In this guide
Two reports can describe the same campaign and show different totals. Identify what each system counts, when it counts it and which events it excludes. Treat the difference as an investigation, rather than immediate proof that one platform is wrong.
Compare the same thing
Align identifiers, placements, dates, time zones, currency and metric definitions. A served impression is not necessarily viewable. A click is not a session. A report with a different attribution window answers a different question.
Write these definitions at the top of your reconciliation sheet. This prevents a discussion about totals in which each person describes a different event.
Consider the counting point
Advertising and analytics systems can record activity at different stages. A user might click an ad but leave before analytics runs. Consent choices, browser settings, redirects and blocking software also affect which systems observe an event.
Google's explanation of Analytics and Ad Manager discrepancies describes examples of differing measurement conditions. Understand the reporting contract rather than expecting every total to be interchangeable.
Wait for the reporting window
Data can arrive late, be filtered or be revised. Compare reports at a consistent point after the period and note when exports were created. A live dashboard and final billing report may have different completeness expectations.
Avoid a universal acceptable discrepancy percentage. Agree an investigation threshold that fits the contract, systems and scale of the campaign.
Break the difference into parts
Reconcile by date, placement, creative, device or another shared dimension. Find where the pattern changes. A difference isolated to one creative suggests a different starting point from a gap across all placements.
Distinguish absolute and relative gaps. Ten missing events out of 20 is a large percentage with a small count; 100,000 out of 10 million is a small percentage with a substantial count.
Make a useful escalation
Include definitions, aligned filters, export times, affected identifiers and a small example. Describe when the issue began and what changed. A screenshot of a total without context rarely helps a colleague reproduce the problem.
Finish with the conclusion the evidence supports. If the cause is uncertain, say what was checked and what evidence is needed next. That is more useful than forcing an explanation to fit the numbers.
Keep learning
Look up a term in the glossary →
