Ecommerce Personalisation: What Works, and What Is Expensive Snake Oil

Ecommerce personalisation: what works, and what is expensive snake oil

Ecommerce personalisation means tailoring content, product suggestions and offers to an individual visitor’s behaviour. Not to assumptions, not to demographic pigeonholes. Two variants in an A/B test: personalised vs. static. A measurable difference in revenue. Everything else is gut feeling with expensive software.

What ecommerce personalisation really means

Most shop owners mean something different by personalisation from what actually brings in revenue. They think of “Hello Max” in an email, or product recommendations everyone sees anyway. That is not personalisation. It is theatre.

Real ecommerce personalisation is based on behavioural data. What did the visitor look at? For how long? What did they put in the cart but not buy? Which category do they keep coming back to? Those signals tell you what to do. A first name or an age does not.

A fashion shop with 80,000 sessions a month showed this in an A/B test: product suggestions based on items actually viewed (not on purchase history) brought +12% revenue per session. The variant with “Hello [first name]” and generic recommendations? Statistically indistinguishable from zero.

The difference lies in the data source. Behavioural data tells you what the customer wants now. Demographic data tells you who the customer is. Who wants to buy matters more than who the customer is.

3 personalisation tactics that measurably increase revenue

1. Dynamic product suggestions based on browsing

The visitor looks at sneakers, then goes to the home page, and sees sneakers again. Not T-shirts. Not accessories. Sneakers. It sounds simple, yet most shops do not do it.

The first test worth running: home page modules that reflect the most recently viewed category instead of a fixed bestseller list. The technology behind it is a simple cookie. No AI, no expensive software.

Worked example: an electronics shop with 45,000 sessions a month swaps its fixed bestseller module for the last viewed category. Clicks on the recommendations rise by 18%, revenue per session by 9%.

The key: the recommendation has to happen within the same session. Someone who looks at shoes and comes back the next day often has a different context. Most tools use 30-day cookies here, which is too long. A 24-hour window is the better starting point; test it against the tool default on your own traffic.

2. Cart recovery with context

The standard “You left something in your cart” email no longer works. Everyone sends it. The email that works says: “The item you looked at is still available in size 42. 3 left in stock.” That is personalisation with urgency.

A sporting goods shop tested this email against the standard version. The contextual version brought 3.2× more revenue from the recovery channel. The difference was not timing: both emails went out after 2 hours. The difference was the information.

Important: the email must name the specific item, not “your selected products”. And it must carry real urgency: stock levels, size availability, a price change. Fake urgency (“Today only!” for an item that has been in the range for months) damages the brand.

3. Prices and offers adjusted by segment

Not every customer should see the same offer. A returning customer who has bought 3 times reacts differently to “10% off” than a first-time buyer. The regular customer is more interested in exclusivity: early access, limited editions.

Two variants are worth comparing: a discount for everyone against an exclusive offer for regular customers, for example early access to a new collection. For regular customers the exclusive offer often beats the discount, because a price cut mainly pushes down the cart value.

Worked example: a home and living shop runs 10% off for everyone against “Early access to the new collection” for regular customers. The exclusive offer brings +22% revenue in that segment. The discount variant brings -5%, because it pushes down the cart value.

The counter-intuitive part: regular customers buy without a discount. They buy out of habit, trust and convenience. A discount is more likely to tell them the product is not worth its price. First-time buyers need a nudge. Regular customers need to feel special.

The most expensive personalisation mistakes

Mistake 1: Personalising without enough data

A shop with 2,000 sessions a month cannot personalise. The segments become so small that no test reaches statistical significance. Below 10,000 sessions a month, personalisation is A/B testing on a shoestring: expensive, slow and without reliable results.

The rule: at least 1,000 sessions per segment and variant for a test running 2 weeks. Anything less is guessing. One shop with 5,000 sessions/month defined two segments: new customers and regular customers. Every test ran 6 weeks instead of 2. After 8 months they had 2 reliable results, and 12 tests that never reached significance.

Mistake 2: Over-personalisation

The more variables you personalise, the harder testing gets. A shop personalising prices, product suggestions, headlines and CTA buttons all at once can no longer tell what caused what. Revenue may rise, but you do not know why, and you cannot repeat it.

Test one variable at a time. Personalisation is not a big-bang project. Start with product suggestions. If that works, add cart recovery. Then checkout. Not everything at once.

A cosmetics shop personalised 12 elements at the same time. After 4 months they did not know which change had affected revenue, positively or negatively. They reset everything and started again, with one variable per test. After 6 months they had 5 measurable winners and a clear ROI.

Mistake 3: The “AI will sort it out” trap

Most personalisation tools sell AI as a cure-all. The reality: AI needs training data. Lots of it. For months. And even then a simple rule-based algorithm is often better, because you can understand it.

If the AI recommends product X and revenue falls, you do not know why. If your rule recommends the “last viewed category” and revenue falls, you know straight away where the problem is.

The comparison to run in your own shop: AI-driven recommendations against a simple rule such as “show products from the last viewed category”. The rule variant is often ahead early on, and it stays explainable. With an AI you frequently cannot say why it recommends a given product, which makes a lost test hard to learn from.

Worked example: a shop with 65,000 sessions a month and 8 months of data history runs exactly that comparison. After 3 weeks the rule variant is 7% ahead on revenue. The AI needs 6 weeks to reach the same level.

What you need before you personalise

Before your first personalisation measure, three things have to be in place:

1. 1
   

   Tracking that records behaviour, not just purchases.

   You need event tracking for product views, add-to-carts, time on page and scroll depth. Without this data, personalisation is flying blind.

2. 2
   

   An A/B testing tool that supports segments.

   Not every tool can test different variants for different user groups. Check this before you buy, not after. Here is a comparison of the leading A/B testing tools.

3. 3
   

   A baseline you can beat.

   Personalisation is relative. You need your current revenue per session, conversion rate and cart abandonment rate. Otherwise you will not know whether personalisation helped.

A furniture shop ignored this order. It bought software, defined rules, and six months later: +2% revenue, not statistically significant. The cost of software and consulting? 45,000 euros. The same budget invested in classic A/B testing would have paid for 8–12 tests with measurable results.

The essentials at a glance

Frequently asked questions about ecommerce personalisation

What is ecommerce personalisation?

Ecommerce personalisation is tailoring shop content, product suggestions and offers to a visitor’s individual behaviour, based on browsing data, purchase history and session signals. Not on demographics or assumptions.

At what traffic level does personalisation pay off?

From about 10,000 sessions a month for simple rule-based personalisation. For AI-driven recommendations, at least 50,000 sessions/month and 6 months of historical data. Below that, classic A/B testing is more efficient.

What does ecommerce personalisation cost?

Rule-based personalisation with an existing A/B testing tool often needs no extra budget. AI tools such as Dynamic Yield, Nosto or Kibo cost 2,000–10,000 euros/month depending on traffic. The bigger cost trap is not the software, it is the time for setup, testing and optimisation.

Does personalisation really increase revenue?

The right kind does. Browse-based product suggestions typically bring +9–18% revenue per session. Contextual recovery emails bring 3× more recovery revenue. “Hello [name]” and generic recommendations bring nothing measurable.

What is the difference between personalisation and A/B testing?

A/B testing compares two variants for all visitors. Personalisation shows different visitors different variants. You need both: A/B testing to learn what works, personalisation to apply it where it counts.

What data do I need for personalisation?

At minimum: product views, add-to-carts, purchases, category preferences, session length. Ideally also: scroll depth, search terms, device type, traffic source. Not needed: age, gender, postcode. Demographic data rarely correlates with buying behaviour.

Is AI personalisation better than rule-based?

Not automatically. AI needs large amounts of data and long training times. Rule-based systems (“if category X was viewed, show Y”) are often quicker to implement, easy to understand and more effective at medium traffic levels. AI only makes sense from 50,000+ sessions/month.

How do I measure the success of personalisation?

Primary: revenue per session. Secondary: conversion rate, average order value, cart abandonment rate. Never: clicks on recommendations, time on site, bounce rate. These vanity metrics say nothing about revenue.

You will find more in-depth material in the Academy track A/B-Testing Mastery ↗.