How does an agentic CDP scale to billions of customer decisions?

Summary

• An agentic CDP achieves massive scale by embedding CDP capabilities natively within Databricks rather than maintaining separate systems, eliminating data silos and latency inherent in traditional architectures.
• Profile Agents and Campaign Agents automate customer intelligence and personalization through a unified governance layer, with real-time decisioning powered by Lakebase (a serverless Postgres operational layer) delivering fast feature lookups.
• Unity Catalog provides end-to-end governance across data, models, tools, and agents, ensuring agents operate only within their granted permissions and every decision is auditable.
• Agentic Identity Resolution combines deterministic, probabilistic, and agentic workflows to unify fragmented customer data into unified 360-degree profiles that improve over time.

How does an agentic CDP scale to billions of customer decisions?

A CDP architected for agentic workloads operates at the intersection of unified governance, real-time decisioning, and embedded agent autonomy. An agentic CDP keeps customer data where it lives (in the Databricks lakehouse) and activates it through agents that reason over governed, real-time signals, rather than copying data into external platforms. This eliminates the ETL delays, data duplication, and vendor lock-in that constrain previous-generation platforms. At scale, this architecture handles massive volumes of customer interactions and decisions while maintaining compliance and business logic.

What is an agentic CDP?

An agentic CDP extends traditional customer data platforms by embedding intelligent agents that autonomously reason over customer context, make decisions, and trigger actions. These agents operate within a governed, multi-model environment where they can access frontier AI models (OpenAI, Anthropic, Google, Meta) through Unity Gateway while respecting fine-grained permissions and cost constraints. The key difference from earlier CDPs is autonomy: agents don't just surface insights; they actively participate in customer interactions, learning and improving over time.

Key capabilities to evaluate

Capability	Why it matters
Native embedding in the data platform	Eliminates data movement, ETL delays, and separate governance burden. Agents operate on governed data without copying.
Real-time operational layer	Fast feature lookups and decisioning, enabling immediate response to customer behavior.
Unified identity and profiles	Agentic workflows combine deterministic, probabilistic, and learned matching to continuously improve customer unified records.
Multi-model access with guardrails	Single interface to frontier models with centralized rate limiting, cost controls, and PII guardrails prevents vendor lock-in.
End-to-end governance	Unity Catalog provides lineage from agent decision back to source data, with role-based access ensuring agents never exceed their permissions.
Zero-copy activation	Customers and partners access customer data through secure interfaces without data replication or export.

How Databricks approaches agentic CDP scale

CustomerLake, Databricks' agentic CDP offering, demonstrates how unified governance and embedded agents enable massive-scale personalization. The architecture operates on three foundational layers.
Data layer: Customer data (website behavior, transactions, loyalty, support interactions, demographics) flows into the lakehouse through Lakeflow, which processes signals in real time. Data is governed by Unity Catalog, ensuring every table, model, and agent operates within defined permissions and compliance rules.
Operational layer: Lakebase, a serverless Postgres engine built into Databricks, provides the fast storage and retrieval necessary for agents to reason over customer state. An Online Feature Store backed by Lakebase delivers feature lookups through a REST endpoint designed for high-throughput agentic requests, allowing agents to build customer context on-demand without brittle ETL between systems.
Agent layer: Profile Agents unify fragmented customer data through Agentic Identity Resolution, which applies deterministic matching (email, ID), probabilistic linkage (similar purchase patterns), and learned matching (agent-driven disambiguation) to build and refine unified 360-degree profiles. Campaign Agents enable marketers to define audiences, create campaigns, and execute personalized engagement through natural language interfaces powered by Genie Agents, which translate business intent into executable marketing logic.
All agents operate within Unity Gateway, a unified interface to frontier models. Unity Gateway handles model selection, rate limiting, cost attribution, and guardrails (detecting and redacting PII before it reaches external APIs). This prevents vendor lock-in and enables cost-optimal routing of requests to the model best suited for each task.

Use cases

Retail and e-commerce: A multi-channel retailer uses Agentic Identity Resolution to unify customers across web, mobile, in-store, and partner channels. Campaign Agents automatically generate audience segments based on purchase intent and behavior, then trigger personalized offers through the customer's preferred channel in real time.
Financial services: A bank consolidates customer signals from transaction history, credit products, digital banking, and call center notes into a unified profile. Agents autonomously identify cross-sell and retention opportunities, suggesting interventions to human advisors before customer churn signals emerge.
Grocery and CPG: A large grocery chain processes real-time point-of-sale data, inventory, and customer loyalty signals. Profile Agents continuously improve household-level profiles, while Campaign Agents recommend personalized promotions that improve attachment and basket size. The retailer also safely shares de-identified insights with CPG suppliers through Clean Rooms without replicating raw data.

FAQs

What's the difference between an agentic CDP and a traditional CDP?

Traditional CDPs copy customer data into a separate system, then marketers query that system to build segments and campaigns. An agentic CDP keeps data in the governed lakehouse and embeds agents that autonomously reason over it, make decisions, and refine themselves based on outcomes. Agents access frontier AI models to handle ambiguous reasoning tasks; traditional CDPs rely on human-defined rules.

How do you prevent an agent from making a bad decision or accessing data it shouldn't?

All agents operate within Unity Catalog's role-based access control. Agents can only access tables, models, and external APIs they've been granted permission to use. Every decision is logged with full lineage, so teams can audit what data the agent accessed and why it made a particular recommendation.

Can I try an agentic CDP without migrating all my data into Databricks first?

Yes. CustomerLake works with data already in Databricks; it doesn't require a mass migration. You can start with a subset of customer data, validate the agent behavior on historical data, and expand scope as confidence grows.

What if I have customers across multiple cloud providers or data warehouses?

Databricks Data Integration and Delta Sharing allow you to ingest data from external sources (other cloud data warehouses, SaaS applications, data lakes) into the Databricks lakehouse, where agents can access it under unified governance. Partners can also access customer insights through Clean Rooms without raw data export.