Marketing reporting reconciliation is not a hunt for one “correct” dashboard. Different systems may count different events, use different clocks, apply different attribution rules or expose different row samples. The audit becomes useful when it explains which question each report answers and where the evidence chain changes.
1. Define the decision and reporting boundary
Write the decision the reconciliation must support: change budget, repair tracking, accept a month, compare channels or explain a pipeline variance. Name the sources, date range, currency, timezone, campaign family, conversion definition and CRM stage in scope. Do not start by copying totals into a spreadsheet without their definitions.
Separate observed events, attributed conversions, qualified leads, opportunities and realized value. A report can be valid for media optimization while being unsuitable for revenue forecasting. Put each measure in its own column and state the owner of the definition.
2. Record metric definitions and grain
For every source, record row grain, primary key, event time, processing time, attribution window, filters, exclusions, currency and refresh latency. Note whether a number is a count of events, users, sessions, records or distinct organizations.
Google explains that Analytics conversions and key events have distinct reporting roles. Use that distinction as a reminder that naming a column “conversions” does not prove that two systems count the same object. Copy the exact field or report label into the reconciliation log.
3. Align clocks before arithmetic
List click, session, event, lead-created, accepted, opportunity-created and closed timestamps. Check timezone, daylight-saving behavior, late-arriving events, import delay and date-of-report versus date-of-event. A month-end difference can be a timing difference rather than a missing record.
Create a small known-record sample and sort it by every clock. Keep open and mature records separate. If an event arrives after the report cutoff, mark it late rather than moving it silently into the earlier period. Retain the as-of timestamp for every extract.
Write the cutoff in the report title or export note. If one system backfills yesterday while another freezes at midnight, both can be internally consistent. The audit should show the observation window, not just the date label that a dashboard happens to display.
4. Compare identifiers and joins
Map campaign ID, source, medium, click ID, session ID, form ID, CRM record ID, account ID and opportunity ID. Document transformations and fields that are dropped. A timestamp-only join can be useful for a diagnostic sample, but it should be labelled probabilistic.
Trace one synthetic or approved record through landing, event, lead, qualification and opportunity. Capture accepted, rejected, duplicate and unmatched states. The objective is to find where the chain changes, not to make the sample look complete.
5. Reconcile attribution rules
Write the attribution model, lookback window, conversion counting, deduplication rule and eligibility conditions for each platform. Keep first touch, latest touch, source of record and assisted influence in separate views. Do not compare a platform-reported credit number with a CRM-created count as if they were one measure.
If the business needs a planning model, calculate it from a declared rule and label it derived. Preserve raw platform values beside derived values. A clean reconciliation should make disagreement explainable, not hide it behind a blended percentage.
6. Inspect the reporting source and refresh path
Google notes that Search Console data can differ from other tools and that reports have different coverage and sampling behavior. Use the same caution for every connected system. Record property, view, API extract, filter, permission and refresh job.
Compare a raw export with the dashboard for a small date range. Remove filters one at a time and identify the first change in total. Check whether one report uses canonical URLs, another uses landing URLs, and a third uses CRM source fields.
7. Build an exception-first reconciliation table
| Difference | Candidate explanation | Evidence | Safe action | | — | — | — | — | | platform count higher | repeated events or wider window | IDs and settings | deduplicate or label | | CRM count higher | untracked or offline intake | lead source and import log | classify source | | dashboard lower than export | filter, permission or refresh | config and timestamps | repair view | | month-end gap | late processing | event and load times | wait or restate | | revenue missing | open or rejected stage | CRM history | exclude from mature value | | channel disagreement | different attribution | model and window | compare by declared rule |
Assign an owner, severity and next evidence date to every exception. An unresolved row is preferable to an invented explanation.
8. Run a bounded reconciliation pilot
Choose one campaign family, one conversion and a short period. Freeze definitions, export raw rows, calculate derived views and compare the same records across tools. Keep a control report unchanged. Do not combine this pilot with a tag migration, CRM stage redesign or budget increase.
Use Search Console performance reporting only for the questions it can answer: impressions, clicks, CTR and page/query patterns. Join it to downstream evidence only when the join rule is explicit. Report valid, unmatched, duplicate, late and unknown rows separately.
9. Close with a decision log
State which totals are comparable, which are not, why each difference exists, what remains unknown and who owns the repair. Include source extracts, report settings, clock definitions, identifier map, exception table and rollback notes.
Approve a budget or reporting change only for the part of the chain that passed. If the data contract is unclear, fix the definitions before scaling activity. Reconciliation is successful when a future reviewer can reproduce the explanation without relying on the author’s memory.
Keep the raw extracts, formulas and decision log together. When a definition changes, start a new version rather than overwriting the old comparison. That prevents a later chart from erasing why an earlier number differed.
How did this article land?
Choose one reaction. You can change it anytime.
