Define the records before counting
For each report, record the metric definition, filter, time zone, date basis and source. Identify retries, transfers, automated events, test activity and unresolved matches. Check whether a platform threshold or reporting delay changes which events appear. Preserve original exports so the comparison can be reproduced.
A worked example
Fictional discrepancy table: the ad platform reports eighteen conversions; the phone log contains twenty-four events; review links those to fifteen inquiries; nine jobs complete and seven have linked payment evidence. These figures can coexist because their units and evidence differ. The missing two payment records are unresolved, not automatically unpaid losses.
Resolve the gap in evidence
Choose one question at a time. Reconcile event counts to inquiry IDs, inquiry IDs to job outcomes and job IDs to payments. Document unexplained differences in a queue instead of forcing totals to match. Do not import a synthetic test into the business total or count a transfer as new demand simply because the platform emitted another event.
Check whether it worked
Publish a small bridge explaining each count and the remaining uncertainty. Recheck after the relevant reporting delay and record corrections. If a definition changes, preserve the old version for historical comparison. Use the reconciliation lab to rehearse the method before handling private exports in approved systems.
Put it to work
A record you can use
- Report, metric and unit
- Window, time zone and basis
- Known duplication or exclusions
- Record-level bridge
- Unresolved difference and owner
Plain text. No email required. Store completed records in your approved business systems.
Next useful step
Need help applying this to your marketing? A strategy call starts with your business and the evidence you have.
Talk through your next step →For a general editorial question, email Chris. Please leave out customer details. Questions are private, and no response time is promised.