Databricks announced CustomerLake on June 16 at its Data + AI Summit: an agentic Customer Data Platform (CDP) built natively inside the Databricks lakehouse. The move takes the data-warehouse vendor squarely into marketing infrastructure, and the implications for measurement teams go well beyond another CDP launch.
Databricks CustomerLake is an agentic CDP embedded in the Databricks lakehouse that uses AI agents to build Customer-360 profiles, construct audiences, and activate them across ad platforms, replacing periodic campaign execution with continuous, automated engagement.
What the Agents Actually Do
CustomerLake ships with two agent types. Profile Agents ingest raw customer data and resolve identities using what Databricks calls “Agentic Identity Resolution (AIR),” producing business-ready Customer-360 profiles without leaving the lakehouse. Campaign Agents take those profiles, build audiences, select next-best actions, and push activations to connected platforms. The partner list at launch includes Meta Audience and Conversions API, The Trade Desk, LiveRamp, Integral Ad Science, Adobe, Braze, Snapchat, and Twilio, among others.
Databricks frames the output as “1:1 personalized experiences a billion times a day” and describes the resulting engagement model as “Infinity campaigns.” Both phrases are Databricks’ own characterization of the system’s intended scale, not independently measured results. The product is currently in Private Preview, with no general availability date announced.
The Pricing Logic Is the Real Story
Traditional CDP vendors charge a platform fee on top of the data infrastructure you already own. CustomerLake drops that layer. Databricks monetizes the compute and storage underneath, offering what it describes as a “consumption model.” For enterprises already running on Databricks, the incremental cost of adding CustomerLake is theoretically just the additional query and compute load. Adweek framed the positioning as a challenger to legacy CDPs, noting Databricks is entering a market long dominated by purpose-built SaaS vendors.
Named early customers include HP, Circle K, AB InBev’s Zé Delivery unit, and Getnet by Santander. Databricks co-founder and CEO Ali Ghodsi put it this way:
“When customer data, AI models, and agents live in one governed platform, marketing stops being a series of campaigns and becomes a continuous loop.”
— Ali Ghodsi, Co-Founder & CEO, Databricks
What Changes for Measurement Teams
The measurement question is concrete. An agent that fires audience segments to Meta CAPI at the scale Databricks describes generates a volume and velocity of activations that no human analyst reviews in real time. That shifts responsibility upstream: the governance, the consent signals, and the attribution logic must be correct before the agent runs, not audited after the fact.
On the UTM side, the same pressure applies. Agent-fired campaigns that push to paid channels still land users on URLs. If the activation pipeline does not attach structured, traceable parameters to every link it generates, the traffic shows up in GA4 as direct or unattributed. The practical fix is straightforward: any team integrating an agentic activation layer needs to ensure their parameter schema is locked before the first agent fires. Tools that let you build clean UTM parameters become a contract, not an afterthought, when the campaign volume is autonomous and continuous.
Attribution gets harder in a different way. When Campaign Agents continuously optimize across Meta CAPI, The Trade Desk, and LiveRamp simultaneously, the question of which touchpoint drove conversion is no longer a reporting exercise. It becomes a question of which agent decision contributed what. GA4’s existing channel framework is already struggling with that complexity, as the GA4 source group attribution model shows: consolidating fragmented traffic from Facebook variants and AI referrers is a recent patch, not a solved architecture.
Consent Is Not an Agent Problem
CustomerLake lives inside a governed platform by design. It is not clear from the launch materials, however, how CustomerLake handles consent-signal propagation when activating to Meta CAPI or The Trade Desk. Those integrations require that consent state travel with the signal. At the activation scale Databricks describes, any gap in consent plumbing is not a compliance edge case. It is the dominant outcome.
The product is in Private Preview, so the documentation on consent handling may still be incomplete. That is worth tracking before any production deployment. Databricks’ full announcement is on the Databricks blog, with the press release at the Databricks newsroom.
The unit of work in the data stack is shifting from the dashboard to the agent. Whether the measurement infrastructure under it is ready is a different question.
