AI Makes E-commerce Personalisation Easier, but When Does It Go Too Far?

E-commerce Does Not Have a Data Shortage

Online retailers already collect plenty of signals. Customers search, click, compare products, abandon baskets, open emails and make purchases. Loyalty programmes and customer accounts add another layer. AI can connect these signals and use them to decide which product, message or offer someone should see next.

That can make shopping genuinely easier. A retailer remembering a preferred size is useful. So is bringing a customer back to an unfinished basket or making it easier to reorder something they buy regularly.

But the same technology can produce the opposite effect.

Look at a product once, and it follows you around the internet. Buy a present for a child, and children’s products dominate your recommendations for weeks. Search for something unusual, and suddenly the homepage seems convinced it has discovered a new side of your personality. In this cas, the technology worked. The interpretation did not.

A Click Is Not a Customer Profile

This is where AI-driven personalisation introduces an important problem. E-commerce data records behaviour, but behaviour often lacks context.

Someone buying a premium handbag may be purchasing a birthday present. A customer suddenly browsing baby products could be shopping for a friend. One household account might be used by several people.

An algorithm sees signals. It does not necessarily know the story behind them.

That makes the difference between what a retailer knows and what it assumes increasingly important.

Signal	Helpful Use	Too Much
Repeat purchase	Make reordering easier	Assume unrelated preferences
Abandoned basket	Offer a simple reminder	Chase the shopper across every channel
Product browsing	Improve on-site recommendations	Build a profile from one visit
Stated preference	Show more relevant products	Keep using it after behaviour changes
AI prediction	Help rank likely relevant products	Treat the prediction as a customer fact

Personalisation becomes risky when an assumption starts being treated with the same confidence as something the customer actually did or told the retailer.

Not Every Signal Needs a Campaign

Marketing technology makes it tempting to act simply because we can.

A customer viewed a category? Create a segment. They left a basket? Trigger a journey. They clicked twice? Add another recommendation.

But not every customer action needs a marketing response. Sometimes a product view should remain a product view.

This is particularly relevant as AI reduces the cost of creating increasingly granular audiences and individualised experiences. The technical barrier is disappearing, which makes human judgement more important.

Before activating another data point, there is a simpler question worth asking: What does this actually improve for the shopper?

If it helps someone find a product faster, removes irrelevant choices or saves them from starting again, there is an obvious purpose.

If the main answer is simply “it gives us another opportunity to convert them”, the customer benefit is much less clear.

Sometimes the Smarter Experience Is the Simpler One

There is a tendency to treat increasingly sophisticated personalisation as progress in itself. It is not.

A good bestseller section may work better than a complicated recommendation engine for a first-time visitor. A single basket reminder may be more useful than appearing in someone’s inbox, social feed and display ads at the same time.

Retailers can also stop trying to infer everything.

Letting customers choose categories they care about, change communication frequency or dismiss irrelevant recommendations gives them some control over the experience. It also provides something algorithms often struggle with: context.

The objective should not be to personalise every available touchpoint.

It should be to personalise the ones where doing so genuinely helps.

Restraint Could Become Part of Good Personalisation

AI will give e-commerce businesses more customer signals, more predictions and more opportunities to act on them.

That does not mean all of them deserve to reach the customer.

The strongest personalisation often feels almost invisible. The right products are easier to find. Irrelevant options disappear. Returning to a purchase is straightforward. The customer does not need to think about why any of it happened.

That may become an increasingly useful benchmark.

Customers do not visit an online shop to see how sophisticated its data stack is. If personalisation makes shopping easier, it is doing its job.