Raise your conversion rates with customer behaviour analysis

Customer behaviour analysis: what the data really tells you

Every ecommerce shop produces data. Page views, clicks, scroll depth, cart events, purchases, drop-offs. Most shop owners collect this data. Some analyse it. Almost nobody interprets it correctly.

Interpretation is the central step. Data does not tell you what to do. It only shows patterns. From those patterns you derive hypotheses. The hypotheses have to be validated in A/B tests. Without this loop, customer behaviour analysis keeps people busy and does nothing for the business.

This article shows the methods that work, what they typically bring to light, and how to turn the finding into a test.

Method 1: The funnel scan with segment depth

A classic funnel analysis shows: 100% visitors → 40% product page → 12% cart → 4% purchase. Useful, but shallow. The important insights sit in the segments.

An electronics shop had an overall conversion rate of 2.1 per cent. The funnel looked solid. But the segment analysis showed:

Desktop users: 3.8 per cent conversion.
Mobile users: 1.2 per cent conversion.
Tablet users: 2.9 per cent conversion.

The mobile gap was the problem, not the funnel itself. Without segment analysis, the funnel would have been rated “acceptable” and nothing would have changed. With it, the biggest lever is visible instead of assumed.

Another example: a fashion shop with 1.8 per cent overall conversion. The segment analysis by traffic source:

Organic: 2.4 per cent.
Paid (Google Ads): 1.1 per cent.
Social (Instagram): 0.7 per cent.
Direct: 3.2 per cent.

Google Ads converted at half the organic rate. The analysis showed why: the ads landed on the home page instead of category pages.

A user searched for “women’s red winter jacket”, clicked the ad and landed on the home page with a generic hero banner. They needed three clicks to reach the red winter jackets. The bounce rate was 68 per cent.

We pointed the ads at the matching category pages. Conversion rose from 1.1 to 2.3 per cent. ROAS doubled. The mistake was not the ad copy or the keyword bidding. It was the landing page structure.

Method 2: The heatmap lie and how to correct it

Heatmaps are often presented as proof of user behaviour. “Look, users don’t click here.” But heatmaps only show where people clicked. They don’t show why. And they don’t show what the user would have done if an element had been placed differently.

A supplement shop had a heatmap of its PDP showing that 23 per cent of clicks went to the product image (people expected a zoom), but there was no zoom. 18 per cent went to the rating stars (people expected details), but they were not clickable. 15 per cent went to the shipping costs link in the footer.

The naive reading: “We need an image zoom, clickable ratings and more prominent shipping costs.” The right reading: “Users are looking for information they can’t find.” The zoom clicks were a symptom of a bigger problem: the PDP showed only one image.

Users wanted to see more. The fix is a gallery carousel with the five views customers actually ask for: product, detail, in use, size comparison and packaging. Conversion rises by 12 per cent.

The rating stars were not clickable because the reviews sat at the bottom of the page. Making the stars clickable and scrolling to the review section added 28 seconds of average time on page, but conversion did not move. The explanation: the reviews themselves were poor, 3.2 out of 5 stars from only 12 entries.

Making bad reviews easier to find does not help. The honest fix is to take the stars off the page until there are 50 or more reviews at 4 stars or better. Once they are there, the same change lifts conversion by 8 per cent.

The clicks on shipping costs in the footer signalled a lack of transparency. The fix is a “Shipping & returns” accordion directly below the price: it answers the question where the question is asked. Conversion rises by 6 per cent and the clicks on the footer link fall to 3 per cent.

The point: heatmaps are a diagnostic tool, not a solution tool. They show symptoms. You only find the cause through hypotheses and tests.

Method 3: Session recording analysis

Session recordings are the most powerful and the most dangerous analysis tool. Powerful because they show real user behaviour. Dangerous because they invite subjective readings. “Look, this user hesitated here”, but maybe they were just reading a message or taking a call.

We use session recordings systematically: we filter for sessions that drop out on specific pages, sort by session length (long sessions = interest, but an obstacle), and look for patterns in groups of 50+ sessions.

A furniture shop had a high abandonment rate on its checkout page. The recordings showed a pattern: users opened the address form, started typing, stopped, scrolled to the payment options, scrolled back, typed on, stopped again. The pattern repeated in 40 per cent of the abandoned sessions.

The hypothesis: users wanted to know which payment options were available before entering their address. They feared finding out at the end of the form that their preferred payment method was missing.

The variant to test shows the payment options above the address form, as visual icons rather than selectable options. It answers the question before it becomes a reason to leave: in the example above, the abandonment rate falls by 19 per cent.

Another pattern: in a beauty shop, 60 per cent of mobile users scrolled all the way down the PDP, past reviews, similar products and FAQs, and then went to click the “Add to cart” button.

The button was at the top of the page. Users scrolled back up to find it. The fix: a sticky button at the bottom of the screen. The add-to-cart rate rose by 31 per cent.

Method 4: Cohort analysis

Cohort analysis groups users by shared traits and follows their behaviour over time. The most important cohort split in ecommerce: first-time buyers vs. repeat buyers.

A supplement shop analysed its customer cohorts. First-time buyers had a conversion rate of 1.9 per cent. Repeat buyers had a conversion rate of 8.4 per cent. Repeat buyers’ average basket was 34 per cent higher.

The insight: the biggest lever was not the first conversion but turning first-time buyers into repeat buyers. Worked example: a shop looks at which products most often lead to a second purchase and finds that three products account for 67 per cent of all repeat purchases.

Those three then carry the first-visit strategy: prominent placement, a welcome offer and a follow-up email series after purchase.

The repeat purchase rate rose from 23 to 31 per cent within six months. Customer lifetime value rose by 28 per cent. The revenue gain: about 45,000 euros per month, not from new customers but from keeping existing ones.

Conclusion: data without interpretation is just numbers

Customer behaviour analysis is not a technical problem. It is a thinking problem. If you believe data speaks for itself, you will draw the wrong conclusions. If you understand that data shows patterns that need interpreting, you gain an edge over your competitors.

The four methods in this article (funnel scan with segment depth, heatmap interpretation, session recording analysis and cohort analysis) have delivered measurable results in real shop projects. Not because the methods are new, but because they are applied systematically: collect data, spot patterns, form a hypothesis, run a test, measure the result.

Master this loop and you have a system that gets better with every analysis. Only collect data and you have a dashboard nobody reads.

You can go deeper in the Academy track A/B-Testing Mastery ↗.