"Our conversion rate is down." It's one of the most common sentences in ecommerce, and one of the least useful on its own. A site-wide conversion rate is an average of very different visitors: people who arrived ready to buy, people browsing on a phone during a commute, people who clicked an ad by accident. When the average moves, the interesting question is always whose behaviour changed.
How an average can mislead
Here's a hypothetical store, with made-up numbers to show the effect. Its overall conversion rate drops from one month to the next, and the team starts planning a checkout redesign.
Down
Overall conversion rate, month on month (example)
Flat
Desktop conversion rate (example)
Flat
Mobile conversion rate (example)
Up
Share of visits coming from mobile (example)
Nothing got worse for any group of visitors. What changed is the mix: more visits came from mobile, where people typically convert less. The overall rate fell because of who was visiting, not because the site got worse. A checkout redesign would have been aimed at a problem that didn't exist.
The splits worth checking first
You can slice traffic a hundred ways. These four explain most of the surprises I see:
1
Device
Phones and desktops behave differently enough that a change in the mix alone can move the overall rate, as in the example above.
2
New versus returning visitors
Returning visitors usually convert better. A big acquisition campaign brings lots of new visitors and can lower the overall rate while the business grows.
3
Channel
Email subscribers, paid social clicks and search visitors arrive with very different intent. Look at each channel's rate before judging the total.
4
Landing page
Someone landing on a product page from an ad is further along than someone landing on the home page. Compare like with like.
Check the mix before the rate
When the overall rate moves, first look at whether the share of each segment changed. If the mix shifted and each segment's own rate held steady, the site is fine and the story is about who's visiting.
When a segment really does drop
Sometimes the split shows a real problem: mobile conversion falls while desktop holds steady, say. That's good news in a way, because a narrow, specific drop is much easier to investigate than a vague overall decline.
Check what changed for that group: a new app, a theme update, a payment option missing on one browser, a slow page. A specific question gets a specific answer, and usually a much cheaper fix than a redesign.
The habit to build is simple: never react to the overall rate on its own. Split it first, and let the split tell you where to look.
