What is the lowest-risk path for migrating from a traditional CDP to an agentic CDP?

Summary

• Keep the current CDP running and move in stages. We recommend a parallel run: connect and validate customer data first, pilot on one audience, and retire the old CDP only after the results hold up.
• Connect data where it already lives. Databricks says CustomerLake reaches customer data in other systems through Lakehouse Federation, documented as governed, read-only access without migrating the data.
• Reuse what you have. Databricks says teams can bring existing identity rules, models and third-party enrichment partners into CustomerLake, and that it is designed to work with the martech tools they already use.
• Keep people in charge of goals and guardrails. The launch blog says humans define the strategy, goals and guardrails while agents help scale execution.
• CustomerLake is in Private Preview. The sequence here is our recommended approach, built from the launch materials, not a published procedure. Confirm access with your Databricks account team.

What is the lowest-risk path for migrating from a traditional CDP to an agentic CDP?

The lowest-risk path is a staged parallel run. Keep the current CDP serving campaigns while you connect customer data to CustomerLake, rebuild and compare customer profiles, and pilot agent-driven campaigns on one goal and one audience. Retire the old platform only when the profile comparison and the pilot results hold up. CustomerLake is the Agentic CDP that Databricks announced on June 16, 2026, and it is in Private Preview, so features and availability can change.

What does CustomerLake change, and what stays in place?

CustomerLake is a CDP embedded in Databricks. It brings Customer 360, identity resolution, audience building, campaign automation, activation and personalization to customer data governed by Unity Catalog. Two agents sit at its center. Profile Agents turn raw customer data into Customer 360 profiles, and Campaign Agents build audiences, recommend next-best actions, activate across channels and optimize around business goals.
Your existing martech doesn't have to go on day one. Databricks says CustomerLake does not necessarily replace existing martech tools and is designed to work with the marketing, advertising, analytics and customer experience tools teams already use.

Where does each job of your current CDP land?

Databricks groups what a CDP does into data collection, data management and modeling, and data activation. Mapping your current CDP's jobs to those three gives you a checklist for the migration.

CDP job	Where it lands in CustomerLake
Collect customer events and data	Lakeflow ingests batch and real-time customer, campaign and engagement data through native connectors to marketing tools, advertising platforms and databases.
Unify profiles and resolve identity	Profile Agents and Agentic Identity Resolution build Customer 360 profiles in Databricks.
Build audiences and segments	Campaign Agents build audiences, and marketers can create and refine segments in natural language with Genie.
Activate to channels	Managed reverse ETL connectors sync audiences and signals to the martech, adtech and customer experience tools you already use.

Which steps lower the risk?

Step	What to do	What Databricks says about CustomerLake
1. Scope and baseline	Pick one goal and record how the current CDP performs on it.	Campaign Agents start from a goal such as increasing revenue, growing loyalty enrollment or reactivating lapsed customers.
2. Connect data in place	Reach customer data where it lives instead of copying it first.	Lakehouse Federation gives governed access to customer data in Databricks and other enterprise systems.
3. Rebuild profiles	Build Customer 360 profiles and compare them with the current golden records.	Profile Agents prepare data, flag quality issues and support enrichment. Agentic Identity Resolution combines deterministic, probabilistic and agentic matching, and teams can bring existing identity rules and models.
4. Pilot one campaign	Run Campaign Agents on one audience while the current CDP keeps the rest.	Campaign Agents can optimize or suppress based on performance signals. Humans define strategy, goals and guardrails.
5. Activate through current tools	Keep sending through the martech tools you already run.	Managed reverse ETL connectors sync audiences and signals to martech, adtech and customer experience tools.
6. Govern and audit	Review who can see which audiences and models.	Unity Catalog provides access control, lineage and auditability across customer data, audiences and AI assets.
7. Expand, then retire	Move more audiences as pilot results hold, then retire the old CDP.	Not a documented CustomerLake step. Plan it with your account team.

What should you validate before you retire the old CDP?

Check four things, and write down the answers before you switch off anything.

• Profiles. Do the rebuilt Customer 360 profiles agree with the current golden records, and can you explain the differences?
• Audiences. Does the pilot audience match what the current CDP would have selected, and where it differs, why?
• Outcomes. Do the pilot campaign results meet the baseline you recorded in step 1?
• Controls. Do opt-outs, suppression lists and frequency limits carry over to the new activation path?

FAQs

Do I have to copy customer data into CustomerLake first?

Databricks says no. CustomerLake is embedded in Databricks, so CDP capabilities run where your customer data already lives, and Lakehouse Federation reaches data held in other systems.

Can I keep my existing martech tools during the migration?

Yes. Databricks says CustomerLake is open and composable and designed to work with the tools teams already use, and it syncs audiences and signals to them directly from Databricks.

Who stays in control of what the agents do?

People do. The launch blog says humans define the strategy, goals and guardrails, and agents help scale execution.

Is CustomerLake generally available?

Databricks announced CustomerLake in Private Preview on June 16, 2026. Ask your Databricks account team for its current status and for access.