Definition
A customer data platform is packaged software that builds a persistent, unified customer database from first-party sources and makes that database accessible to other systems. Three capabilities define the category. It ingests data from any source, event streams and batch files alike, without a bespoke integration project for each one. It resolves identities, so that a web visitor, an app user, an email subscriber and a store purchaser who are the same person become one profile with one history. And it exposes profiles, traits and audiences to other tools for activation and analysis, rather than keeping them for its own reporting.
The profile is the product. It combines identifiers (email, customer number, device, cookie), declared attributes (preferences, consent choices), behavioural events (viewed, added to basket, purchased, opened, contacted support) and computed traits such as predicted LTVTotal value a customer brings over time. or churn risk. Because the data is first-party and tied to a consented relationship, the profile can persist for years, which is the sharpest difference from the anonymous, short-lived records a DMPPlatform that organises audience data for targeting. keeps.
How it works
Data enters through SDKs on sites and apps, server-side event APIs, connectors to SaaS tools and scheduled loads from a warehouse. Each record carries at least one identifier. The identity layer applies matching rules, usually deterministic (same hashed email, same customer number, same login) and sometimes probabilistic, to decide which records belong to the same person, and merges them into a profile while keeping the source records for audit. A trait layer computes derived fields, and a segment builder lets a marketer define audiences from traits and event sequences without writing queries. Destinations then receive audiences or profile updates: ad platforms for targeting and suppression, email and messaging tools for journeys, personalisation engines for on-site content, and the warehouse or BI tools for analysis.
| Capability | What it does | Where it goes wrong |
|---|---|---|
| Ingestion | Accepts events and records from every source | Inconsistent event names and identifiers across sources |
| Identity resolution | Merges records about the same person | Over-merging households or under-merging a person's devices |
| Profile store | Persists traits, events and consent per person | Retention and deletion rules not aligned with consent |
| Segmentation | Builds audiences from traits and behaviour | Segments defined once and never revisited |
| Activation | Syncs audiences to ad, email and personalisation tools | Match rates at the destination lower than expected |
Two architectures exist. A packaged CDP stores the profiles in its own database. A composable or warehouse-native CDP leaves the data in the organisation's warehouse and adds identity, segmentation and activation on top of it. The data flow and integrations guide covers where either design sits among the other systems, and the martech stack hub covers the choice between them.
In practice
A CDP earns its place when the same person appears in several systems and no single one of them can see the whole relationship. The typical first uses are suppression (stop showing acquisition ads to existing customers), cross-channel journeys triggered by behaviour in one channel and delivered in another, and value-based audiences, where profiles are ranked by predicted value and the top tier is used as a seed for prospecting or as a priority list for service. Each of these depends on the identity layer working, so most implementations start with an audit of which identifiers each source actually carries.
Unifying a retailer's customer records
A retailer has 2,400,000 email subscribers, 1,100,000 online purchasers and 650,000 loyalty-card members in three separate systems. After deterministic matching on hashed email and loyalty number, the CDP resolves these into 2,950,000 profiles, of which 38% carry a purchase history and 21% carry both online and in-store activity. The first activation is a suppression list of 710,000 active customers sent to two ad platforms, which cuts acquisition spend wasted on existing buyers by an amount the team estimates at £42,000 a month. The second is a value-tier trait that feeds a 5% top-customer seed audience for prospecting.
What it is not
A CDP is not a CRM. A CRM is the system of record for relationships that sales and service staff manage, built around accounts, contacts, deals and tickets; the CRM fundamentals guide describes that role. A CDP usually ingests from the CRM and sends traits back to it, but it does not replace it. A CDP is not a DMPPlatform that organises audience data for targeting.: the DMP works on anonymous identifiers for advertising, the CDP on known people across every channel. It is not a data warehouse, although a composable CDP runs on one; the warehouse stores everything, the CDP is opinionated about people. It is not an analytics tool, and it is not an AttributionAssigning conversion credit to touchpoints. model, even though its unified event history is the best raw material for both. And it is not a consent management platform, though it must store and enforce the consent each profile has given.
Related terms
The DMPPlatform that organises audience data for targeting. is the term most often confused with CDP, and the two are best understood as answers to different eras: anonymous cross-site identifiers then, consented first-party profiles now. LTVTotal value a customer brings over time. is the computed trait that justifies most CDP investment, because value-based audiences and suppression lists need a per-person value to rank on. AttributionAssigning conversion credit to touchpoints. benefits from a CDP indirectly, since a unified, deduplicated event history across devices is what multi-touch analysis is usually missing. On the systems side, the martech stack hub places the CDP among the other platforms, and the data flow and integrations guide shows the pipes in and out. The CRM is not a glossary entry here but belongs in the same conversation: the CRM fundamentals guide describes the system a CDP most often reads from and writes back to, and the boundary between the two is where ownership disputes usually start.
Common mistakes
- Buying a CDP before fixing event collection. Identity resolution cannot merge records that carry no shared identifier; the tagging and event naming work comes first.
- Expecting the CDP to replace the CRM or the warehouse. It sits between them and depends on both.
- Measuring success by profile count. Profiles that lack a usable identifier or any recent activity add nothing to activation.
- Ignoring destination match rates. An audience of known customers only reaches the share of them the ad platform can match, and that share varies by platform and by identifier type.
- Treating consent as a field rather than a rule. Every destination sync must honour the purposes each profile consented to, and deletion requests must propagate.
- Letting segments multiply without owners. Hundreds of unnamed, unused audiences make the activation layer as hard to govern as the spreadsheets it replaced.
