What Is a Customer Data Platform (CDP)?

David Raab of Raab Associates proposed the term in his Demand Gen Report column of 30 April 2013, where he saw "an entirely new category of system: the Customer Data Platform (CDP)". His definition had three parts: a customer-level database that links the identifiers of one person (email address, browser cookies, customer account), predictive models built almost automatically, and real-time answers to the systems that deal with customers. According to Raab, the CDP does not replace CRM or marketing automation and feeds them data and recommendations.

According to the CDP Institute, Raab named the category in a blog post in April 2013. Spanish and German have translations (plataforma de datos de clientes, Kundendatenplattform), but searches in both languages mostly use the English name.

Raab's version included predictive models and decisioning; the current definition centres on the profile. Dates according to the CDP Institute:

• September 2013: Raab publishes the first industry report on CDPs, profiling eleven systems.
• 2016: the CDP Institute is founded, and CDPs appear on the Gartner Hype Cycle for the first time.
• Since 2020: the Institute has expanded internationally and now has regional chapters, including one for Iberia.
• 2021: its definition speaks of "packaged software", bought and controlled by business users, as opposed to a data warehouse built by IT.
• 2025: the Institute becomes part of the Customer Data Alliance.
• 2026: the new definition drops "packaged" and names three architectures: own-store, warehouse-native (composable) and hybrid.

According to Gartner, a CDP must collect first-party customer data, online and offline, in real time or near real time. The flow has five steps:

• Collection: web and app events, purchases, CRM and in-store data.
• Identity resolution: deciding which identifiers belong to the same person.
• Profile: each customer's history, kept in full detail for as long as the company chooses, according to the CDP Institute.
• Segments: groups built by rules or by calculated values such as trends or model scores.
• Activation: profiles and segments for other systems, through APIs, queries or file extracts, according to the Institute.

Gartner names two kinds of identity resolution, deterministic and probabilistic. Deterministic matching links records when an identifier matches; Adobe documents that its Identity Service creates a link when the identity namespace and the value both match. Probabilistic matching estimates a match from signals that carry no certainty, and it can merge two different people.

Deterministic matching fails too. Twilio Segment describes a shared iPad in a store: if several customers sign up on it, their events carry the same device ID and, without rules, end up in one profile. Segment also describes a hard-coded test value in the user_id field that reaches production and merges hundreds of profiles by mistake.

The CDP Institute sorts vendors into four types. Each type includes the functions of the one before.

Source: CDP Institute, "What is a CDP?", section "Types of CDPs" (accessed 13.11.2026).

By architecture, an own-store (packaged) CDP keeps its own copy of the data; a composable CDP works on the company's data warehouse with components from several vendors; a hybrid does both. According to the CDP Institute, advocates argue that the composable approach avoids another copy of the data and gives more control; the Institute warns that most data warehouse technology handles real-time updates poorly, which may require copying the data.

In an example from Adobe's documentation, a person logs in to an online shop from a laptop, and the system receives their customer ID and the browser's ID. Later they log in from a phone and a second browser ID arrives. Because both events share the customer ID, the identity graph links all three.

An illustrative case: a UK retail chain with a website, an app and stores. A customer buys in the app with her email address and pays in a store with a loyalty card tied to the same address. The CDP joins both trails, and the chain can leave her out of acquisition campaigns, provided it has a lawful basis for that purpose.

A CDP pays off when several of these conditions apply:

• Customer data sits in several sources, such as the website, app, stores and CRM.
• Customers buy or interact through more than one channel.
• There is a reliable identifier, such as a customer ID or a verified email address.
• A team defines the use cases and maintains the integrations.

With one source and one channel, a CRM with an email tool covers the need. The costs are integration work and the risk of merging different people; the composable variant, according to the CDP Institute, takes more engineering effort.

Source: CDP Institute, "What is a CDP?", section "How a CDP Differs from CRM, DMP and Data Warehouses" (accessed 13.11.2026).

The difference from a DMP lies in the kind of data: first-party data about identified customers against anonymous third-party segments.

Products their makers describe as CDPs, in alphabetical order, without ranking.

Source: each vendor's product page (accessed 13.11.2026).

The CDP Institute keeps a directory of more than 240 vendors.

The reference definition comes from an institute that, according to its FAQ, is funded by CDP vendors and companies with related products. Its RealCDP certification is audited by an independent third party and paid for by the vendor.

The line between CDP and data warehouse keeps moving. Gartner writes that CDPs operate in alignment with enterprise data platforms, often using federated or zero-copy access. According to the CDP Institute, if the warehouse lacks many capabilities, a conventional CDP is often more effective.

In July 2026 Gartner Peer Insights summarised lessons from CDP users (minimum of 75 responses): document use cases, clarify pricing before deployment, cleanse data before migrating and roll out in phases.

In the EU, a CDP processes personal data under the GDPR: every processing purpose needs a lawful basis (Article 6), people may object at any time to direct marketing, including profiling (Article 21(2)), and when data is erased (Article 17), the controller must tell each recipient, unless that proves impossible or involves disproportionate effort (Article 19); in a CDP, those recipients are the connected systems. In the UK, regulation 6 of PECR, as substituted on 5 February 2026, restricts storing or accessing information on a user's device, subject to the exceptions in Schedule A1.

• Write down the use cases before you look at vendors.
• List your data sources and the identifier that links them.
• Check the lawful basis and consent for each source and each purpose.
• Choose the architecture based on the state of your data warehouse and your technical team.
• Start with one use case and one or two sources.
• Decide before launch how you will measure the result.

• Normalise identifiers before loading them, for example email addresses in lower case.
• Block test or default values as identifiers, as Segment recommends.
• Connect data erasure to every system that receives profiles.

• Buying the platform before defining the use cases.
• Assuming every CDP treats identifiers the same way: according to their documentation, Adobe is case-sensitive and Segment is not.
• Using probabilistic matching for purposes that need certainty about the person.

This entry was written in November 2026 with the help of AI. The information comes from David Raab's 2013 column, the CDP Institute, Gartner, vendor documentation and product pages, and the legal texts. It was reviewed before publication and contains no client data.