How to find where users drop off
Short answer
Read the step-to-step drop rather than the end-to-end rate. The overall number tells you how the month went; the step drops tell you what to change. Then size a fix by what it would produce, not by how many people it touches.
The funnel below is the demonstration signup flow used across this site, compared against the week before. The counts are illustrative; the report is the standard one.
Build the funnel from what people actually do
A funnel is a sequence of steps in the order people move through them. For most sites the useful one is short: landed on the site, viewed pricing, started signup, finished signup, paid.
Resist adding steps you do not intend to act on. Every extra step splits your numbers thinner and makes the interesting drop harder to see.
Read the drops, not the totals
Each step shows two numbers that matter: the share of the original audience still present, and the share lost since the step before. The second is the one to act on.
A funnel converting 2.8% end to end sounds like a broad problem. If 75% of the loss happens between starting signup and finishing it, it is not a broad problem at all. It is a form.
- Find the single largest step-to-step drop.
- Check how many people it affects in absolute terms.
- Multiply: a 10% improvement on the worst step usually beats doubling the top of the funnel.
What it told us. Traffic and the first two steps moved by a couple of points. The last step lost a sixth of its people, and since everything above it grew, that step is the whole of the week.
Late steps can matter more than early ones
It is tempting to focus on the first drop because it involves the most people. But people deep in a funnel have already shown intent, so recovering them is usually cheaper than acquiring new visitors, and they convert at a much higher rate afterwards.
Split the funnel before you rebuild anything
Before redesigning a step, break the funnel down by device and by source. A step that looks broken on average is often only broken for one group. Mobile signup completing at a third the rate of desktop points at the form on small screens, not at the offer.
That distinction decides whether you spend a day on CSS or a month on positioning.
What it told us. Desktop is where it was. The entire loss is on mobile, which turns a question about the offer into a question about a form on a small screen — a day of work rather than a month.
Measure the fix against the same funnel
Keep the funnel definition fixed after you ship a change. If you edit the steps at the same time, you will not know whether the number moved because the product improved or because the measurement did.
Where that leaves you
The month's conversion rate fell by six tenths of a point, which on its own suggests nothing in particular. Split into steps it is one step, and split by device it is one step on one device. That is a change somebody can actually ship, and the same funnel definition will say next week whether it worked.
Where this evidence is weak
A funnel counts only people who entered at step one and it counts steps in order, so anybody who arrived straight at signup from an email never appears here at all. The last step is also the thinnest: at 847 people a quiet weekend moves it more than a broken form would.
Questions this guide answers
4- How do I find where users drop off?Read the drops
- Which funnel step should I fix first?Late steps
- Should funnel steps be pages or events?Build the funnel
- Why does the average funnel look fine when conversion fell?Split the funnel
Do this with your own numbers
Every table above is a standard report over a standard range. Nothing here was built for the example.
14-day free trial. No credit card required.