Personalisation: it will get worse before it gets better.

Kind. Personal. Proactive.

Except I had not checked in yet.

A good intention. A wrong assumption.

I described that email in 2016. I liked the idea: a hotel checking whether a guest needed anything, rather than waiting for a complaint.

But the timing made it clear that the message did not reflect my actual situation.

I did not know what had triggered it. Perhaps a scheduled check-in time rather than an actual arrival. Whatever the mechanism, the system treated me as a guest already staying in the hotel when I was not.

The message tried to say “we are paying attention”. Instead, it revealed what the hotel did not know.

Personal is a bigger promise.

A general message makes a modest claim. A personalised message implies something more: we know enough about you to make this relevant.

When that knowledge is wrong, the mistake becomes the thing you notice.

In the original post, I listed other examples:

• YouTube recommending Dora after my daughter had used my iPad.
• Retargeting for a dress intended as a surprise gift appearing when the recipient used the same computer.
• A tourist guide to Amsterdam sent to me when I was Dutch and regularly stayed in the city for work.
• A retailer addressing me as “Misses De Maagt”.

Different mistakes, but a similar gap: activity is not identity. A booking is not an arrival. A destination is not a reason for travelling.

Knowing something about a person is not the same as understanding their situation.

“95% right” can still feel completely wrong.

That was the shorthand I used in 2016, not a measured accuracy score.

The point was that getting most of the details right does not rescue the one assumption that makes the whole interaction irrelevant.

Right hotel, right guest, wrong moment. Right city, wrong reason for being there.

The customer experiences the message as a whole. They do not award partial credit for the fields you populated correctly.

Learning for the company. Friction for the customer.

My original argument was that personalisation would get worse before it got better. Companies would experiment, make mistakes and learn their way towards more useful experiences.

But a company's learning process is also somebody else's customer experience.

And customers are not dealing with just one company learning how to personalise. They encounter the experiments of hotels, retailers, banks and platforms alongside one another.

A small mistake in each company's dashboard can add up to a lot of irrelevant attention in one person's day.

That does not mean mistakes are an inevitable price customers must pay. Nor does collecting more data automatically solve them.

Sometimes the better choice is a smaller claim. A question instead of an assumption. A general message instead of a falsely personal one. Or no message until the situation is clear enough.

Before making it personal.

Ask what you actually know, what you have inferred and which assumption would make the message fail if it were wrong.

Then ask whether personalisation adds something useful. If the same message would help without pretending to know the customer's situation, use that version.

The ambition should not be to sound as if you know me. It should be to help me without getting in the way.

First published 4 July 2016; rewritten in October 2026 from my original post. The title keeps the original prediction. “95% right” is rhetorical, not a research finding. Related: How Amazon Echo made me a worse person and 95% right. 100% wrong.