41% of companies have implemented a CDP. Only 22% report high utilization. That's a utilization gap wide enough to park an entire martech budget inside, and it tells you something uncomfortable: the category's problem was never adoption. It was whether the thing actually gets used after procurement signs off.
Databricks announced CustomerLake on June 16 at its Data + AI Summit in San Francisco. It's a warehouse-native CDP built on the Databricks Lakehouse, governed by Unity Catalog, and loaded with autonomous AI agents (Profile Agents, Campaign Agents) designed to run what the company calls "infinity campaigns." HP, Circle K, AB InBev, and Getnet by Santander are in private preview. GA is expected later this year.
The hype is predictable. The real question is whether this product breaks the pattern that's defined CDPs for years: technically interesting, operationally underused.
Why This Isn't Really a CDP Story
Forrester's Joe Stanhope put it well: CustomerLake doesn't fundamentally alter the direction of the CDP market. It accelerates existing trend lines. The shift from Packaged to Composable to Agentic CDPs has been underway, and CustomerLake is Databricks' bet that the third stage is ready for enterprise money.
The more interesting signal is what Databricks is actually selling. CustomerLake is the company's second vertical application after Lakewatch (security, launched March 2026). This isn't a data infrastructure company dabbling in marketing. It's a platform company building opinionated applications on top of its own substrate. That's a different competitive posture than, say, Snowflake partnering with CDP vendors to run composable workloads.
For marketing ops teams, the warehouse-native architecture matters because it means customer data doesn't get copied into a separate system. Governance stays in Unity Catalog. Identity resolution, segmentation, and activation all run where the data already lives. In theory, this reduces pipeline latency and eliminates the governance headaches that come with syncing data across tools.
In theory.
The Utilization Gap Won't Close Itself
Here's the part worth slowing down on. Experts have been arguing for years that CDPs must cover four capabilities to deliver real value: data collection, identity resolution, segmentation and intelligence, and real-time activation. Miss any one of those, and the CDP becomes an expensive data repository.
CustomerLake claims to cover all four, with agents handling the decisioning and orchestration layer. But the critique that keeps surfacing in analyst commentary is pointed: many CDPs fail marketers by optimizing for technical architecture rather than enabling marketers to actually build, preview, and analyze campaigns tied to revenue outcomes. The architecture can be pristine. If the marketer can't figure out whether a campaign moved pipeline, it doesn't matter.
Agentic features alone won't fix organizational barriers, either. The 41%-vs-22% gap isn't purely a software problem. It's a skills problem, a governance problem, an operating model problem. Throwing autonomous agents at a team that hasn't sorted out data contracts or cross-functional handoffs between marketing and data engineering is a recipe for expensive shelfware with fancier branding.
What to Actually Evaluate
If you're a VP of Marketing or CMO weighing CustomerLake (or any agentic CDP), the diligence questions are blunt:
- Warehouse commitment: CustomerLake mandates Databricks at the center. Are you already there, or does this create a new dependency?
- Agent-first readiness: Does your team have the operating model, data quality, and governance to let agents make decisions autonomously? What's the failure mode when an agent gets it wrong?
- Functionality parity: Has CustomerLake met or exceeded what industry-standard CDPs deliver today? It's in private preview. Many claims are forward-looking.
- Revenue attribution: Can you tie "infinity campaigns" to measurable lift? Vendor-stated scale claims like "1:1 personalized experiences a billion times a day" are not independently verified. Demand proof connected to pipeline and retention metrics before committing budget.
The partner ecosystem is broad (Adobe, Meta, Braze, The Trade Desk, LiveRamp, Iterable, Bloomreach, Snapchat, Twilio, IAS). That's a good sign for teams with entrenched stacks. But breadth of integrations and depth of activation are different things. Test the connectors that matter to your GTM motion, not the ones that look good on a slide.
The Hypothesis Worth Testing
With 51% of organizations prioritizing agentic AI investment in the first half of 2026, and Gartner projecting 75% of enterprises operationalizing AI in marketing by year-end, the pressure to move isn't abstract. But "operationalizing" is doing a lot of heavy lifting in that sentence.
The falsifiable hypothesis: if warehouse-native agentic CDPs reduce activation latency and keep governance centralized, then the utilization gap should shrink measurably within 12 months of deployment. If it doesn't, the problem was never the architecture. It was everything around it.
CustomerLake is worth watching precisely because it's the first product that puts all the agentic CDP promises into a single, opinionated package backed by a platform company with real distribution. But "worth watching" and "worth buying" are separated by exactly one thing: evidence that it moves revenue, not just data.
The CDP market is projected somewhere between $4B and $10B in 2026, depending on who's counting. Composable vendors are growing employment at 7.8% versus 1.3% for the industry average. The structural shift is real. Whether Databricks' bet on agents closes the gap between CDP adoption and CDP value is the question that matters. And right now, nobody has the answer. Not Databricks, not analysts, and definitely not the conference keynote.
