Quick answer
Choose a source of truth for each stage: website activity, submitted lead, qualified lead, opportunity and customer. Define the event, timestamp, deduplication rule, identifier and owner for each stage, then reconcile totals using those definitions.
Expect differences when attribution windows, time zones, consent, identity matching, event counting or offline updates differ. Investigate unexplained changes, but do not label every mismatch a tracking failure.
What I would check first
Confirm that one real submission creates one CRM record and one intended analytics event. Check duplicate forms, thank-you-page reloads, cross-domain journeys, missing identifiers and the difference between lead creation time and qualification time.
Compare a small sample of records end to end before building a dashboard. A record-level trace often reveals mapping and timing errors that aggregate totals conceal.
Decision framework
Separate collection problems from attribution differences and business-process differences. Collection problems include missing or duplicated events; attribution differences include different models or windows; process differences include a lead being qualified later than the original submission.
Document acceptable reconciliation ranges and escalation rules. The target is a trustworthy explanation of the data, not a forced equality between systems that answer different questions.
Implementation steps
Create a metric dictionary and a field map across the CRM, GA4 and ad platforms. Standardize event names and status values, preserve campaign context, and record how offline outcomes are imported or reported.
Test the reconciliation after website releases, CRM changes and tag updates. Google recommends configuring cross-domain measurement when a journey spans related domains; otherwise, a checkout or booking handoff can split one journey into separate users or sessions.
Common mistakes
Do not sum platform conversions as though they represent unique customers, use form submission as a proxy for every later stage or change definitions without annotating reports.
Do not export more personal information than the workflow requires. Access, retention, consent and secure handling should be part of the measurement design.
Measurement and business implications
Track match rate, duplicate rate, missing-stage rate, reconciliation variance and the time between lead creation and final outcome. Use a small exception report to investigate the cases that matter most.
Connect this process to the offline conversion import, consent-mode QA and digital reporting articles so the business can optimize for meaningful outcomes without overstating precision.