I had already decided where the problem was.
On a marketing review call, I said the issue was on the next page. Checkout was the suspect, and I was ready to inspect it.
That is a useful feeling. You have a place to work. Move the fields. Change the button. Shorten the form. Rewrite the sentence near the card field. There are enough things to adjust on a checkout page to keep you occupied for the rest of the afternoon.
Then I opened it.
The checkout used the same URL as the page before it.
That was a problem for the heat-map view we were using. I wanted to see how people interacted with checkout, but the view could not isolate that stage from the page around it.
We had a question about the buyer's next step and an instrument that couldn't distinguish the step.
A buyer can move from reading an offer to opening checkout without the address in the browser changing. To the buyer, something has happened. The task has changed. Reading about a purchase and entering payment information ask different things of a person.
Our measurement needed to recognize that change.
The limitation belonged to the view we had in front of us. A checkout can work on the same URL, and a different event or recording target might have made the distinction visible.
Until then, we could not use that heat map to answer the question we had brought to it.
How many people opened checkout? How many interacted with it? Where did they stop? Did they encounter an error?
Those answers were missing.
I had named a suspect before I had a way to observe it.
Unknown is an uncomfortable answer when a business wants a decision.
Checkout abandonment sounds more useful. So does form friction. Either gives you a reason to open the editor and start changing things.
But a plausible explanation can get ahead of the evidence. Once it does, every detail on the page starts looking like support for it. That field looks unnecessary. That button looks weak. Surely someone would hesitate at that sentence.
Perhaps they would. We still need to see whether they did.
There is a difference between finding something you dislike on a page and finding the thing that is costing the business orders. The first can take seconds. The second requires a record of what buyers had the chance to do and what happened next.
Without that record, a redesign can consume a week and leave the original question unanswered.
The conclusion we could support was narrow: checkout behavior was not visible as its own stage in the current view.
The number of visitors who reached that point remained unknown. An abandonment rate or an explanation of why people left would have required evidence we did not have.
The next useful move was to make the stage visible.
That means giving checkout opening a signal of its own, then distinguishing interaction, errors, and completed orders. It also means knowing which visitors had the chance to take each step. A count of orders means something different when it follows ten eligible visits than when it follows a thousand.
Those are the conditions for a diagnosis.
The record of this call ends with the visibility problem and the decision to improve it. It does not establish that the new measurement was implemented or that performance improved afterward.
I would rather leave that ending open than attach a result the source cannot support.
I don't think the lesson is to stop using instinct. Suspicion got me to open the checkout and look.
The useful job for instinct is to lead us to a question we can answer. When it becomes the answer itself, we can spend a lot of time fixing the page we happened to blame first.
Before changing a suspected weak point, I want to know whether the measurement can distinguish it from the steps on either side. That requires a record of who arrived, what they did, and what followed.
If we can't, improving that visibility is real work. It earns its place ahead of the button change.
Once the stage can be seen, we may find the problem we suspected. We may find a different one. We may discover that there isn't enough evidence to justify a change yet.
Any of those answers gives us more to work with than a confident label on missing data.
On that call, I started with the next page.
The first thing that needed fixing was my ability to see it.
Source note: Anonymized from a May 14, 2024 marketing review call. The record supports the measurement limitation and the decision to improve checkout visibility. It does not establish checkout abandonment, a completed implementation, or a later performance result.
I turn this kind of reading into a working document for one business at a time: who your buyers imitate, what they want beneath the surface, and the words they use when nobody is selling to them. The first bounded read is free. Mimetic Intelligence, built by hand with source-linked evidence and stated confidence.