Why is my conversion rate dropping? Rank the suspects
A dropping conversion rate always has a cause, and nobody can retrieve it with certainty. Here is how to confirm the drop, rank the suspects by cost, and read any stated reason as a hypothesis.
A dropping conversion rate always has a cause, and nobody can retrieve that cause with certainty. A ranked hypothesis you can test beats a confident verdict you cannot.
Here is the objection in the words it usually arrives in, from someone asking why their conversion rate is dropping. "Your tool tells me what is wrong with the page. It cannot tell me why people actually left."
Half right. There is a why, and Revslip produces one. It will not hand that why over as a verdict. Nobody can honestly do that for one visitor on one page, and anyone claiming otherwise is selling certainty they do not own.
Did your conversion rate actually drop?
Confirm the drop is real before you hunt a cause. Open your order count in your store or CRM, then your analytics, for the same two weeks. If orders held and analytics fell, the tracking broke rather than the page. ConvertCart's diagnosis guide calls this the first false alarm, and it is the cheapest to rule out.
Then split the rate by device and by channel. A rate that fell across every segment at once usually means measurement. A rate that fell on mobile only, or on paid only, is a change you can go and find. Our four-check split between traffic and site covers the channel half.
Not sure which suspect is yours? Revslip ranks them by cost.
Why can nobody tell you why people left?
Because the reason a person gives for their own behaviour is reconstructed afterwards. Nisbett and Wilson established this in Psychological Review in 1977: people often cannot report accurately on what influenced a decision. Ask a visitor why they abandoned your checkout and you get a plausible story, not the cause.
The demonstration. In a study by Nisbett and Schachter, 12 subjects took a placebo pill described as causing the symptoms of arousal, then accepted four times as much electric shock as those who had not. Only 3 of the 12 said they had thought about the pill.
The pill moved them. They could not see it moving them.
That is your exit survey. That is also the person in the session recording, whose cursor you can watch and whose reasoning you cannot.
What does the research on stated reasons show?
It shows that behaviour observed is evidence and reasons reported are not. Jakob Nielsen of Nielsen Norman Group put it bluntly in 2001: self-reported data is typically three steps removed from the truth. He cites a correlation of 0.44 between measured user performance and stated preference, which accounts for roughly a quarter of the outcome.
- 0.44 correlation between what users do and what they say they prefer, per Nielsen Norman Group
- 3 of 12 subjects aware of the pill that quadrupled their shock tolerance, Nisbett and Schachter
- 30 days that Microsoft Clarity keeps an ordinary session recording before it is gone
Sources: Nisbett and Wilson, Psychological Review 84(3), 1977, pages 231 to 259, and Jakob Nielsen on why not to listen to users. Retention from Microsoft Clarity's own FAQ, which also confirms Clarity applies no sampling and no traffic limits.
Where the two camps part company: the standard playbook says run an exit survey to get the why, and Nielsen says do not listen. Both can be right. A survey answer is real data about what your visitor believes, and a hypothesis about what moved them, not a measurement of it.
One figure we will not repeat. Several tracking articles claim studies show around 70% of ecommerce stores have broken tracking. We chased it. No study, no sample and no method is cited behind it, so it stays out.
What do founders say when the number slides?
The pattern in the forums is recognition, not proof. One Shopify seller watched sessions and conversion ratio "fall off the radar mid August and in September" while reporting that "my sales haven't dropped and remain consistent", with UK sessions reading 3,366 one month and 51 the next.
Steady money, collapsing rate. That is a measurement failure wearing the costume of a conversion problem. Read it as the shape of the question people ask, never as evidence.
How do you rank the suspects?
Price them. A list of possible causes is an argument that never ends, so convert each one into a monthly euro figure and the order settles itself. Revslip uses one published formula for exactly this, and you can run it by hand on any suspect you have named.
traffic × CVR gap × AOV × mobile weight = monthly leak
- Name it narrowly. Not "mobile is worse". The checkout step where mobile falls away.
- Size the gap. Use your own better segment as the target, not a blog average.
- Multiply it out. 9,400 sessions, a 0.7 point gap, a 74 euro order and 62% of sessions on a phone gives roughly 3,019 euros a month.
- Fix the top one only. One change, dated, so the next movement has two candidates instead of twenty.
Want the ranked list instead? Run a free audit.
How long before the rate tells you anything?
Give it four weeks of the same traffic mix, and compare conversions rather than the rate. Below roughly 500 conversions a month a single change almost never reaches significance, so you are reading direction and not proof. Date the change, watch one metric, and accept a probable answer.
Our piece on telling a real lift from a quiet week sets out the before and after design. If you have the traffic to split properly, the threshold where testing becomes possible is the better route and Revslip is the wrong instrument for it.
When is hunting the cause the wrong move?
When what you need is one person's screen. Revslip does not record sessions and never shows you a cursor, so if you want to watch somebody rage-click a dead button, install Microsoft Clarity, which does that free. Use it for the broken interaction, and keep the ranking exercise above for choosing which one to watch.
Where our method stops being measurement Revslip's free estimate is arithmetic on four inputs, and the target rate is assumed rather than measured, which swings the figure more than the other three combined. Halve it before it goes in a plan, and on a paid plan replace it with tracked impact on the changes you shipped.
A word on our own numbers. Revslip has audited 134 businesses, a self-selected group, because a site gets audited once its owner already suspects something. Those figures describe worried sites, so check any pattern against your own analytics. The roughly 200 signals behind the report are published for the same reason.
Questions people ask when the rate slides
Can a tool tell me why my conversion rate is dropping?
It can tell you what changed and rank what it costs. The cause itself arrives as a hypothesis, because Nisbett and Wilson showed in 1977 that people cannot reliably report what influenced their own decisions. Treat any tool that states a cause as settled fact with suspicion, including this one.
Should I check session recordings or run an audit first?
Audit first, then record. An audit covers every page and prices what it finds, which tells you where to look. Session recordings then show you that one interaction in detail. Watching recordings before you know where the money leaks means sampling whichever visitors you happened to catch.
How do I read a heatmap without fooling myself?
Use a heatmap to confirm a suspicion you already named, not to go looking. A heatmap reports where attention landed and never why, and reading a cause into a warm patch is the error this whole post is about. Pair it with the ranked euro figure instead.