How data drift happens
Nothing breaks loudly. A developer renames a button class and the click trigger that depended on it stops matching. The checkout gains an extra step, so the old thank-you page rule fires for fewer people. A platform changes what it counts as a session or a conversion, and the same report now measures a slightly different thing than it did last quarter. Marketing adds new campaigns without campaign tags, so a growing slice of real traffic lands in the wrong bucket.
Each change is small and defensible on its own. Data drift is what they add up to: a setup that was accurate when it was built and is quietly less accurate every month afterwards, without ever throwing an error. That is what separates drift from an outage. An outage announces itself when a number falls to nothing; drift only shows up when someone compares today’s definition of a metric with the one used when the target was agreed.
Why data drift matters
Decisions get made against the trend line, and drift bends the trend line. A channel that looks like it is fading may simply have lost its tags. A conversion rate that looks like it is climbing may be counting an extra event added some months back. Neither reading survives contact with the business’s own records, but nobody checks until a large decision depends on it.
It also erodes year-on-year comparison, which is the comparison seasonal businesses in Nepal rely on most — a trekking operator or an education consultancy is always asking how this season compares with the last one. If the measurement changed in between, the comparison is not measuring the market.
Common mistakes with data drift
The biggest mistake is treating tracking as a project that finished. Once the setup is signed off, nobody owns it, and no one is told when the site changes. Developers ship a redesign without a word to whoever built the data layer, and the tags come back partly broken.
The second is patching over history. When a metric jumps, people redefine it, add a filter or switch to another report until the chart looks sensible again. That fixes the appearance and destroys the ability to compare across time. Restating a number is fine; restating it without a written note of what changed and when is not.
How to act on it
Write down what each key metric means, which event feeds it and where the event comes from, then treat that document as the reference when a number looks odd. Without it, every investigation starts from scratch and usually ends in a guess.
Re-test the important paths on a schedule — an enquiry form, a call click, a purchase — and re-test them again after every release, plugin update or theme change. Reconcile with a source outside the analytics platform: the enquiry inbox, the CRM, the invoices. Where a proper analytics and tracking setup exists, drift is caught in weeks rather than discovered a year later in a board meeting.
