Which KPIs to track in your e-shop and how to evaluate them properly

An e-shop can have growing traffic and falling profitability at the same time. That is why it pays to track not only visits and revenue, but also the path from visit to order and the costs that come with the result.

Tie your metrics to the purchase funnel

Traffic, add-to-cart rate, completed orders and average order value help you pinpoint where a change happened. If orders are falling, total revenue alone will not tell you whether the problem is in traffic, conversion or basket value.

• Number of orders and conversion rate.
• Average order value by category.
• Gross margin, based on the costs you have available.
• Channel results with a clear attribution method.

Compare segments and check measurement quality

Break results down by device, channel, campaign, category, or new versus returning customers. Check whether measurement, attribution, product range or product availability has changed.

How to read a change in the funnel

If traffic stays similar but orders fall, check conversion by device and channel. If orders held up but revenue dropped, look at average basket value, discounts and product mix. If margin changed, break it down by category and include only the costs you can attribute reliably. Breakdowns like these help you decide what to ask next; they do not automatically reveal the cause.

Run a simple consistency check: do order counts in your analytics match your commerce system, are cancellations and returns counted the same way, and do campaigns use consistent tags? A change in tagging can make a channel look like it dropped even when the business result has not changed.

From metric to decision

Say average order value dropped in one segment. Check whether the mix of purchases changed, whether higher-priced products were available, and how large the discounts were. Only then choose an action you can evaluate, such as testing an adjusted offer in a limited segment. Compare the result against a period you defined in advance, and watch secondary metrics too, so that growth in orders does not mask a drop in margin.

For marketing attribution, always record the model and time window you used. Assigning an order to the last click is not the same as measuring the incremental effect of a campaign. If you do not have that data, word your conclusion accordingly: the report shows a correlation or a breakdown, not proof that advertising caused the sale.

A practical weekly e-shop check

Start with an overview of orders, revenue and average order value against a comparable period. If one indicator changed, break it down by category and channel. Then verify stock availability, pricing, discounts and any campaign changes. Keep a running log of website or tracking outages so that a technical event is not mistaken for a change in buying behavior.

For every test, set a primary metric and one or more guardrail metrics before launch. A page change can increase add-to-cart events without increasing completed orders; an aggressive discount can improve conversion and hurt margin at the same time. Summarize the evaluation together with the limitations of the data, and do not roll out a successful result to all segments if the test ran on only a small share of traffic.

Account for seasonality and assortment changes

Before evaluating a campaign or a sale, compare periods with a similar calendar context and check whether the product range was available. A holiday, a clearance sale or a stock-out changes both behavior and order mix. In a year-over-year comparison, explain it if the offer or the way you measure changed significantly. A single summary chart can hide the fact that one category is growing while another weakens, so prepare a breakdown that matches the business decision you need to make.

Draw the line between a report and an experiment

An observational report shows what happened; a controlled experiment helps estimate whether a change to your offer influenced the result. If you change a price, a page or your messaging, record the scope, the duration and the group the change applies to. Do not cherry-pick the period with the best result and present it as generally valid proof. With a small number of orders, the result can be uncertain, so state the limitations and avoid overly precise conclusions.