How should customer data be modeled and prepared before implementing an agentic CDP?

Summary

• Consolidate all customer data sources (transactional, behavioral, demographic) into a single governed lakehouse rather than copying data into a separate CDP system.
• Establish a unified customer identity through identity resolution before activation, combining deterministic, probabilistic, and agentic matching to create golden profiles.
• Govern all data through Unity Catalog to enforce access controls, audit trails, and lineage across data teams and marketers.
• Prepare for real-time streaming of customer interactions and third-party enrichment, with activation flowing through integration tools without copying data.
• Structure data once in the lakehouse; marketers access trusted datasets through purpose-built interfaces without requiring custom data pulls.

How should customer data be modeled and prepared before implementing an agentic CDP?

Successful agentic CDP implementations begin not with the CDP itself, but with how you organize and govern customer data in your lakehouse. Rather than duplicating data into a proprietary system, the goal is to prepare a unified, governed foundation where customer data stays put and AI capabilities run on top of it.

What is customer data preparation for an agentic CDP?

Customer data preparation is the process of consolidating customer information from multiple sources (transactional systems, web and mobile apps, email platforms, CRM systems, and third-party data providers) into a single governed location. This differs from traditional CDP approaches that copy data into an isolated proprietary database. In an agentic CDP context, preparation means organizing that data so agents can access it with full lineage, governance, and real-time context.

Key capabilities to evaluate

Capability	Why it matters
Unified identity resolution	Without a reliable way to recognize the same customer across systems (web sessions, email, in-store purchases, mobile app), your profiles remain fragmented and AI decisions rest on incomplete context. Identity resolution combines deterministic (exact matches), probabilistic (statistical matching), and agentic (AI-driven) techniques.
Governed data foundation	Marketing agents must operate on data that is accessible, auditable, and compliant. Data governance through enterprise access controls and audit trails prevents unauthorized use and satisfies regulatory requirements.
Real-time data ingestion	Customer interactions (browsing, purchase, support ticket) happen in real time. Preparation must support streaming of behavioral signals so agents can make decisions with current context rather than stale snapshots.
Data lineage and quality	Knowing which system data came from, when it was last refreshed, and whether it meets quality thresholds is essential for agent decision-making and compliance audits.
Activation without data duplication	Integration to email platforms, ad networks, and martech tools should not require copying customer data outside the lakehouse; instead, data flows through controlled integration layers.

How Databricks approaches customer data preparation

Databricks CustomerLake is built on a foundation where data preparation happens within the unified lakehouse, not in a disconnected silo.
Unified data consolidation begins with Lakeflow, which ingests customer data from transactional databases, SaaS systems, and streaming sources into Delta tables within a single lakehouse. Data remains governed and queryable without duplication.
Identity resolution is handled by Agentic Identity Resolution (AIR), which automatically stitches customer records across sources using deterministic matching (email, phone), probabilistic scoring (name and address similarity), and agentic techniques that learn from your data patterns. The result is a golden customer profile per individual.
Profile preparation is automated through Profile Agents, which transform raw customer records into business-ready Customer 360 profiles. These agents handle normalization, deduplication, and the creation of customer segments without manual ETL scripting.
Governance through Unity Catalog ensures all prepared data is tagged with access levels, audit logs, and lineage information. Marketers can discover trusted datasets and use them directly through purpose-built interfaces without requesting custom data exports from data teams.
Real-time streaming and enrichment is supported through Lakeflow's streaming connectors and integration with third-party data providers. Behavioral signals (web clicks, email opens, ad impressions) flow continuously into the lakehouse and are immediately available to agents.
Activation happens through Delta Sharing and API integrations, which send customer segments and messages to marketing tools without moving data out of Databricks.

Use cases

Multi-channel marketing orchestration: A retail company ingests point-of-sale data, email engagement, and web behavior into a lakehouse. Identity resolution creates a unified profile per customer. Agentic CDP then dynamically recommends which channel and message will be most effective for that customer at that moment, without data teams running manual queries.
Real-time personalization for e-commerce: An online retailer streams web session data, purchase history, and browsing signals into a real-time customer profile. Agents access this profile to decide what product recommendations to show, ensuring freshness and relevance.
Privacy-compliant customer insights: A financial services company consolidates customer data within Unity Catalog, which tracks every access and supports data masking rules. Compliance and marketing teams can verify that customer data is used only for intended purposes.

FAQs

Do I need to move my customer data to a new system?

No. CustomerLake operates on data that lives in your Databricks lakehouse. You prepare and organize that data once in Delta tables, and then activate it without copying. Existing systems (your CRM, email platform, data warehouse) stay in place while Databricks becomes the trusted source of customer truth.

How long does identity resolution typically take to set up?

Identity resolution happens automatically as part of data ingestion through Agentic Identity Resolution. Deterministic rules (email matches) work immediately. Probabilistic matching improves over time as AIR learns from your data. The specific timeline depends on data quality and volume, but many implementations see usable golden profiles within weeks of ingestion starting.

What if our customer data quality is poor?

Data quality issues are common. The principle is that agents are only as good as the data they see. Start by ingesting your data into the lakehouse and running data profiling to identify gaps, duplicates, and inconsistencies. Profile Agents and identity resolution reduce the manual work, but data teams should prioritize the most critical fields first (email, customer ID, purchase amount) before activating for marketing use.

Can I start with a subset of customer data?

Yes. Prepare a focused dataset first, perhaps a single source system and a narrow customer segment, to validate your identity resolution and profile quality. Expand incrementally as you refine the model and gain confidence in the results.