Setup runs in three phases: design match rules, configure reconciliation, and run the first materialization. Plan iteration; the first rule set rarely matches business reality on day one.
- Inventory source-system identifiers
Before writing rules, list every identifier each source system has: email, phone, member ID, login ID, hashed identifiers. Coordinate with each source-system owner to confirm what identifiers are reliable enough to match on.
- Build the match rule set
Data Cloud, Identity Resolution, New. For each source-pair, add a match rule. Start with high-confidence exact-match rules (email, phone) and add fuzzy rules (name plus DOB) as needed. Order rules by confidence; high-confidence rules run first.
- Configure reconciliation rules
For each Unified Individual attribute (FirstName, EmailAddress, City), pick how conflicts resolve. Most Recent is the common default; Source Priority works when one source is canonically trusted (CRM beats marketing platform).
- Run the initial materialization
Trigger the Identity Resolution job. Wait for completion (can take hours on large source data). Review the unified Individual count versus expected count; large gaps indicate match rules need adjustment.
- Audit and iterate
Sample unified Individual records. Verify the right source records merged. Look for false merges (different people merged together) and missed merges (same person not merged). Tune rules and reprocess.
Exact match, normalized match, fuzzy match. Each has different precision and recall trade-offs.
Most Recent (latest update wins), Source Priority (configured source order), Custom (per-field logic). Pick per attribute.
Hourly, every 6 hours, daily, on-demand. Match to data freshness need and credit budget.
Standard normalizations for email, phone, name. Custom normalizations for industry-specific identifiers.
- Fuzzy match rules risk false merges. Two people with the same name and birth date merge incorrectly without a stronger disambiguator.
- Reconciliation choice affects every unified Individual attribute value. Wrong reconciliation produces a unified profile that contradicts both source records.
- Full reprocessing is expensive but sometimes necessary after major rule changes. Plan during maintenance windows.
- Unified Individual IDs are not guaranteed stable across reprocessing. Downstream systems that bookmark IDs need a reconciliation strategy.
- Match quality is measurable but easy to ignore. Sample audits catch false and missed merges before they cascade.