Conversion rate is the headline metric in most ecommerce operations. Visitors who buy, divided by total visitors, expressed as a percentage. It's tracked obsessively and optimised continuously.
It has a structural problem: it's a ratio, and ratios can be improved by shrinking the denominator. Which means a business can improve its conversion rate by attracting fewer customers.
How this happens in practice
Suppose you're running broad advertising that brings a large volume of loosely qualified traffic. Some of it converts; much of it doesn't. Your conversion rate is modest.
Cut that campaign. Traffic falls substantially, and the traffic that remains is your existing customers and people searching directly for you. Conversion rate rises sharply.
Revenue has fallen. The dashboard looks better. If anybody is being evaluated on conversion rate, they've just been rewarded for shrinking the business.
This isn't hypothetical. It's one of the more common ways optimisation efforts go wrong, and it's difficult to spot because every individual decision looked like an improvement.
Traffic quality dominates
The broader point: conversion rate is largely determined by who arrives, not by what happens once they do.
Someone searching for a specific product name converts at a dramatically higher rate than someone who clicked a display advert while reading something else. That difference swamps almost any change you can make to a checkout flow.
Which means comparing conversion rates between businesses, or between periods with different traffic mixes, tells you mainly about traffic composition. Benchmark figures published as industry standards are close to meaningless for this reason.
A useful discipline: segment conversion rate by traffic source and never look at the blended figure. The blended number moves when your mix changes and tells you nothing about whether anything got better.
What to measure instead
A few alternatives that are harder to game.
Revenue per visitor. Combines conversion rate and order value. Immune to the shrinking problem — cutting traffic that converts poorly but does convert will reduce this figure, correctly.
Contribution per visitor. Better still. Revenue minus variable costs, including the cost of acquiring that visitor, per visitor. This is the number that determines whether growth is worth having.
Total contribution. The absolute figure. Ratios can improve while the business declines; absolute contribution cannot.
Repeat purchase rate and customer lifetime value. A conversion at the cost of a customer who never returns is worth much less than one that begins a relationship, and conversion rate cannot distinguish them.
The discounting trap
The most common way conversion optimisation destroys value.
Offering a discount reliably increases conversion rate. It also reduces margin on every sale, including all the sales that would have happened anyway.
If a discount converts a small additional share of visitors while reducing margin on everyone, the maths frequently doesn't work. It looks like it does, because conversion rate went up and revenue went up, and contribution went down.
This is why measuring contribution rather than revenue matters so much. Discounting always improves revenue metrics in the short run and frequently destroys profitability.
Where optimisation genuinely helps
Not an argument against improving the experience, which does matter. The gains are just concentrated in specific places.
Removing friction that has no purpose. Forced account creation, excessive form fields, unclear delivery information, unexpected costs appearing late. Each of these loses people who wanted to buy.
Cost transparency. Unexpected delivery charges at the final step are consistently the largest single cause of abandonment in the research. Showing costs early loses some people earlier and loses fewer overall.
Speed. Page load time has a measurable relationship with conversion, particularly on mobile, and it's one of the few interventions with a clean causal story.
Payment options. Offering the methods your customers actually use. Varies enormously by market and is frequently an oversight.
What these have in common is that they remove obstacles for people who already intended to buy. That's a genuine improvement rather than a shift in the mix.
Testing properly
One methodological note, since most ecommerce testing is done badly.
Tests need to run long enough to reach adequate sample size and to cover full weekly cycles, because behaviour differs by day. Stopping a test early because it's showing a positive result is the most common error and it produces false positives at a high rate.
And most tests show no effect. A programme reporting that most of its tests produced wins is almost certainly measuring noise, not because the team is dishonest but because early stopping and multiple comparisons make it very easy to find effects that aren't there.
The honest expectation is that the majority of changes do nothing, a few do harm, and occasional ones help meaningfully. That's what a well-run programme looks like and it's not what most reporting shows.
The mobile split
One segmentation worth doing before any other: separate mobile from desktop. Conversion rates on mobile are consistently lower across nearly every category, sometimes by a wide margin, and the reasons are partly behavioural and partly execution.
Behaviourally, a lot of mobile sessions are browsing rather than buying — people research on a phone and complete on a laptop. That inflates the mobile denominator with sessions that were never going to convert on that device, and no amount of optimisation fixes it because the visitor is doing something else.
On execution, the friction that costs you is different. Form filling, payment entry and image inspection are all harder on a small screen, and the fixes are specific: wallet payment options, minimal typing, and product photography that works at thumbnail size.
Reporting a blended figure across both hides all of this. The two behave like different businesses and should be measured that way.