AI Just Hollowed Out the MarTech Stack. Winners Are Racing to the Last Mile.

The composable CDP market is collapsing in on itself. Databricks launched CustomerLake, an agentic CDP built natively inside its data lakehouse. The day before, Hightouch, the leading composable CDP, announced its own Agentic CDP. BlueConic acquired Blueshift. Fivetran absorbed Census. And most CMOs are still buying from a market that's folding underneath them.

How we got here

Every CDP forced brands to do the same thing first: clean, standardize, and centralize their customer data in a cloud data warehouse. A CDP layered on top of messy, fragmented data was useless. So brands spent years building that foundation.

The ones who started early are now years ahead. L'Oréal has run its Consumer 360 and CRM on the Databricks Lakehouse since 2019. Etienne Bertin, the company's Group CIO, has referred to the partnership as the thing that makes L'Oréal "the front-runner in the world of Beauty Tech." Across retail and beauty, that same pattern has played out at scale. e.l.f. Beauty turned its six-million-member Beauty Squad into what its leadership calls a real-time data engine. Ulta Beauty calls its loyalty file "the richest dataset in all of beauty," with roughly 95% of sales flowing through more than 44 million members.

The foundation is built. The standardized first-party data that CDPs once promised to unify now lives inside Snowflake, Databricks, and BigQuery.

And the warehouse noticed.

The turn

Databricks made the move the composable CDP market had been bracing for. CustomerLake does identity resolution, audience segmentation, and campaign activation natively inside the lakehouse. No middleware; no data duplication; no separate vendor. Tasso Argyros, the Databricks VP of Engineering leading CustomerLake, told Adweek that "the CDP, as middleware, is going to go away." He adds that agents are "collapsing the layers" between the warehouse and the execution tools marketers depend on.

The irony runs deep. Argyros is the former CEO of ActionIQ, one of the composable CDPs his product now threatens. His team includes former ActionIQ and Census engineers. Census was acquired by Fivetran. The people declaring standalone CDPs dead are, in many cases, the people who gave them life in the first place.

Hightouch Co-CEO Tejas Manohar, who helped build Segment's warehouse and personas products before founding the company, conceded the warehouse point on launch day: "Welcome to the party, Databricks. We've been waiting years for this." Then he drew a line: "Being the CDP for marketing teams isn't a 'feature,'" his point being that the middle-layer CDP vendors aren't vanishing overnight. Argyros himself notes he's "not saying the vendors are going to go away." But they're being forced to reposition toward execution, verticals, and AI decisioning as the core data-unification job collapses into the warehouse.

Salesforce showed how this structural move plays out when it bundled Data Cloud into its execution stack and pressured independent CDPs. But the comparison only goes so far. Salesforce absorbed the CDP to lock data inside a CRM walled garden. Databricks is running the same play from the opposite direction: an open lakehouse that pushes execution outward to any partner.

That leaves brands in a better position than most of them realize. "Brands should see this as a clarifying moment," says Ryan Willette, Global VP of Customer Success and Partnerships at Attentive. Willette previously served as CRO of Treasure Data and held leadership roles at Agilone, one of the first CDPs on the market. "Customer intelligence and long-term modeling belong in the warehouse. But real-time decisioning, the kind that determines what message hits which customer at what moment, that's an activation problem. Those are different disciplines, and they belong in different layers."

Why CMOs should care about a data platform war

Databricks didn't try to build the execution layer. It partnered with it. The CustomerLake launch named messaging platforms, ad networks, and marketing clouds, including Adobe and Meta, as activation partners. The warehouse owns the data and the intelligence; the execution tools own the last mile. Getting a message to the right person at the right moment, with the right offer, through the right channel, and doing it at scale under compliance constraints, is an engineering problem the warehouse wasn't built to solve. The platforms already doing that work have spent years earning marketer trust, building channel-specific infrastructure, and scaling delivery workflows. That operational depth doesn't ship with a product launch.

Ekta Chopra, Chief Digital & AI Officer at e.l.f. Beauty, captured the stakes in a recent post: "A CMO told me this week: 'We've trained 2,000 marketers on Claude. Our output is up 40 percent. Our impact is up roughly zero.'" Her point: the technology is moving faster than the operating model around it. Applying AI to the old organization produces output, sure, but building the organization that uses AI natively produces outcomes. The same logic applies to the data stack. A warehouse full of clean customer data doesn't generate revenue on its own; the execution layer is where impact lives.

"Brands have spent years getting their data into the warehouse. The question was always going to become: now what? The brands generating real revenue from their data aren't sitting on the cleanest lake. They're activating against it in real time," says Willette.

Abercrombie & Fitch Chief Digital & Technology Officer Samir Desai framed the ownership piece: "98%, maybe 99% of our customers we're interacting with directly. Meaning we have first-party customer data." That direct relationship, combined with a centralized data foundation, is what gives the execution layer something worth acting on.

The platforms built for that last mile are already warehouse-native, and the competitive dynamics are accelerating that. Databricks launching its own CDP sharpens the value of Snowflake's partner ecosystem, where activation platforms with existing native integrations have a head start. Attentive runs a bi-directional data share with Snowflake, pulling profiles and loyalty data in to power campaigns and pushing engagement events back out to the warehouse. The warehouse's move into the CDP layer is a tailwind for that model, not a threat. "The plumbing is consolidating, and that's good news for marketing leaders," says Willette. "Fewer vendors, less duplication, and a clearer line between data and the customer. Every brand going through this transition should be asking one question: is our execution stack built on top of the warehouse, or sitting alongside it?"

The early mover gets the margin

The brands already a step ahead have something in common: they own their data, they've centralized it, and they've invested in the tools that convert it into revenue. Ulta Beauty treats its loyalty data as a competitive moat. DICK'S Sporting Goods rolled out AI-powered personalization at scale through its 25-million-member ScoreCard program earlier this year.

CustomerLake is still in private preview. The decisioning layer is still human-assisted, not fully autonomous. The full impact will take time, but the structural direction is undeniable: the warehouse absorbs the data layer, the middle collapses, and the value that remains pools in the last mile.

The brands that see it coming have time to position for it. The ones that don't will find out from their data team.