I looked at, connected it to everything else it knows and calculated what I might want next.
What it has not necessarily understood is why I was looking in the first place.
That distinction matters.
Personalization has become very good at recognition. What people actually want is relevance.
They are not the same thing.
My name is not a relationship
Somewhere along the way, personalization became mail merge with a data warehouse.
Put my first name in the subject line. Remember the shoes I clicked. Show me “recommended for you.” Rearrange the homepage because I once bought something vaguely similar. Congratulations: the experience is now personalized.
Except my name is probably the least interesting thing a company knows about me. I already know it.
“Hi Gledis” was impressive when the person behind the counter actually remembered Gledis. When a database inserts it using {{first_name}}, the emotional achievement is somewhat smaller.
Recognition is useful, of course. I would rather my bank know who I am than greet me every morning like an unexplained visitor. The problem begins when recognition is mistaken for understanding.
McKinsey found that 71% of consumers expected personalized interactions from companies, while 76% became frustrated when those interactions were not relevant. Those numbers are usually presented as evidence that consumers want more personalization.
I think they suggest something more demanding.
People do not necessarily want companies to prove how much they know about them. They want companies to use what they know intelligently enough not to waste their attention.
That is a much harder brief.
A click is an event, not a confession
Behavioural data records what happened remarkably well and why it happened remarkably badly.
I clicked the sofa.
Maybe I wanted to buy it. Maybe I was comparing prices. Maybe I was researching something for work. Maybe I hated it and sent it to somebody with the message, “Who would put this in their house?”
The click remains identical.
The meaning does not.
That gap between behaviour and intent is where a surprising amount of bad personalization begins.
Recommendation systems increasingly incorporate context such as time, location, social circumstances and changing patterns of behaviour because historical activity alone is an incomplete explanation of what somebody wants now.
Human beings, unfortunately for databases, are full of temporary interests.
Watch two videos about watches and the internet suspects you have developed a horological calling. Research hotels in one city and apparently you intend to relocate there. Click an unusual lamp at 11:47 p.m. and somewhere a server quietly updates its understanding of your aesthetic philosophy.
The machine is not stupid.
It is literal.
And literal interpretation is dangerous when behaviour is ambiguous.
The algorithm remembers every version of me
People change their minds constantly.
We buy gifts for people who are nothing like us. We research places we will never visit. We become fascinated by something on Tuesday and have completely forgotten about it by Friday. Sometimes we click things because we are bored.
This is terrible behaviour from a clean-data perspective.
Recommendation systems need continuity because prediction depends on patterns. But when every past action keeps contributing to the profile, temporary curiosity starts hardening into identity.
Past interest becomes future targeting.
Past purchase becomes assumed preference.
Past curiosity becomes a segment.
The algorithm remembers every version of you and appears to have terrible separation anxiety.
A person can notice that your interests have changed. A database needs a mechanism for deciding that yesterday has stopped being useful.
That may be one of the least discussed skills in personalization: forgetting.
If I bought the sofa, perhaps the sofa problem has been solved.
If I booked the hotel, I probably need directions or check-in information more than another advertisement asking whether I have considered booking it.
If I searched for something once six months ago, perhaps we can all agree to move forward with our lives.
The smartest profile is not necessarily the one that remembers the most. It may be the one that knows what to discard.
Relevance is a judgment problem
Recognition asks: Who is this?
Relevance asks: What would be useful to this person in this particular situation?
The first can largely be solved with data.
The second requires judgment.
It means distinguishing persistent preferences from temporary intent. It means understanding that a purchase can signal affinity or simply completion. It means allowing some signals to grow stronger through repetition while letting others decay.
And sometimes it means doing nothing.
That last part is uncomfortable because much of digital marketing is built around action. Send something. Recommend something. Retarget something. Generate another message because the customer has produced another signal.
But the best personalized experience is often invisible.
You barely notice that the irrelevant options disappeared. The system simply made a good decision and then had the discipline not to announce it.
That feels considerably more personal than receiving an email whose primary evidence of intimacy is knowing my first name.
There is also such a thing as knowing me too well
There is another problem.
Personalization can feel impersonal because it misunderstands us. It can also become uncomfortable when it understands us with suspicious enthusiasm.
Research into personalized advertising repeatedly finds this tension. Greater relevance can improve people's attitudes toward advertising, but awareness of how personal information was collected can also increase feelings of vulnerability.
Humans already understand this boundary offline.
A hotel concierge remembering that I prefer a quiet table feels attentive.
The same concierge saying, “We noticed you walked past the restaurant three times last night but did not enter” feels like the beginning of a police procedural.
Same information.
Different emotional experience.
This is why maximum personalization is probably the wrong objective.
A better principle might be minimum necessary intimacy: know enough to improve the experience, but not so aggressively that I become preoccupied with how you know it.
The best personalization feels like service.
Bad personalization feels like evidence.
Context beats memory
AI will make all of this considerably more sophisticated.
Systems can combine more signals, infer patterns faster and generate individualized communication at enormous scale. A brand can theoretically produce a different message for every customer, every moment and every behavioural cue.
But scale does not solve the conceptual problem.
If the instruction remains “use everything we know about Gledis to sell Gledis something,” AI will simply become much better at executing a slightly stupid instruction.
The more interesting question is:
What matters to Gledis right now?
That shifts the goal from building the biggest possible customer memory to interpreting the present moment accurately.
Because people do not experience themselves as customer profiles. We experience ourselves as situations.
I am the same person when I am working, travelling, buying a present, choosing a restaurant, researching a sofa or aimlessly browsing online at midnight. But I am not the same customer in each of those moments.
The difference is context.
And context is where personalization stops being a database function and starts resembling actual understanding.
Perhaps personalization needs better manners
None of this makes personalization useless. Done well, it reduces friction, improves discovery and makes enormous digital environments easier to navigate.
But better technology raises the standard.
When a company knows almost nothing about me, irrelevance is understandable. When it knows my identity, history, purchases, behaviour, device, location and preferences, irrelevance starts to feel less like an information problem and more like poor listening.
That is why “Hi Gledis” no longer impresses me.
The interesting part is what happens next.
The next generation of personalization may not be defined by systems that know more about us. We already have plenty of those.
It may be defined by systems that know which information matters now, which information has expired and which information they have the manners not to mention.
Until then, “Hi Gledis” will remain what it increasingly feels like:
a stranger reading my name badge.