Is your website data telling you the truth?

Analytics configurations are set up once, carefully, by someone who understood them. Then the site changes. A checkout is rebuilt, a subdomain is added, a consent banner is installed, a developer leaves. None of these events announces that it has broken the measurement, and so the numbers keep arriving, looking exactly as reliable as they did before.

Why configurations drift

Nobody breaks analytics deliberately. It happens because tracking is invisible: a page loads correctly whether or not the tag fired, so there is no failure to notice. The reporting continues, the graphs continue, and only the relationship between them and reality has changed.

The cost is not the wrong number. It is that decisions get made on it: budget moved to a channel that looked efficient because half its conversions were being credited from somewhere else, or a product line discontinued because the tracking on its category page had been broken since a redesign.

The nine checks

None of these needs a specialist. Most take a few minutes, and any competent developer or agency can answer the technical ones the same day.

1. Is the tag on every page? Including checkout, thank-you, blog, landing pages built by the marketing team, and anything on a subdomain. Missing pages are the most common single fault and usually appear after a redesign.
2. Is it on any page twice? Double-tagging halves your reported bounce rate and inflates sessions. It happens when a tag is added directly and again through a tag manager, and it makes historic comparisons meaningless.
3. Is internal traffic excluded? Your own team, your agency and your developers browsing the site. In a low-volume business this can be a material share of reported sessions and it behaves nothing like a customer.
4. Do the conversion goals still exist? Goals defined by a URL break the moment that URL changes. A confirmation page moved from /thank-you to /order/complete and nobody updated the goal. Conversions simply stop, or worse, drop by an amount small enough to look like a bad month.
5. Is revenue passed with the transaction? Not just the count. Without the value you cannot compute return on spend, and any channel comparison is a comparison of volume rather than worth.
6. Are refunds and cancellations subtracted? In categories with high return rates, gross reported revenue can be dramatically higher than what is retained, and the channels that drive the most returns are frequently the ones that look best.
7. Do the referral exclusions cover your payment provider? If a customer leaves for a hosted payment page and returns, the original source can be lost and the sale attributed to the payment provider. Direct traffic that converts suspiciously well is the classic symptom.
8. What does the consent banner actually do? Where analytics only fires after consent, the reported figures represent consenting visitors, not visitors. That is not wrong, but it must be known, particularly when comparing periods either side of the banner being installed.
9. Is anyone still looking at bot traffic? Filtering is better than it was, but automated traffic still inflates sessions on smaller sites, and it is concentrated at times of day that make the pattern look like a real audience.

The one that matters most

If you only do one thing on this page, do this: take last month’s revenue as reported by your analytics, and compare it with the revenue banked from online orders in the same period.

They will not match exactly, and they are not supposed to. Timing differences, refunds processed in a later period, failed payments and tax treatment all account for some gap. But you should be able to explain the difference in a sentence, and it should be stable month to month.

Where the gap is large, moves unpredictably, or nobody has ever calculated it, every decision that used the analytics figure was made on a number nobody had checked. In our experience this single comparison finds more real problems than the other eight checks combined.

A number nobody has ever reconciled is not data. It is a rumour with a chart attached.

What to do with the answers

Resist the urge to fix everything at once, and above all resist restating history. Reprocessing past data to “correct” it usually produces a second set of figures that disagrees with the first, and now nobody trusts either.

1. Fix forward. Correct the configuration and draw a line. From this date, the numbers are reliable.
2. Write down the line. Record what was wrong, when it was fixed, and how it affects comparisons across that date. In six months somebody will ask why a metric stepped.
3. Do not compare across it naively. Where a fix materially changed what is captured, say so on any chart that spans the change rather than letting a reader assume a trend.
4. Diarise the re-check. Every six months, and after every significant site change. Configuration drift is not a one-off event; it is a permanent condition of a site that is still being developed.

None of this is glamorous work, and it is regularly the highest-return hour anyone spends on a commerce business in a given quarter, because everything downstream of the numbers depends on the numbers.

The analytics audit described here is the first step of most eCommerce engagements, before any margin or conversion work begins. It is general guidance rather than advice on your particular configuration.