CDP in Transition: Why Customer Data Will Be Activated Where It Resides in the Future

On June 16, 2026, at its Data + AI Summit, Databricks presented its own Customer Data Platform, CustomerLake — natively embedded in the Lakehouse, driven by AI agents and currently in private preview, with Getnet by Santander as an early adopter from the financial sector. One can read the announcement as just another product in a crowded market. It is more interesting as a symptom: CDP capabilities are moving to where the customer data already resides. For marketing and data executives in banks, a closer look is worthwhile — because the architectural question increasingly decides speed, governance, and ultimately marketing results.

From copying data to activating it in place

For a decade, Customer Data Platforms followed a common basic pattern: data from all source systems was copied into a proprietary storage of the platform. Identity resolution, profiling, and segmentation took place there; subsequently, target groups were synchronized into the activation channels. This model has brought much to the industry — but it structurally creates a second data copy with its own governance, its own latency, and its own costs.

The countermovement goes by terms like "composable" or "warehouse-native": the CDP capabilities run directly on the company's data warehouse or lakehouse, and the data remains in place. Open table formats reinforce the trend because the same physical files can be read by analytics, activation, and BI tools. Profiling close to the data platform reduces latency and prevents segment logic from diverging in copied tool silos.

What CustomerLake concretely does differently

Databricks consistently implements this idea. Customer data is not copied into a separate CDP database, but remains in the lakehouse and is subject to the same governance as all other data in the company. Two agent families take over the work that was previously passed back and forth between marketing and data teams: Profile Agents create Customer 360 profiles, combining deterministic, probabilistic, and AI-powered identity resolution; Campaign Agents build target groups and orchestrate activation. Through federation mechanisms, the platform can also read data in other systems without moving it.

It is also remarkable what CustomerLake does not attempt to do: replace the activation landscape. Connecting established marketing tools is part of the concept. The CDP thus becomes less of a place and more of a layer of capabilities over the existing data infrastructure.

The market is reorganizing

The announcement comes at a time when the CDP market is visibly regrouping. In the Gartner Magic Quadrant for Customer Data Platforms 2026, Salesforce is the only remaining Leader, while Hightouch, a warehouse-native provider, has newly entered the quadrant — a clear signal that composable architecture has arrived in the mainstream. At the same time, the overall market is growing strongly: market researchers from Mordor Intelligence value it at around 4.6 billion US dollars for 2026 and expect it to more than triple by 2031. It is therefore not a matter of displacement in a shrinking field, but of which architecture will absorb the growth.

Where the limits of the approach lie

As clear as the trend is, it is hardly a universal answer. Three limitations deserve attention:

• Maturity of the data platform: Those who do not operate a viable warehouse or lakehouse with clean pipelines have gained little from a composable CDP. Packaged platforms remain the faster path to initial results for many organizations.

• Identity resolution as a core capability: In some organizations, the identity resolution of the existing CDP is actually foundational. Replacing it without an equivalent substitute compromises profile quality. Hybrid transition models may make more sense here than a hard cut.

• Data location is not an end in itself: Optimizing for "zero copy" does not replace a strategy. The decisive factor is whether better decisions and more relevant customer interactions emerge from the profiles — not in which system the tables are located.

Evaluation checklist

Those deciding on their own CDP architecture in the coming months can quickly reach a solid assessment with five questions:

1. Where does the customer data really reside today? The more consolidated the data base is in the warehouse or lakehouse, the more the starting point speaks for a composable approach.

2. Who resolves identities — and how well? The quality of identity resolution should be measured before deciding on its location.

3. What latency do the use cases require? Real-time triggers present different requirements than weekly campaign selections.

4. How many copies of customer data exist — and who governs them? Every copy is a governance and cost factor that belongs in the overall calculation.

5. Which teams should work with the profiles? An architecture that commits data and marketing teams to the same data repository reduces coordination effort — but requires shared processes.

The answers vary depending on the organization. This is precisely why the choice of architecture is a management decision and not a pure tool question.

What this means for banks

For financial institutions, the architectural question has a special nuance, as almost no other industry is subject to stricter requirements regarding data storage, access control, and traceability. When customer profiles are created where encryption, authorization concepts, and audit processes already apply, it reduces duplicate work in governance and lowers the risk of shadow data repositories. At the same time, this brings the data platform closer to the business departments: segments, scores, and next-best-action logic can run on the same governed data as risk and reporting processes.

However, the actual value creation only occurs one level above — in the decision-making logic along the customer lifecycle. An architecture that leaves data in place creates the prerequisite for making personalization, campaign management, and sales impulses faster and more consistent. But it does not replace the question of which customer signals an institution wants to detect, what actions follow, and how their impact is measured.

For the priority list, this means: first clarity on the desired outcomes in the customer lifecycle, then clear responsibilities for identity, consent, and measurement — and only on this basis the architectural decision. The market is recognizably moving towards the lakehouse. Those who leverage the movement instead of just following it will connect the new architecture with a resilient decision layer for customer management.