Duplicate Property Detection Playbook

A duplicate property exists when one physical hotel is represented by more than one canonical entity. The symptom is usually discovered indirectly through split offers, duplicated result cards, inconsistent reviews or conflicting supplier mappings.

Trigger

Near-identical hotels appear in results, offers are split across entities or provider mappings point to two canonical IDs.

Candidate generation

Use normalized name, geo distance, address/postal code, phone, official website, provider cross-references and brand.

Decision tree

Large geo distance rejects merge. Strong geo/name/address agreement increases confidence. Conflicting phone/brand or separate-building evidence requires review.

Merge safety

Measure impact on offers, reviews, images and mappings before merge. Keep merge reversible and audited. Preserve both source identities after the merge so late supplier updates can still be routed to the surviving canonical entity.

Failure modes

False merges are more damaging than temporary duplicates. Nearby sister properties, apartment buildings with shared addresses and chain hotels with similar names should be treated as explicit contradiction cases.

Metrics

Track candidates, sampled auto-merge precision, merge reversals, unresolved duplicates and duplicate search impressions.

Prevention

Run candidate matching before creating a new canonical ID.