As data and agentic systems reshape customer engagement, organizations need ways to access secure, governed data in real time. Zero Copy deployment may help, but it raises questions around technology, data ownership, governance and turning data into action.
Organizations have turned to Customer Data Platforms (CDP) and marketing technologies to deliver more personalized customer experiences for years. Now, agentic artificial intelligence (AI) systems that reason and act across data in real time have raised the stakes—making integration with existing enterprise data lakes and warehouses more critical than ever. These agentic architectures don’t just consume data; they depend on it being clean, governed and instantly accessible.
As needs evolve, organizations tend to move through two distinct phases. The first is simply getting access to the right data by consolidating customer information from disparate sources to power personalized experiences. The second is about the conditions of that access: speed, governance, security and cost. Real-time performance, privacy controls and cost management have fundamentally changed what “good” looks like.
This is where Zero Copy deployment methods enter the picture. They change how CDPs access enterprise data, often directly from the warehouse or lake, and—depending on implementation—what data persists in the CDP. As more organizations explore this approach as a potential deployment model, it is important to recognize that not all implementations are the same. There are multiple ways to adopt Zero Copy, each with different implications for how data is accessed, governed and ultimately put to work.
Zero Copy should therefore be evaluated not simply as a CDP feature, but as an enterprise architecture and operating model decision. It changes where customer data is processed, where controls must be enforced, and which teams own the work required to make activation possible. In practice, its success depends as much on alignment across marketing, data, security and IT as on the CDP itself.
So as organizations pursue greater return on data investments with faster time-to-market, tighter cost control and simplified governance, the real question becomes: Does Zero Copy data architecture actually bring you closer to those outcomes?
Zero Copy vs. composable
Zero Copy and composable are often used interchangeably because both are commonly discussed in warehouse-native environments, but they describe different things.
Three data movement paradigms
Zero Copy is best understood in contrast to the traditional CDP data model. For the past decade, CDPs have generally followed the same pattern: moving data from source systems into the CDP’s lake, transforming and unifying it there, building customer profiles from it, and then activating those profiles across channels. This approach works, but it creates multiple copies of data and requires replicating data integration work that may have already been done elsewhere.
In response, newer architectures have sought to reduce data movement and rely more directly on governed enterprise data platforms. Zero Copy is one of these patterns, but it is not a single monolithic approach. We see three distinct data movement paradigms, distinguished not by whether data moves at all, but by what persists inside the CDP versus what remains in the warehouse or lake:
These aren't “better” or “worse” approaches for data movement; they're optimized for different organizational contexts and priorities. The question isn't which paradigm is correct, but which aligns with an organization’s data infrastructure, capabilities and objectives.
These paradigms are not mutually exclusive. Different data types can, and often should, move differently. Real-time data, such as active customer profiles used in personalization, may justify full ingestion and persistence in the CDP to meet latency requirements. Other datasets that don't require real-time availability can remain in the warehouse and be queried on demand. Organizations can draw from more than one paradigm, applying each where it fits.
The importance of alignment becomes clearer at the technology layer, where more than one architectural approach can support Zero Copy. Those underlying foundations shape how the model works in practice and how well it fits within a given enterprise architecture. These differences become visible through three common implementation patterns: data sharing, query federation and push-down query.
Data sharing technologies enable multiple systems to access the same physical data through standardized formats, eliminating duplication.
Apache Iceberg provides a table format for large-scale data sharing across data lakes. It specifies how to organize data files and metadata so multiple systems can read them efficiently. When a CDP uses Iceberg-based Zero Copy, it’s reading through this shared format. Data physically resides in cloud storage, but Iceberg provides the queryable structure. In this model, the CDP creates virtual data lake objects, such as pointers with metadata, rather than copying underlying data.
Delta Lake emerged with similar goals but a different table architecture. It prevents partial updates and adds versioning on top of data lakes, so only completed changes are visible. CDPs built on Delta Lake can query external data platforms without replication. The main difference from Iceberg lies in table change tracking and tool usage: Both make data lakes more reliable, but they track table changes differently and are used with different sets of tools.
Query federation represents a different approach where the CDP issues queries to multiple source systems like an organization’s warehouse, CRM, transaction database and then, stitches results together in its own environment. The CDP doesn’t maintain complete source table copies, but does create its own representation of combined results. This provides flexibility and works with systems that don’t support Iceberg or Delta Lake, though the “Zero Copy” claim becomes nuanced since aggregated data does move and get stored.
Push-down query, common in composable CDPs, keeps all heavy lifting in the data warehouse. When a segment is built or analysis run, the system translates the request into SQL and pushes that query to the warehouse. The warehouse processes it using its own compute resources, and only results (such as a list of customer IDs) return to the activation platform. This keeps nearly everything in one place but requires support for real-time or near-real-time queries.
The practical implication is that an organization’s data platform and CDP choice need to align. When the platform and CDP are built on different architectural patterns, integration becomes more complex. The “Zero Copy” label alone does not indicate whether a solution will work smoothly with existing infrastructure; the underlying technology paradigm still matters.
Hidden considerations: What the vendor materials don’t emphasize
Vendor presentations about Zero Copy emphasize speed, cost savings and simplified architecture. These benefits can be real, but several important considerations receive less attention.
SECURITY APPROVAL PROCESSES MAY ACTUALLY BECOME MORE COMPLEX.
Zero Copy requires opening new access patterns—a CDP needs direct connections to an organization’s data warehouse, which means configuring network access, opening ports and creating authentication mechanisms. Security teams often prefer well-understood patterns like encrypted file transfers over opening database connections that allow external systems to query internal data stores.
The approval process involves demonstrating exactly what data the CDP can access, how queries are controlled, what derived data may be created or retained within the CDP, what prevents unauthorized extraction, and how access and downstream data movement get audited. Organizations that assumed Zero Copy would simplify security approval sometimes find the opposite—at least initially. Engage security organizations early and budget realistic time for reviews.
THE COST EQUATION IS MORE NUANCED THAN “WE'LL SAVE ON STORAGE.“
Reducing duplicate data storage saves money, but Zero Copy shifts costs rather than eliminating them. Data warehouses now handle query load for CDP operations. Depending on query patterns and data volumes, compute costs can be substantial. Cloud providers also charge for data egress—moving data out of their environment. The economics depend on specific architectures, data volumes and usage patterns.
The point isn’t that Zero Copy is more expensive, because it often isn’t. The point is that the cost model changes, so it’s important to analyze specific scenarios rather than assuming automatic savings. Organizations that measure carefully sometimes find the financial benefit smaller than expected, with real value coming from governance, control or time-to-value improvements instead.
DATA PREPARATION REQUIREMENTS ARE NON-NEGOTIABLE AND OFTEN UNDERESTIMATED.
Organizations consistently discover their warehouse data isn’t ready for CDP consumption. The warehouse exists and contains customer data, but that data hasn’t been modeled for marketing activation. Tables are organized for analytics. Identity resolution is incomplete. Data quality issues acceptable for business intelligence become problematic for real-time activation.
Before Zero Copy works effectively, customer data must be prepared: records unified across source systems, identity resolution completed and data organized into structures the CDP can efficiently query. In traditional CDP implementations, much of this happens within the CDP platform. With Zero Copy, it often needs to happen in a warehouse first, using data engineering teams.
This also shifts more responsibility to the enterprise data organization. Ownership of customer data models, identity logic, quality thresholds and policy enforcement must be defined upstream rather than assumed to sit within the CDP. As a result, Zero Copy changes both the technology architecture and the distribution of responsibility for customer data capabilities.
The timeline for realizing Zero Copy benefits depends more on this data preparation work than on the CDP implementation itself. Organizations with clean, unified, marketing-ready customer data in their warehouse move quickly. Those that don’t may find Zero Copy takes longer than expected because of prerequisite data work.
Practical recommendations for executives
Zero Copy is an enterprise architecture and operating model decision, and realizing its value requires deliberate choices across architecture, governance, security, cost and organizational readiness. Executives should weigh these key considerations when evaluating whether and how to adopt Zero Copy.
Adopt Zero Copy with a direct link to a core business priority, like eliminating redundant storage costs, accelerating use-case delivery, or strengthening compliance through centralized data governance. Without a clear outcome, the architectural complexity isn’t justified.
Select a CDP that aligns with the warehouse’s underlying technology and access patterns. Exact parity is not always necessary, but misalignment in formats, connectors or query models can create friction that undermines Zero Copy benefits. Confirm compatibility early based on the organization’s architecture and priority use cases.
Plan for full security reviews around new query access, authentication and monitoring requirements. Zero Copy changes the risk profile. Ensure all controls are reimagined, not just inherited from previous patterns. Budget time for this upfront.
Assess storage savings versus projected increases in compute and data egress. Base projections on campaign level query and activation patterns unique to your organization. Model realistic scenarios before assuming cost reduction.
Prioritize agility, faster go-to-market and improved data governance as key measures for Zero Copy returns. Cost savings are valuable but only one part of the strategic upside. Set success metrics that reflect the full value proposition.
About the authors
David Chan is a managing director at Deloitte Digital in the Customer Marketing Data, Performance & Optimization practice, where he helps clients in retail, consumer products, telecom and technology modernize their organizations through real-time personalization, customer data platforms and first-party data strategies.
Syed Mir is a managing director at Deloitte Digital in the Customer Marketing Data, Performance & Optimization practice, where he helps companies use agentic AI, MarTech, analytics and data to create personalized customer experiences that increase retention, loyalty and profitability.
Dominic Sagar is an associate vice president and business architect at Deloitte Digital in the Customer Marketing Data, Performance & Optimization practice, where he helps organizations apply AI, machine learning and analytics to their data to make better, more informed business decisions.
Molly Enbysk is a manager at Deloitte Digital in the Customer Marketing Data, Performance & Optimization practice, where she brings together MarTech, AdTech and brand strategy to help retail and consumer brands unify their customer data and turn a clear vision for the future into marketing activation.
Sebastián Toledo is a senior consultant at Deloitte Digital in the Customer Marketing Data, Performance & Optimization practice, where he helps life sciences and health care companies advance their marketing strategy, technology and AI transformation.
Himanshu Hooda is a manager at Deloitte Digital, where he helps organizations define and execute customer data and marketing technology strategies. He specializes in customer data platforms (CDP), business intelligence and data analytics, primarily serving retail clients with additional experience across financial services, life sciences and health care.
