A store’s overall conversion rate can stay steady while mobile checkout quietly loses orders. The average hides it because other segments absorb the decline. Data analysis in e-commerce means breaking the funnel down by device, channel, page, and customer cohort until a leak has a location, an owner, and a number attached to it.

Cart abandonment remains a major source of potential lost revenue. Baymard Institute’s running average puts online cart abandonment at 70.22%, with extra costs, slow delivery, lack of trust, forced account creation, and complicated checkout among the reported reasons.

This guide looks at how analytics can identify which of those leaks are recoverable and where fixing them can improve conversion and customer lifetime value.

What Should Data Analysis in E-Commerce Actually Measure?

Data analysis in e-commerce is the measurement and modelling of how visitors become buyers and buyers become repeat customers, at a granularity that locates where value is lost and who can recover it.

The measures fall into three groups:

• Funnel measures: sessions, product views, add-to-cart, checkout starts, checkout completions and orders, each split by device, channel, landing page and new versus returning.
• Value measures: revenue per session, average order value, contribution margin per order and, over time, customer lifetime value by acquisition cohort.
• Friction measures: page load time on the checkout and search pages specifically, zero-result search rate, payment failure rate by method, and behavioural signals such as rage clicks and dead clicks on the pages that matter.

The third group is one many stores do not instrument, and it can help explain the first two. A rising abandonment rate is a symptom. A checkout page whose load time doubled after a plugin update, or a search index that returns nothing for a new product line, is a possible cause, and it can be visible days before the abandonment rate moves.

In many implementations, instrumenting the friction group is a useful starting point. Depending on the stack and the data-quality and privacy requirements, this may combine client-side analytics, server-side tagging or server-to-server events with behavioural and performance monitoring.

GA4 supports standard e-commerce events such as add_to_cart, begin_checkout, add_payment_info and purchase without server-side tagging, and server-side tagging complements client-side measurement rather than universally replacing it.

The GA4 side of that build is covered in our guide to real-time custom insights with GA4 APIs.

Why Do Averages Hide the Leak?

Because a store is several stores, and data analysis in e-commerce has to treat it that way. Mobile and desktop shoppers arrive with different intents, on different networks, facing different checkout layouts, and they convert at different rates for different reasons.

Paid social traffic and organic search traffic land on different pages with different expectations. New and returning customers face different friction: the new customer may meet the account-creation wall, the returning one the saved-card failure.

Blend all of that into one conversion rate and the number moves only when several segments move together, by which time the loss may be a quarter old.

The fix is structural. Every funnel metric on every report is split by device, channel, landing page and customer status before it is shown as a total. Revenue per session sits next to conversion rate so that a segment losing orders shows up as money lost.

Segments should be diagnosed against the right comparison. Baymard’s 70.22% is an aggregate across many studies, not a device, channel or customer-segment target, so it should not be used to rank which segment to fix first. A better comparison is each segment’s own historical baseline, adjusted for seasonality and promotions, plus comparable cohorts and the business context.

Let’s take an illustrative case. Traffic holds steady, the blended rate holds steady, and mobile checkout completion has slipped against its own trend for a quarter while desktop absorbed it. Split by device from the start, and the slip can become a ticket in week one instead of a board question in month four.

What Are Five Signals Behind Cart Abandonment?

These are five signals we suggest instrumenting early in e-commerce analytics work, because they can surface friction before the abandonment rate reports it. A note on evidence:

Baymard’s reasons are self-reported by shoppers in surveys. They show what shoppers say drives them away, not that each reason maps to a single analytics signal or that instrumentation alone recovers the lost order.

The signals below are diagnostic tools. Each should have an owner on the client’s team who can act on what it shows.

1. Abandonment split by device and step

Checkout is tracked step by step, cart to shipping to payment to confirmation, separately on mobile and desktop. The step with the largest drop relative to its own baseline on the weaker device is a strong candidate for the first investigation. In many stores this split is a useful starting point because it needs little beyond correct event tracking.

2. Rage clicks and dead clicks on the pages that matter

A behavioural analytics tool records where shoppers click repeatedly in frustration (rage clicks) or click on elements where no response is detected (dead clicks). On the cart, checkout and search pages, these events can be reviewed or alerted on. These signals can reveal interaction problems that may contribute to checkout frustration.

They are not the same as website crashes or errors, so application and error monitoring should be used to detect actual failures. Some tools, such as Microsoft Clarity, separately flag error clicks associated with JavaScript errors, which is a closer proxy for technical faults.

3. Zero-result and low-result searches

Site search queries that return nothing are logged, ranked by volume and reviewed on a regular cadence. Each one may point to a synonym to add, a product to index, or a demand signal for merchandising. A shopper who searches and gets nothing rarely complains. Zero-result logs are one of the clearest places to identify unmet search demand, and they work best alongside funnel events, exit behaviour, session replay and clickstream data, which show other forms of departure.

4. Load time on the checkout and search pages specifically

Site-wide performance averages can hide a slow checkout. The checkout and search URLs can be monitored individually with page-level tools and real-user monitoring, with alerts when their core metrics regress.

The business case is illustrated by a Vodafone landing-page experiment documented by Google. In a 50/50 A/B test of roughly 100,000 clicks a day per version, a 31% improvement in Largest Contentful Paint came with an 8% increase in total sales, an 11% uplift in cart-to-visit rate and a 15% uplift in lead-to-visit rate.

This was one experiment on one telecom landing page. It is not a conversion formula, and it does not show that a given speed improvement will produce the same uplift in another retail context. What it supports is measuring page performance alongside conversion outcomes, instead of assuming a universal uplift.

5. Payment failure rate by method

Declines, gateway errors and abandoned authentication steps are tracked by payment method from the provider’s events. A rising failure rate on one wallet or one card type may be a configuration or provider problem that looks like abandonment in the funnel. Baymard’s survey figures for declined cards and for too few payment methods are relevant context here.

Tracking helps diagnose these losses, and analytics alone does not recover the shopper. A decline can stem from issuer decisions, available funds, authentication, fraud controls or other causes, and “not enough payment methods” is a different problem again.

The signal helps isolate payment-related loss and prioritise investigation and recovery actions. The integration side is covered in our guide to secure payment systems for e-commerce.

How Does Predictive Consumer Data Analytics Raise Lifetime Value?

Predictive consumer data analytics moves the store from measuring what happened to deciding what to do next, and one decision with potentially large returns is who to acquire and who to keep.

The marketing science literature has had the tools for two decades. Gupta, Hanssens, Hardie and colleagues’ 2006 review in the Journal of Service Research defines customer lifetime value as the metric marketing exists to maximise. It reviews several implementable models and positions them as tools for market segmentation and the allocation of resources across acquisition, retention and cross-selling.

The e-commerce application is direct: model expected purchases and expected margin per customer from their transaction history, group customers by predicted value, and set acquisition cost and retention effort per group.

The paper reviews multiple modelling approaches and does not prescribe one sequence. As an illustrative implementation pattern, a store might build up from:

• A repeat-purchase model that predicts whether and when a customer buys again, from recency, frequency and monetary value plus behavioural signals.
• A margin model that predicts contribution per order, including returns, so that high-revenue, high-return customers are not mistaken for high-value ones.
• A churn or missed-order model, which makes sense where there is a meaningful expected purchase or subscription cadence, such as subscription or replenishment segments. Where purchasing is irregular, it may add little.

Which models to build, and in what order, should follow the business model and the data available. Whatever is built is validated on a held-out period and measured on lift over the current rule, which is often spend by last-purchase recency. The AI side of merchandising and personalisation that sits on top is covered in our guide to AI retail optimisation.

What Does an E-Commerce Data Visualization Dashboard Need to Trigger Action?

An e-commerce data visualization dashboard is judged by what happens after a number moves, and the design that produces action has four properties:

1. 1. Every metric is segmented by default, with the blended total shown last.
2. 2. Each critical metric has an owner, a threshold and a defined action, so a mobile checkout completion drop lands with the person who owns mobile checkout rather than in a weekly meeting.
3. 3. Revenue per session sits beside conversion rate on every page, so a leak is expressed in money.
4. 4. Every number reads from one governed definition of orders, revenue and margin that finance agrees with, because the first time the dashboard’s revenue disagrees with the finance pack, it loses its authority.

The refresh question for an e-commerce data visualization dashboard is answered by the decision, not by a fixed rule. Refresh frequency should follow the latency of the decision being supported. Some merchandising decisions are weekly, and a daily batch update is enough for those.

Other marketing decisions may justify hourly refreshes. A flash sale or a payment outage may need alerts within minutes, which usually means the behavioural, performance and payment monitoring tools alert directly rather than waiting for the dashboard. Fraud and payment decisions are real time and belong in the payment system.

Building a real-time analytics warehouse to serve decisions that are made weekly can be an over-investment, and the trade-off is the same one we describe for executive dashboards in general.

What Do Retailers Get Wrong About Data Analysis in E-Commerce?

Common mistakes include:

• Leading with traffic. Sessions and pageviews get reported first, before anyone asks how many of those sessions reached checkout, and more traffic is poured into a funnel nobody has examined.
• Relying on the blended rate. One conversion figure across every device and channel can hide the segment that is losing, and the loss compounds until it is large enough to move the average.
• Treating friction as minor. A slow checkout page or a bad search result gets logged and never revisited, while the behavioural signals that would have flagged it are not instrumented at all.
• Building dashboards without owners. A page of segmented metrics with nobody named against each one produces the same silence as a blended one.
• Using predictive models without a baseline. A lifetime value model that cannot show lift over spend-by-recency has not earned a budget, and the discipline for proving that is in our guide to the benefits of predictive analytics.

Split the Average, Find the Leak, Name the Owner

Data analysis in e-commerce pays back when it stops averaging. Roughly seven in ten carts are abandoned across the documented studies, though part of that is browsing and comparison shopping.

Shoppers’ self-reported reasons point to causes a store can investigate with instrumentation, and the market those carts sit in is growing at double digits. Friction signals can help find a leak before the conversion rate reports it, and segmentation locates it. Lifetime value modelling can inform where recovered margin is best reinvested, and a dashboard with owners and thresholds helps make sure someone acts.

Fix the tracking, split every metric, instrument the checkout and search pages, model value on lift, and put a name against every number. Talk to Ariel about an e-commerce analytics review, and we can help identify which segments of your store may be losing orders inside a healthy average.

Frequently Asked Questions

1. What is data analysis in e-commerce?

Data analysis in e-commerce is the measurement and modelling of how visitors become buyers and buyers become repeat customers, at a granularity that locates where value is lost and who can recover it. It covers funnel measures split by device, channel and cohort, and value measures such as revenue per session and lifetime value. It also covers friction measures such as checkout load time, zero-result searches, behavioural click signals and payment failures.

2. What is the average cart abandonment rate?

Baymard Institute’s running average of documented online cart abandonment is 70.22% across 50 studies. It is an aggregate, and Baymard notes that much abandonment comes from browsing, comparison shopping and shoppers who were not ready to buy. Baymard’s survey of self-reported reasons puts extra costs such as shipping, tax and fees at 40% and slow delivery at 20%. Lack of trust with card details follows at 19%, forced account creation at 18% and a long or complicated checkout at 17%. The 70.22% figure is not a per-segment target. Segments are best compared against their own historical baselines and comparable cohorts.

3. How does page speed affect e-commerce conversion?

Page speed can affect conversion, but the size of the effect varies by site. In one Vodafone landing-page A/B test documented by Google, with roughly 100,000 clicks a day per version, a 31% improvement in Largest Contentful Paint came with an 8% increase in total sales. Cart-to-visit rate rose 11% and lead-to-visit rate rose 15%. That is a single experiment, not a general formula. The practical lesson is to measure page performance alongside conversion outcomes, and to monitor checkout and search pages individually because site-wide averages can hide a slow checkout.

4. What is predictive consumer data analytics in retail?

Predictive consumer data analytics uses transaction and behavioural history to forecast what a customer will do next. That can mean whether and when they will buy again, what margin they will contribute, and, where there is a purchase cadence, whether a subscriber is about to lapse. The models group customers by predicted lifetime value so that acquisition cost and retention effort can follow value. Each model is validated on a held-out period and judged on lift over the current rule.

5. What should an e-commerce data visualization dashboard include?

Every funnel metric segmented by device, channel, landing page and customer status, with the blended total shown last. Revenue per session beside conversion rate so leaks show in money. An owner, a threshold and a defined action on each critical metric. And one governed definition of orders, revenue and margin that finance agrees with. Refresh frequency should follow the latency of the decision the dashboard supports, and urgent friction or payment alerts are often better fired directly from the monitoring tools.

6. How do you increase customer lifetime value with data?

Model it, then act on it. Depending on the business, that can include a repeat-purchase model that predicts who buys again and when, a margin model that predicts contribution per order including returns, and, where there is a purchase cadence, a churn model that catches lapsing customers. Customers are grouped by predicted value, acquisition spend is set by channel according to the value it brings, and retention effort is targeted at high-value customers showing early lapse signals. Measure lift against a baseline such as spend-by-recency before scaling.