How can I connect anonymous digital behavior with known customer profiles?

Summary

• Connecting anonymous digital behavior to a known customer is identity resolution, also called identity stitching: matching web and app sessions, device identifiers, and cookies to a real person's email, account, or loyalty record.
• The dependable approach combines deterministic matching on known identifiers with probabilistic matching on behavioral and device signals, then reviews the uncertain matches rather than guessing.
• Resolving identity where your customer data already lives, under one governance and consent model, avoids copying sensitive data into a separate identity tool.
• Databricks CustomerLake is an agentic CDP embedded in the lakehouse; its Profile Agents use Agentic Identity Resolution to stitch anonymous signals into golden customer profiles governed by Unity Catalog.
• CustomerLake is in Private Preview, so confirm current availability before you design around a specific capability.

How can I connect anonymous digital behavior with known customer profiles?

You connect anonymous behavior to a known customer through identity resolution: matching the signals a visitor leaves before they identify themselves (web and app sessions, device identifiers, cookies, clickstream events) to the records you already hold (email, account, loyalty, order history). Do it well and a returning visitor's anonymous activity joins their real profile, so personalization and measurement reflect the whole person. The reliable way to get there is to resolve identity where your customer data already lives, under one set of permissions and consent rules, rather than shipping raw behavioral data into a separate system.

What is identity resolution?

Identity resolution, or identity stitching, is the process of deciding that many disconnected records and signals refer to the same person, then merging them into one profile. It draws on two kinds of evidence. Deterministic matching uses exact matches on stable identifiers such as email, phone, or account ID. Probabilistic matching infers likely links from behavioral and contextual signals, such as shared devices, sessions, and cookies, when a hard identifier is missing. Anonymous visitors are the hard part: they arrive with device and session signals but no login, so the system links them by inference and confirms the link once they authenticate.

Key capabilities to evaluate

Capability	Why it matters
Deterministic and probabilistic matching	Exact-match rules plus signal-based inference catch links that either method alone would miss.
Edge-case handling	Uncertain matches should route to automated and human review, not a silent, incorrect merge.
Runs on governed data	Resolving identity where the data already lives keeps one permission and consent model and avoids extra copies.
First- and third-party inputs	Blending owned data with external identity and enrichment data raises coverage of anonymous visitors.
Golden profile output	One resolved record per person becomes the source of truth for segmentation, personalization, and measurement.
Consent and lineage	Each merge and attribute should carry consent status and lineage so privacy and audit hold up.
Freshness	Anonymous-to-known links should update as new sessions arrive, not wait for a slow nightly batch.

How Databricks approaches it with CustomerLake

CustomerLake is Databricks' agentic CDP, embedded in the lakehouse so identity resolution runs where customer data already sits and inherits security and lineage from Unity Catalog. Its Profile Agents turn fragmented source data into unified golden customer profiles without copying data out.
The matching engine is Agentic Identity Resolution, and it works in layers. Rules handle high-confidence matches on known identifiers. A model layer resolves probable matches from behavioral and device signals that rules miss. For matches outside existing rules, a fallback LLM review proposes a decision, and genuine edge cases route to human review. The agent then suggests rule updates based on those model and human decisions, so accuracy improves over time. You can bring your own ML models, and pull third-party identity graph and enrichment data through a data marketplace to resolve visitors your first-party data can't place on its own.
The output is a golden customer profile: anonymous web, app, and device signals stitched to a known person, with duplicate records collapsed into one governed record. Consent and preference data travel with the profile, and because everything runs under Unity Catalog, access to sensitive attributes follows the same permissions whether a person or an agent reads them. Data lands through Lakeflow Connect, identity pipelines run on Lakeflow Pipelines, and profiles are stored as Delta tables. Marketers can then explore the resolved data in plain language with Genie. For background, see What is an agentic CDP? and how unified profiles feed customer segmentation. CustomerLake is in Private Preview, so confirm availability before you build around a specific capability.

Use cases

• Retail and e-commerce. A shopper browses anonymously on a phone, then signs in. Their earlier session links to their loyalty profile, so offers reflect full history rather than treating them as a new visitor.
• Media and subscription. Anonymous views across several devices resolve to one person or household, giving a consistent experience and cleaner reach and frequency for measurement.
• Financial services. A prospect researches products without logging in, then authenticates to apply. The pre-login activity attaches to the known applicant, while restricted attributes stay masked from anyone not cleared to see them.

FAQs

What does it mean to connect anonymous behavior with known profiles?

It means identity resolution: linking pre-login signals like device IDs, cookies, and session activity to a known record such as an email or loyalty account, then merging them into one profile.

How does CustomerLake match anonymous sessions to a real person?

Agentic Identity Resolution combines deterministic rules on known identifiers, probabilistic matching on behavioral and device signals, and an agentic layer that reviews uncertain matches with LLM and human review, learning better rules over time.

Do I have to copy customer data into a separate identity tool?

No. CustomerLake resolves identity inside the lakehouse where the data already lives, so you keep one governance and consent model and avoid extra copies.

How is customer consent respected during matching?

Consent and preference records travel with the profile, and Unity Catalog applies the same access, masking, and lineage rules to every attribute, whether a person or an agent reads it.

Is CustomerLake generally available?

CustomerLake is in Private Preview. Confirm the current status of Profile Agents and Agentic Identity Resolution before relying on them.