A team repairs a conversion tag. The dashboard shows fewer sales attributed to paid media the next morning. Finance asks whether the channel has deteriorated; marketing wonders whether the previous figures were wrong. Both may be asking the wrong question. The first job is to establish what changed in the ruler.
Tracking changes can alter which events are captured, how they are classified, when they arrive and which campaigns receive credit. A cleaner implementation can therefore reduce reported conversions while business outcomes remain stable. The reverse is possible too: a broader event definition or a new attribution setting can raise the reported total without creating another customer.
Build the comparison before explaining the movement
Record the old and new definitions side by side. Which event counts as a purchase? Does the event fire after payment, after booking confirmation or after a lead form is submitted? Are refunds and cancellations removed? Which value is sent, and in what currency? Has a supporting event become a primary optimisation action, or are two actions describing the same customer journey?
Then map the technical change. Note the tag or SDK version, consent handling, deduplication logic, attribution window, channel parameters, CRM status, data import and any transformation between source systems and the dashboard. A change to any of these can affect coverage or classification. Record the deployment time and the first date from which the new data is comparable.
This is a measurement bridge, rather than an attempt to force the new total to match the old one. If the previous setup counted duplicate bookings, retaining that higher number for continuity would preserve a known error. The bridge should explain what can still be compared, what can’t, and which series should be used for each decision.
Separate historical correction from forward measurement
Some platform settings affect only future data. Google Analytics, for example, states that data filters are evaluated from the point of creation forward and do not change historical data; an applied exclude filter can permanently remove that incoming data from reports and BigQuery. That means a before-and-after series may reflect different treatment of internal traffic, rather than a change in customer behaviour. Check the system’s own documentation before assuming that a setting can repair the historic record.
Where the old and new definitions cannot be reconciled, show both periods with a visible break. Do not splice them into one seamless line. Where possible, run a short parallel period in which the old and new measurement methods observe the same underlying outcomes. Reconcile against a more stable business source such as orders, payments or completed bookings, while checking that the business source itself has not changed.
Use a tracking incident decision log
A useful log records the change, its owner, the reason, the effective date, the affected events and channels, the expected reporting effect, the validation checks and the decision rule for using the new data. Include the measured gap against the source of truth and any unresolved difference. Keep a copy of the definitions that were live before and after deployment.
For a performance review, label the evidence as comparable, directionally comparable or not comparable. “Comparable” means the events, coverage, values and timing are sufficiently consistent for the decision being made. “Directionally comparable” means the signal may still help, but a known break limits precision. “Not comparable” means the series should not be used to claim a change without another source or test.
These labels are judgement calls, so explain the reason. A 12% gap may be acceptable for a broad budget decision in one business and fatal for a small campaign test in another. The right standard depends on how much money is at stake and how sensitive the recommendation is to the measurement difference.
A tracking repair does not prove that reported performance improved. A fall in attributed sales after a repair does not prove the channel got worse either. First establish what the system now counts, when the change took effect and whether customer outcomes moved in the same way.
If your investment decisions depend on numbers that changed meaning midstream, GV Marketing can help rebuild the comparison before recommending what to scale or stop.