Keep cleaned CRM records from being recreated by imports and integrations by using stable source identifiers and reviewing ambiguous matches.
A team spends a week reviewing duplicate contacts before connecting an AI assistant to the CRM. The customer history looks coherent on Friday. The overnight partner import then runs, fails to match an address that changed in July and creates a second record. By Monday, the assistant can see an old conversation on one contact and a current contract on another.
During CRM deduplication, find out how the duplicate originally got there. Removal is one job. Preventing the same import from creating it again is another. Look at the connector’s matching keys, what it does with missing data and how it decides to create. Use a few known duplicate histories to test the receiving system before giving the assistant access to it.
Use a key a source can keep
Ask each input owner what identifier uniquely names a contact in that system and whether the identifier remains stable after an email change. A stable key must include the source context if IDs from different systems can overlap. For example, customer 802 in an event system is a different identifier from customer 802 in a store. Store that mapping alongside the CRM record where an authorized integration can retrieve it. Keep a record of which feed supplied the mapping if a reviewer needs to trace its origin.
HubSpot’s contacts API guide documents contact operations using a custom unique identifier with the idProperty option. Confirm which operations and properties your actual connector supports. If a single CRM property cannot represent several external identities cleanly, an integration can maintain a mapping with one entry per source and external ID. Start with the import that recreates duplicates most often. It is easier to validate one stable link than to change every source at once.
Define the decision when a key cannot be used
An import should have a defined route when its stable identifier is missing or unfamiliar. Older customers may have only an email address on file. A role inbox can belong to several people over time. Have an administrator decide where a secondary match is reliable and where it needs review. A customer name alone is usually a poor basis for creating or combining identities. For uncertain matches, give a reviewer the evidence and allow the record to wait.
| Incoming record | What the connector can establish | Proposed action |
|---|---|---|
| Known source ID with changed email | Existing CRM mapping names one contact | Update the mapped contact under the field rules |
| New source ID with a shared role address | Email may represent more than one person | Place the candidate in a review queue |
| No source ID and a familiar display name | Name provides weak evidence | Ask the source owner for a stable key before automatic creation |
| Former contact ID on a delayed event | The identifier may have an authorized survivor mapping | Resolve to that record or hold if the mapping is ambiguous |
| Previously unseen valid ID with no plausible existing match | Creation criteria are met | Create once and save the source mapping with the result |
After merging records, retain authorized aliases for integrations that may still send old IDs. A mapping should point to the final survivor in a way the connector can resolve. Avoid relying on a chain of manual redirects that nobody checks. Keep enough lineage for staff to spot a bad association and to tell a later writer why the event was routed to that contact.
Reproduce the duplication with test data
Take one import row similar to a row that recreated a contact. In a test environment, link it to an existing contact with a stable source ID. Change the email address in the test source and run the import. Verify that the intended contact updates. Then replay the same row and check how many records and associations exist. The second run should leave a single contact for that source identity, with any changes traceable to the incoming row.
Next, remove the source ID and use a shared inbox. Check who receives the review item and what evidence the reviewer sees. Run a separate test using two different sources that both send the same numeric ID. The connector should maintain separate source contexts unless a reviewer has established that both identities belong to the same person. Finally, try a delayed event for a former record. Verify which contact the event updates and whether the operator can trace the route.
Let operations see what keeps coming back
Track new contacts by source during the pilot and compare against a normal week. Review how many records were matched on a stable key, held for evidence or created. When staff find a duplicate, tie it back to the first event that created each record. That tells the team whether to fix an intake rule, a mapping or a source export. Hold a brief review with the owner of any feed that regularly sends identities the CRM cannot interpret.
Keep an eye on the review queue after the assistant launches. A delay may matter when a caller is trying to open an urgent case. The assistant can hand over the observed account details and the reason the match needs confirmation. It should leave the pending records available to an authorized reviewer. Add one alert for a sudden rise in unrecognized source keys, which can indicate that a partner changed its export or rotated identifiers.
Also test two arrivals at the same time. When a partner sends repeated events after a delay, both may look new until the first write completes. Use the supported unique key or an application constraint to let a single mapping win, and ensure the following event can read the result. If a CRM write succeeds while recording the mapping fails, make that incomplete operation visible to the owner. The owner should be able to link the existing contact from confirmed evidence, without creating a replacement simply to make the next import run.
Emerging Stacks can design matching and intake rules through Zoho CRM services and full-stack development. Contact Emerging Stacks with one recurring duplicate and the integration that recreates it.
Ready to work with us?
Get in Touch


