Human-Reviewed AI Entity Matching for Dealer Customer Records

The Same Person, Household, and Vehicle Are Different Things

An AI matching tool may suggest that two records describe the same person, but similarity is not identity. A model can overvalue a shared surname, address, or email and erase a separate customer’s preference or service history. If the dealership uses an approved tool, pass only the fields needed for a candidate match, keep the source record IDs, and require a human to review the evidence. Never let a generated confidence score trigger an irreversible merge by itself. The review outcome and reason belong in the data-governance record.

A dealer CRM may contain several records for one person, two people sharing an address, a business vehicle, and a family with more than one vehicle. Treating those cases as one duplicate problem can erase useful context or send a message to the wrong person. Start by defining the entities the system must represent: individual customer, household or business relationship, vehicle, opportunity, repair order, and communication preference. A clean identity model helps staff see the relevant history without assuming that similar names or addresses prove the records belong together.

Set Evidence Rules for Matching

Start with a review queue that favors false negatives over false positives when the cost of a mistaken merge is high. A duplicate that remains separate can be investigated again; a mistaken merge can join service history, communication choices, and sales notes that belong to different people. Show reviewers the fields that caused the candidate match and let them record why it was accepted or rejected. Reuse those reasons to improve data capture at intake, where a missing apartment number, shared family email, or nickname may be easier to clarify than after years of activity.

Use a match hierarchy that reflects the risk of a mistaken merge. A verified customer identifier or confirmed contact may be strong evidence; a shared phone number, address, or surname is only a clue. Treat recycled phone numbers, shared email inboxes, and misspelled names carefully. Create statuses such as likely duplicate, confirmed duplicate, possible household relationship, and do not merge. Give a data steward or manager authority to review ambiguous matches. Automated suggestions can reduce the queue, but a person should approve merges that alter customer history or communication eligibility.

Protect History During a Merge

Before merging, capture the source record IDs, ownership, active opportunities, service history references, vehicle associations, preferences, notes, and consent evidence. Choose a surviving record according to a documented rule, then preserve the losing ID as a reference if the platform allows it. Verify that open tasks, appointments, forms, call history, and source attribution remain connected. Never overwrite one person’s preference with another’s simply because the records share an address. If the CRM cannot preserve a field safely, stop the merge and involve the platform owner instead of exporting a manual replacement.

• Define individual, household, business, vehicle, opportunity, and preference records before deduplication.
• Use evidence thresholds and a review queue for uncertain matches rather than forcing every candidate together.
• Snapshot record IDs, active work, history, source, and preferences before a merge.
• Test lead routing, service reminders, appointment tasks, and opt-outs after a sample merge.
• Report merge volume, rejected matches, unresolved duplicates, and customer-impacting exceptions monthly.

Make Household Relationships Useful, Not Promotional Shortcuts

A household relationship can help staff understand shared vehicles or a preferred contact path, but it should not become an excuse to market to every person at an address. Record who consented to which channel and who owns a conversation. Let customers correct names, relationships, vehicles, and preferences through a controlled process. For business accounts, identify the operational contact and the entity that owns the vehicle or contract. Those distinctions matter when service, sales, parts, billing, or privacy questions arise.

Measure Data Quality Through Customer Work

Do not judge the cleanup by the number of records removed. Monitor duplicate lead rate, failed contact tasks, conflicting preferences, missing vehicle links, repeated service questions, and staff corrections after the merge. Sample a few customer journeys from inquiry to appointment or repair order to confirm that the identity model helps rather than hides history. A careful CRM cleanup is ongoing governance. It reduces avoidable confusion for customers and employees while preserving the evidence marketing needs to understand how a relationship actually developed.