I'm the sole developer of the booking platform for a Tokyo photo studio. Customers book from 20+ countries before they arrive in Japan: they pick a plan, a location, and a time slot in their own language, then pay a deposit. Every one of those steps is a place someone can quietly leave.
Instrument every step first
Before changing anything, I made sure each step of the booking flow reported to GA4. A single bookings number tells you how the business is doing. It doesn't tell you where it's leaking. With an event on every step, the funnel becomes a sequence you can read, and a drop between two steps stands out on its own.
I added Microsoft Clarity to the same flow. GA4 answers how many people moved from one step to the next. Clarity answers what those people actually did on the screen.
39%
drop-off between date and time selection
GA4
events on every funnel step
Clarity
session-level replay
Where people left
The step that stood out sat between choosing a date and choosing a time: 39% of people who picked a date didn't go on to pick a time slot.
That's a strange place to lose people. By the time someone picks a date, they have already chosen a plan and a location. They're close to done. Intent isn't the likely problem at that point, which pointed at the interface rather than the offer.
Why replay mattered
A funnel chart can show where people leave. It can't show why. Watching session-level recordings of people at the date and time steps turned a percentage into specific moments on the screen, and those moments are what the fixes were aimed at.
What I'd tell a client
If you're about to redesign a checkout, instrument it first. The fix you need is often one step, not the whole flow.
What I shipped
I shipped fixes to that part of the flow. The measurement that matters next is the same funnel over a comparable period after the change, which is why this post doesn't quote an after-fix number.
The general lesson
- Put an event on every step, not only on the conversion.
- Use the funnel to find the step, and replay to understand it.
- Fix the step that leaks before touching the ones that don't.
It's the same approach I take on client work: measure before automating or rebuilding anything, so the time goes where the problem actually is.