Skip to main content

3-year roadmap to evolve from a traditional CDP to a composable CDP architecture

Summary

  • A three-year move from a traditional CDP to a composable architecture runs in three stages: build a governed customer data foundation, move audiences and activation onto it, then retire the legacy CDP and automate engagement.
  • Keep the existing CDP serving live campaigns until each use case reaches parity on the new stack, so no campaign depends on an unfinished system.
  • Connect activation to your channels early. A profile layer that can't reach email, ads or apps doesn't let you switch anything off.
  • On Databricks, CustomerLake is the agentic CDP embedded natively in the platform and governed by Unity Catalog. Campaign Agents are in Beta, Profile Agents are in Private Preview, and the two can be adopted independently.

3-year roadmap to evolve from a traditional CDP to a composable CDP architecture

Spend year one building a governed customer data foundation on your own data platform, year two moving audience building and activation onto it while the old CDP keeps running, and year three cutting over, retiring the legacy CDP and automating engagement. The order matters: data and identity first, activation second, retirement last, with a parity check at each gate. A composable CDP makes your data platform the system of record for customers, so this roadmap is really a migration of where customer truth lives. On Databricks, the end state can go one step further, with CustomerLake running CDP capabilities inside the platform itself.

What is a composable CDP?

A traditional CDP is an all-in-one bundle: it collects customer data into its own storage and models it around predefined users and events. A composable CDP puts your existing data infrastructure, such as a lakehouse, at the center instead. Ingestion, identity resolution, modeling, segmentation, activation and governance become layers over one governed copy of customer data. That lets you model whatever entities the business needs, such as households, apply one set of access controls, and run analytics and ML on the same tables marketing uses.

The three-year roadmap

Phase Goal Key moves Gate to the next phase
Year 1: Foundation One governed copy of customer data Inventory sources and live campaigns; ingest CRM, commerce, event and service data; define the customer data model, identity rules and access policies Unified profiles reconcile with the legacy CDP for priority segments
Year 2: Audiences and activation Marketers work from the new foundation Rebuild priority audiences; connect owned and paid channels; move cohort campaigns first, then one-to-one owned email; set up measurement Migrated campaigns run on the new stack and match the legacy baseline
Year 3: Cutover and automation Retire the legacy CDP Move the remaining journeys, including latency-sensitive paid media; shut off feeds into the old CDP; add next-best-action and always-on engagement Legacy CDP decommissioned; every audience traces to governed source tables

Guardrails that keep the migration safe

  • Run in parallel. The legacy CDP keeps serving live campaigns until each use case has moved and been checked against it.
  • Activate early, not last. Programs that defer activation integrations can end up with finished segments and nowhere to send them, and segments that can't reach a channel don't let you retire anything.
  • Version audience definitions. When a segment's logic changes during migration, have the campaign owner re-approve it.
  • Plan for the operating model. A composable CDP needs in-house data engineering, data stewardship for identity rules, and marketing operations for audiences. Budget for people and process, not only licenses.

How Databricks approaches the composable CDP

Foundation (Year 1). Lakeflow Connect managed connectors ingest from SaaS applications and from databases using change data capture, and they produce ingestion pipelines governed by Unity Catalog. Lakehouse Federation queries external databases and catalogs without migrating the data, which helps while sources are still being moved. Unity Catalog captures lineage down to the column level, so you can show where regulated customer data originates and which assets consume it.
CustomerLake, the CDP embedded in Databricks. CustomerLake is "an Agentic CDP embedded natively within Databricks" (product page), governed by Unity Catalog, and it works without copying sensitive data into a separate CDP (introducing CustomerLake). It has two parts:

  • Profile Agents (Private Preview) turn raw customer data into Customer 360 profiles directly in Databricks. They use deterministic, probabilistic and LLM-based matching, golden-record survivorship rules, incremental updates as data changes, and data steward overrides. This is the Year 1 identity layer.
  • Campaign Agents (Beta) let marketers build audiences in natural language with Genie, sync them to engagement platforms, and move from one-off campaigns to always-on engagement with next-best-action recommendations. Activation runs through native integrations and Reverse ETL with bi-directional pipelines. This covers Years 2 and 3.

Profile Agents and Campaign Agents can be adopted independently, and teams can start with campaign workflows on their existing customer data (CustomerLake enters Beta). CustomerLake is designed to work with the marketing, advertising, analytics and customer experience tools you already use, so it can join the roadmap at whichever phase you've reached.

FAQs

Do I have to switch off my current CDP at the start?

No. Keep it serving live campaigns while you build the foundation and rebuild audiences, then retire it use case by use case once each one reaches parity.

Should I start with profiles or with campaigns?

Either works. On Databricks, Profile Agents and Campaign Agents can be adopted independently, so start where your current CDP is weakest or where the business case is clearest.

Is CustomerLake generally available?

No. CustomerLake entered Beta in October 2026: Campaign Agents are in Beta and Profile Agents are in Private Preview. Access goes through your Databricks account team.

What skills does a composable CDP need?

Data engineering for ingestion and modeling, data stewardship for identity rules and overrides, and marketing operations for audiences, activation and measurement.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.