A CDP that never leaves the warehouse
Zeotap announced on August 5 that its Composable Customer Data Platform is now live on Snowflake Marketplace as a Snowflake Native App, built on Snowpark Container Services. The pitch is architectural rather than feature based: identity resolution, audience segmentation, journey orchestration, and AI models for propensity and churn all execute inside the customer's own Snowflake account. No customer record leaves the account boundary to reach a separate CDP backend, and no reverse ETL job has to run on a schedule to keep a second copy of customer data current somewhere else.
Zeotap founder Projjol Banerjea framed the point directly, saying running the entire platform inside the customer's Snowflake account is how the company keeps customers in control of every record. That answers a specific objection enterprise buyers raise constantly: every CDP historically required copying customer data into a new system, which meant a new attack surface, a new compliance boundary to defend during an audit, and a new vendor holding raw identity data the security team did not choose to trust with it.
Why regulated industries are the target buyer
The capability list reads like a checklist built from procurement objections rather than a marketing brainstorm: deterministic and probabilistic identity matching, SQL-free segmentation with live audience size estimates, real-time journey triggers, and more than 250 activation endpoints for pushing segments out to ad platforms and messaging tools. Consent and governance dashboards are built to map to GDPR, CCPA, and the EU AI Act, which signals who Zeotap expects to close first: banks, insurers, and other regulated data controllers who cannot tolerate an unaccountable copy of customer personal data sitting inside a marketing vendor's own cloud.
AI models for propensity, lifetime value, and look-alike audiences run through Snowflake Cortex rather than a bespoke Zeotap model layer. That choice trades some technical differentiation for something a regulated buyer values more: a security team can audit Cortex the same way it audits every other Snowflake workload already running in production, instead of onboarding an entirely new AI vendor's model governance process, documentation, and risk review from a standing start.
Snowflake needs wins like this one
Snowflake VP for EMEA Partners and Alliances Dan Waters called Zeotap's approach a strong addition to the composable ecosystem, the language Snowflake has used throughout 2026 to describe its answer to Databricks on data application strategy. Native Apps let Snowflake claim application-layer revenue and stickiness without building a CDP itself, while giving customers one invoice, one identity model, and one governance boundary to defend to auditors instead of stitching together evidence from two or three separate vendor contracts every audit cycle.
For Snowflake, every Native App that handles a workload as sensitive as customer identity is a proof point that the platform can host more than analytics queries. That matters commercially, because the harder it becomes to argue a workload needs to leave Snowflake for a specialized system, the harder it gets for a customer to justify a second platform contract at renewal time, and the stickier Snowflake's own negotiating position becomes across every account running this pattern.
Composable CDPs finally close the last gap
Composable CDPs built on top of the warehouse have been pitched since at least 2022 by vendors like Hightouch and Census, arguing that reverse ETL made a standalone CDP unnecessary. Zeotap's move goes a step further: it skips syncing data to the warehouse entirely and deploys the whole application layer as a containerized workload that never requests an export in the first place. That closes the gap earlier composable CDP vendors left open, where the application logic itself still ran somewhere outside the warehouse even after the data stopped moving.
The timing lines up with a broader shift in 2026 buying behavior, where CIOs increasingly ask vendors to prove data locality before they will even evaluate a feature list. Snowflake's own Cortex and Native Apps investments over the past year exist partly to give vendors like Zeotap a credible answer to that question, and partly to keep more of the AI compute spend flowing through Snowflake's own billing relationship rather than a partner's separate invoice.
What this means for the CDP replacement cycle
Enterprises running legacy CDPs from Adobe, Salesforce, or Segment are mid-cycle on renewal decisions that increasingly hinge on where AI workloads and customer data intersect. A CDP that runs inside the warehouse removes an entire category of vendor risk assessment work: there is no separate SOC 2 report to chase down, no separate breach notification chain to negotiate into a contract, and no separate data processing agreement covering a second copy of records the security team already fought to minimize elsewhere.
The tradeoff CTOs should scrutinize is lock-in of a different kind than the one they are used to evaluating. A CDP embedded in Snowflake ties customer data architecture to a single warehouse vendor's roadmap and pricing decisions for years. Teams should model the real cost of a future warehouse migration under this pattern before assuming the governance win is free, because Native App architectures are not built to be portable across warehouses by design, and that constraint compounds every year the deployment stays in place.
The procurement conversation this actually changes
The practical shift for a CTO running a CDP evaluation in the second half of 2026 is where the conversation starts. Instead of opening with a feature comparison matrix, the first question worth asking every CDP vendor on a shortlist is where customer records physically live during activation, and how many hops the data takes between ingestion and an ad platform or messaging tool receiving a segment. Vendors that cannot answer precisely, or that answer with a diagram showing three intermediate systems, are the ones carrying the compliance risk a security team will eventually have to underwrite.
That question also reframes how to think about build versus buy for CDP functionality generally. A composable, in-warehouse CDP narrows the gap between buying a packaged product and building activation logic in-house on top of an existing Snowflake investment, since much of the integration risk a custom build would carry is already handled by the platform itself. For data teams that already run identity resolution or segmentation logic natively in Snowflake, evaluating Zeotap against an internal build is now a more honest comparison than it was when every CDP option meant exporting data first.
