TLDR: Most ecommerce sites have analytics installed and still cannot trust what it tells them. When the numbers are quietly wrong you make real decisions on them anyway, and handing the same broken data to an AI just produces confident fiction faster. Before anything can analyse your data, human or machine, you have to be able to trust it. That is what this series builds.
Click to discuss this article with your favourite AI
In one month, a site I audited recorded 10,979 purchases and only 7,972 payment-step events.
Read that again.
More people completed a purchase than reached the step where you hand over your payment details. That is not a rounding error, it is an impossibility, and it had been sitting quietly in the checkout funnel for months while everyone made decisions around it.
Once you find it, the cause is almost mundane. The tracking fired its payment event after the payment step finished, and for anyone paying by a redirect method like Afterpay or PayPal that step happens on someone else’s site. The customer left, paid and came back, and the event that was meant to mark the moment never got the chance to fire. Around 3,000 purchases a month carried no payment record at all.
And 3,000 is only the floor, because it counts just the people who came back and bought anyway. It says nothing about the customers who chose a redirect method, landed on the provider’s page, changed their mind and left. They never return to the site, so they never appear in the data, and they are precisely the ones worth knowing about. Someone who reached for Afterpay and then backed out is a sale you nearly made and cannot even see.
More people bought than reached the payment step
Look at what that single quirk did to the reporting. On the dashboard the payment step looked like a disaster, roughly a quarter of buyers seeming to evaporate in the instant before they paid. Anyone sensible would read that funnel, decide the payment page was broken and set about fixing it.
But the payment page was fine. Those buyers were never missing, they were just invisible to one misfiring event. The real problem, a clumsy handoff to the redirect payment providers, was sitting somewhere else entirely, and the numbers were quietly steering everyone away from it. A quarter’s worth of effort would have gone into repairing something that already worked, while the thing that did not stayed untouched.
The danger is not missing data, it is data that is quietly wrong
Missing data at least announces itself. A gap in a chart makes you stop and ask what happened. The dangerous kind is the number that looks precise, lives inside a tidy dashboard and is wrong in a way nothing on the surface betrays. No one interrogates it, because there is nothing visibly to interrogate, so it slides straight into a decision.
That is the quiet condition most ecommerce analytics is in. The tag is installed, the dashboards are built, the figures carry two decimal places, and underneath it all the data is disagreeing with itself in ways nobody has thought to check. It is not that these businesses have no data. It is that they are trusting data they have no reason to trust, which is the more dangerous place to be, because it feels exactly like knowing.
The same trap, faster, with an AI
All of this matters more now that we are starting to hand the questions to machines. The pitch for AI analysis is seductive and mostly true: ask your data a question in plain English, get an answer in seconds. But point that same machine at the checkout funnel I just described and it lands on the same wrong conclusion a person would, only faster, and it hands you the verdict with a confidence no careful analyst would risk. Ask it why the payment page is leaking and it will tell you, fluently, precisely and completely wrong.
That is the trap. A model cannot tell the difference between data you can trust and data you cannot, and it will never stop to warn you. It just answers. So the smarter the analysis gets, the more it leans on the one thing nobody thought to check first. Person or machine, the constraint has not moved since the first spreadsheet: the answer is only ever as good as the numbers underneath it.
What this series builds
This is the first piece in a series on building a measurement foundation you can actually trust, for an ecommerce business, from the first conversation through to a live reporting layer. Across it I work through the whole thing in the order I actually do it on a job: name the decisions the business is really trying to make, walk the site by hand to map what a customer can do, audit whether any of it is tracked correctly, define each metric with an honest note on what you can trust, fix what is broken in the order that actually helps and, last, build a reporting layer that tells you when to worry and, just as often, when not to.
Every one of those steps exists to close a single gap, the distance between having data and being able to trust it. Until you close it, every hour someone spends analysing the numbers is an hour spent on a guess, whether that someone is you, an analyst or an AI.
And it does not begin where most people assume, with the tooling. It begins with the decisions the business needs to make. That is the next article.
Not sure if your data is trustworthy?
If you suspect your reporting is telling you a story you cannot quite trust, that instinct is usually right. I audit GA4 and GTM setups for ecommerce businesses and find the errors quietly shaping decisions.