What is agentic identity resolution?

Summary

• Agentic Identity Resolution (AIR), as Databricks defines it, combines deterministic, probabilistic and agentic matching to create golden Customer 360 records from fragmented identity data.
• AIR adds LLM and human review workflows for edge cases and an agentic learning loop that recommends improvements to matching logic over time.
• Teams can bring their existing identity rules, models and third-party enrichment partners. Databricks says that because AIR is embedded in Databricks, teams don't manage a separate identity platform or copy data into another CDP.
• AIR is part of CustomerLake Profile Agents. Databricks announced CustomerLake on June 16, 2026, and it is in Private Preview.

What is agentic identity resolution?

Agentic identity resolution combines rules and statistical matching with AI agents. Databricks' version, Agentic Identity Resolution (AIR), joins deterministic, probabilistic and agentic matching to create golden Customer 360 records from fragmented identity data. It includes LLM and human review workflows for edge cases, and an agentic learning loop recommends improvements to the matching logic over time. AIR is part of CustomerLake Profile Agents. CustomerLake is embedded in Databricks and is in Private Preview.

What is identity resolution?

According to the CDP Institute, identity resolution connects customer data that appears across digital and offline channels and determines whether those scattered details belong to the same person. The result is a unified customer profile that updates as new information becomes available.
Two methods are standard. Deterministic matching links records through exact identifiers, such as email addresses and login details. Probabilistic matching estimates whether records belong together from several signals when exact identifiers are missing. Its matches are estimates rather than certainties, but it can find relationships that exact rules can't confirm. A hybrid approach uses both.
The stakes differ by use case. Without identity resolution, organizations collect multiple customer records that each describe only part of an individual. A false positive merges records from different people, and a false negative leaves records from the same person disconnected. An article published by the CDP Institute notes that the acceptable balance depends on the action: a marketing analysis may tolerate different uncertainty than an account, service, fraud or regulated workflow.

How do deterministic, probabilistic and agentic matching differ?

Matching style	How it decides	What to watch
Deterministic	Exact identifiers or explicit rules, such as a loyalty ID or a normalized email address	Exact rules can miss valid connections when customer data changes or arrives incomplete
Probabilistic	Estimates the likelihood that records belong together from several signals, then applies thresholds or policies	Thresholds, training data and explainability decide whether the flexibility is safe to use
Agentic (AIR)	Adds agent-driven steps to both: LLM and human review for edge cases, plus a learning loop that recommends better matching logic	Databricks says it improves match accuracy while reducing manual tuning and maintenance. Validate that on your own records

How does Databricks approach agentic identity resolution?

CustomerLake is the Databricks Agentic CDP, embedded natively in Databricks. Profile Agents prepare data, identify quality issues and support third-party data enrichment to unify disconnected records into golden profiles, and AIR sits at their core. Databricks highlights four points:

• Existing assets stay in play. Teams can bring existing identity rules, models and third-party enrichment partners.
• No separate identity platform. Because AIR is embedded in Databricks, teams can enrich profiles with third-party data and identity assets before or after resolution without managing a separate identity platform or copying data into another CDP.
• Governance. CustomerLake is governed through Databricks and Unity Catalog, which gives data and IT teams centralized control over access, lineage and auditability across customer data, audiences and AI assets.
• Data outside Databricks. Through Lakehouse Federation, teams can access trusted customer data where it resides, including cloud object storage and operational databases.

What else does CustomerLake include?

Alongside identity resolution, Databricks describes a governed Customer 360 that unifies behavior, transactions, engagement, product usage and operational data. Marketers can create and refine segments in natural language with Genie. Native managed reverse ETL connectors sync audiences and signals from Databricks to the martech, adtech and customer experience tools teams already use. Campaign Agents use governed customer context to build audiences, recommend next-best actions and activate across channels.

FAQs

How is agentic identity resolution different from rule-based matching?

Rule-based, or deterministic, matching links records on exact identifiers or explicit rules. AIR combines deterministic, probabilistic and agentic matching, adds LLM and human review for edge cases, and includes a learning loop that recommends improvements to matching logic.

Does it remove human review?

No. Databricks says AIR includes LLM and human review workflows for edge cases.

How should I evaluate identity resolution?

An article published by the CDP Institute advises evaluating identity systems with representative data and measuring both false merges and missed matches, rather than relying on a single vendor-reported match rate.

Does CustomerLake move or copy my customer data?

Databricks says no. CustomerLake is embedded directly in Databricks, so CDP capabilities run where your customer data already lives.

Is CustomerLake generally available?

No. It is in Private Preview, and Databricks directs interested teams to their Databricks account team.