Marketing automation data quality is not a cleanliness score. A database can have complete fields and still send the wrong message, route a request to no owner, or suppress a customer incorrectly. Diagnose the path from consent and identity to automation, handoff, and mature outcome before buying a new tool or adding another workflow.
1. Define the quality decision
Write what the review must change: field standard, integration, audience, automation, suppression, routing, report, or owner. Name the object, segment, channel, date range, and outcome. Keep “field is populated” separate from “field is accurate and actionable.”
2. Sample actual failures
Collect duplicate, missing, stale, invalid, conflicting, wrong-owner, suppressed, and incorrectly messaged records. Record object ID, source, field history, automation version, timestamp, consent state, and outcome. Redact data not needed for the diagnosis.
Do not inspect only the cleanest records. Sample accepted, rejected, and unknown records so the data contract reflects real edge cases.
3. Map objects and identity
List contact, company, account, request, opportunity, subscription, campaign, activity, and customer objects. Define identifiers, relationships, merge rules, source of truth, and what happens when a key is missing. HubSpot’s data model builder can support an object review; it does not define your business rules.
Test shared email, role change, multiple opportunities, partner, existing customer, household, and merged records. An email address is not automatically a person, company, or buying unit.
4. Audit field definitions and values
Create a dictionary with meaning, type, allowed values, owner, collection point, normalization, required status, expiry, and privacy class. Identify free text, copied values, timezone mismatch, outdated industry, invalid location, and fields changed by automation.
Add unknown, not applicable, and not collected states. Replacing missing data with “other” hides the reason a workflow fails.
5. Check consent and suppression
Record consent source, purpose, timestamp, channel, region, unsubscribe, bounce, complaint, preference, and suppression owner. Test a contact who unsubscribes after enrollment, a hard bounce, a customer preference change, and a consent withdrawal during a batch.
Google’s Gmail sender guidelines are a deliverability boundary for senders. They do not replace local privacy and legal review. Stop if the system cannot show why a recipient was eligible.
6. Inspect automation and handoff rules
For each workflow record trigger, conditions, priority, enrollment, re-entry, delay, update, notification, owner, error alert, and exit. Look for contradictory automations, hidden lists, stale branches, race conditions, and manual overrides.
Trace Marketing to Sales, customer success, support, and partner queues. Record response SLA, disposition, capacity, and fallback. A data-quality defect may actually be a handoff or serviceability defect.
Check whether automation writes a value back into the same field that a person uses for qualification. Record precedence, timestamp, actor, and rollback for each write. A field that alternates between an import, a workflow, and a manual edit can look inconsistent even when every individual process is behaving as configured. Decide which system is authoritative and how conflicts are shown.
7. Reconcile reports and events
Compare eligible, enrolled, sent, delivered, suppressed, bounced, clicked, replied, routed, accepted, qualified, opportunity, and mature outcome counts. Use the same cohort and date logic. Keep duplicate, unknown, and late states visible.
For website interactions, GA4 event guidance can define the event layer. An event does not prove data accuracy, consent, qualification, or revenue. Join only through an approved key and state the attribution limit.
8. Repair one bounded contract
Choose one field, object relationship, automation, suppression path, integration, or report. Snapshot current values, rules, lists, access, and outcomes. Repair it in a test environment, replay synthetic records, then observe a controlled live cohort.
Stop if a vendor silently transforms fields, access is excessive, privacy cannot be reconstructed, the receiving team has no capacity, or the change would message suppressed contacts. Preserve the prior version and rollback. Run the repair through a sample of records that should enter, should be excluded, should be routed elsewhere, and should remain unknown. Ask the receiving owner to confirm the outcome and record whether the data was actionable without manual reconstruction. Preserve the sample and the response in the quality register so the same failure is not reclassified as a new issue next month. Add the field definition and owner to the same register so the fix remains understandable after the workflow changes. Add a date for the next sample and an escalation path if the failure reappears. Review that sample at the same meeting as the workflow owner and report owner, not only after a deliverability incident. Archive the decision with the field dictionary.
Before closing the incident, compare the repaired cohort with the original sample and note whether the error rate, exception path, and owner response changed. If the evidence is incomplete, keep the issue open with a date and named reviewer rather than declaring the data clean.
9. Apply the data-quality gate
| Gate | Required evidence | Hold if | | — | — | — | | meaning | dictionary, owner, allowed values, unknown | field name is the only definition | | identity | stable keys, relationships, merge rules | email is an undocumented universal key | | consent | source, purpose, preference, suppression | eligibility is assumed | | automation | priority, delay, exit, error alert | workflows conflict | | handoff | owner, SLA, capacity, disposition | queue has no responder | | reporting | cohort, event, stage, outcome, limits | filled field is called quality | | rollback | versioned rules and safe replay | change cannot be reversed |
Data quality is ready when a reviewer can explain why a record entered a workflow, what it received, who owned the handoff, and what outcome followed. Improve the contract before increasing automation volume.
How did this article land?
Choose one reaction. You can change it anytime.
