Web analytics & conversions
Funnels leak in one place; find it with a verdict
Where users drop, and whether the change is real, funnel reads that end in a z-test, not a hunch.
Conversion rate slipped, so you changed the landing page, the form, the pricing table and the checkout across three sprints. It came back up. Nobody can tell you which change did it.
Worse, your funnel is harder to reason about than before you started. Four things moved at once, and none of them left a record of what it was worth.
The cheap version of that quarter takes an afternoon. Find the one step where the funnel actually leaks. Confirm the leak is real. Then point the sprint at that step alone.
Funnels leak in one place at a time
A funnel check lines your counts up in order, sessions into conversions into transactions, and computes the drop at each step. One step is almost always the outlier, losing visitors at a rate the others do not come close to. That is the leak, and the check names it outright instead of handing you a chart to interpret.
The concentration is the useful part. A team that knows its leak sits between conversion and transaction stops rewriting landing pages, because the landing pages were never this quarter's problem and no amount of copy work was going to move the number.
The same check sizes the prize. A leak's width multiplied by the traffic reaching it says what fixing it could be worth. That is the number that gets a sprint prioritized over the redesign wishlist.
Ordering the steps takes one decision, and it is yours: the sequence your business already believes in, top to bottom. Everything after that is arithmetic.
Ask it yourself
Where do users drop off between sessions, conversions, and transactions, and which step leaks the most?
Test the leak before you redesign
A leak that widened this month gets one question before anybody opens a design tool. Is the change real, or is it the ordinary week-to-week wobble this metric has always had? The significance check answers with the test that fits the metric, and hands back a plain verdict with the p-value beside it.
Rates need this more than counts do. A conversion-rate wobble on a modest session base can look dramatic and mean nothing at all. The article on telling a real change from noise covers why waiting is so often the correct answer.
The test protects your good weeks too. A leak that narrowed gets the same treatment before anyone claims the fix worked, so you record your improvement as carefully as you recorded your problem.
You do not have to choose the test. Rate changes get the proportion test on the underlying counts, volume changes get the comparison suited to counts, and the answer arrives fitted to your metric.
What this funnel check does not do
It works on macro steps built from your analytics count metrics. It finds which stage of the journey leaks, and it does that well. What it will not do is trace individual people through button-by-button event sequences, and the card says so on its face rather than implying a precision it does not have.
For most funnel decisions the wide-angle view is the one that matters anyway. Knowing the leak lives between visit and signup narrows your investigation to one stretch of the journey. The event-level microscope comes out afterwards, pointed at that stretch.
Saying the scope out loud sets up the right handoff. The macro check names the step. The event-level tools go inside it. Neither one pretends to do the other's job.
The leak arrives with a name
Your funnel's stages come from the same named metrics the rest of your reporting uses: sessions, your goals, transactions. Conversions here always mean the named goals your team defined in your analytics tool, by name. So a drop between Demo Request and Purchase is a sentence your revenue team can act on without translating it first.
That vocabulary is doing real work. It turns a leak into an assignment somebody can accept.
The ordering is yours, and it is worth choosing once. A funnel whose steps mirror how your business actually talks about its pipeline produces a report nobody has to reinterpret.
Run the same funnel next month
Then your fix shows up as the leak narrowing, in the same terms that found it. The before and the after live in one chart. Across a few quarters your team learns which interventions move which steps, which is knowledge that compounds.
The worked example below shows the verdict half in action: a conversion movement arriving with its statistical reading attached, before anyone reacts to it.
Repeat it long enough and your funnel's seasonality surfaces as well. Which steps always widen in which months, so next year's wobble arrives pre-explained.
The workflow that does this: Conversion spike: real? →