CDP (customer data platform): when a large company really needs one

A CDP, or customer data platform, is a system that brings together first-party customer data from multiple sources, resolves each person's identity and maintains unified profiles that can be segmented and sent to other tools, such as media, email, website and customer service. What sets it apart from other data systems is the combination: it is built for marketing to activate data, not to store it.

In large companies, the same person shows up in e-commerce, in stores, in the app, in customer service and in the loyalty program, often with different records in each brand of the group. First-party data, collected through the direct relationship with the customer, is the data the company controls and can explain. The question for leadership is whether a CDP is the way to organize it, or whether the existing CRM, data warehouse and tracking already solve the problem.

What a CDP is and what it does

A CDP performs five functions. Not every product does all of them with the same depth, and not every project needs all of them.

• Ingestion: receives events from the website and the app, registrations, orders, customer service interactions and campaign responses, in real time or in batches.
• Identity resolution: decides when different records belong to the same person. It can be deterministic, using verified identifiers such as email, CPF (Brazil's individual taxpayer ID) or login, or probabilistic, estimated from indirect signals. The latter requires more technical and legal caution.
• Unified profile: consolidates each person's attributes, interaction history and consents into a single queryable record.
• Segmentation: lets the marketing team build audiences using rules or models, without depending on a technical query for every request.
• Activation: sends audiences and attributes to media platforms, email and messaging tools, website, app and customer service, respecting each profile's consent.

The value lies in the full cycle. A unified base that does not reach the channels becomes just another repository.

CDP, CRM, DMP and data warehouse: what is the difference

All four systems handle customer data, but they answer different questions.

• CRM (customer relationship management): records the sales and service relationship, such as accounts, contacts, opportunities and tickets. It is operated by people, record by record. In enterprise sales, it is usually the source of truth for the funnel.
• DMP (data management platform): a system geared toward programmatic media, historically based on third-party cookies and identifiers, with profiles that are generally anonymous and organized into audience segments.
• Data warehouse: a central analytical repository that stores sales, finance, operations and marketing data for analysis and reporting. It is the foundation of business intelligence, but it was not designed for marketing to build and send audiences without technical support.
• CDP: unifies identified first-party data at the person level and makes it activatable by marketing, with identity and consent in the same profile.

In practice, the boundaries overlap. Some CRMs have segmentation and activation features, and a data warehouse can feed channels directly through integration tools. That is why the choice should start from the missing function, not from the product category.

When a large company needs a CDP

A CDP tends to make sense when several of these conditions appear together.

• Customers interact with the company through many touchpoints, such as stores, e-commerce, the app and customer service, and those records do not connect today.
• There are activation use cases with an estimated value, such as reactivating customers with no recent purchase or excluding recent buyers from acquisition campaigns.
• Demand for new audiences is frequent, and the technical team has become a bottleneck for every segmentation.
• Activation needs to happen in minutes or hours, not in weekly cycles.
• There is a reliable identifier in meaningful volume: login, loyalty registration, email or CPF linked to purchases.

When CRM, data warehouse and server-side are enough

In many scenarios, the stack the company already has solves the problem at lower cost and lower risk.

• Enterprise sales with few high-value customers, where the CRM already concentrates accounts, contacts and opportunities. The topic is covered in the article on CRM and revenue between marketing and sales.
• Batch use cases, weekly or monthly, that the data team handles with queries in the data warehouse and an integration tool connected to the channels.
• Measurement and sending conversions to media platforms, which server-side tracking with Consent Mode handles without a unified profile.
• A base with few verified identifiers. In that case, the priority is to create reasons for customers to identify themselves, not to buy a system to unify what does not yet exist.

Packaged CDP or composable CDP: two approaches

A packaged CDP is a product that combines ingestion, storage, identity, segmentation and activation in a single platform, with its own copy of the data. It tends to speed up the start of operations and offers a ready-made interface for marketing. In return, it creates yet another copy of customer data, which must be synchronized, governed and protected, and it can make it harder to use attributes and models that live outside it.

A composable CDP, also called a warehouse-native CDP, uses the company's data warehouse as its foundation. The data stays where it already is, and specific tools handle identity, segmentation and delivery to channels. It leverages definitions and models the data team already maintains and reduces copies. In return, it requires an organized data warehouse, a team capable of maintaining the identity layer and more coordination among vendors.

Neither is superior in the abstract. Centralized data and a mature technical team favor the composable approach. Scattered data and a need for fast marketing autonomy favor the packaged one. Hybrid architectures are also possible.

Prerequisites before any CDP project

• Identifiers: which keys link the records (email, phone, CPF, login, loyalty number) and in what share of the base each one exists.
• Consent: where and how it is collected, for which purposes, and how revocation reaches each system.
• Data quality: duplicates, required fields and registration standards across brands and channels.
• Prioritized use cases: a short list, with estimated value, activation channel and success metric.
• Owners: who owns the identity rules, who approves new audiences and who is accountable for incidents.

How to choose a CDP starting from use cases

• List the use cases and estimate the value of each one with the company's own data, such as the size of the inactive base and the average margin per purchase.
• For each case, describe the data required, acceptable latency, activation channel and consent rule.
• Map what the current architecture already covers. Some of the cases can be served by the CRM or the data warehouse with little effort.
• Define the approach (packaged, composable or hybrid) based on the gaps found, not on a preference for a product.
• Evaluate vendors with the company's data, in a proof of concept with two or three real use cases and success criteria written in advance.
• Calculate the total cost: license, integration, storage, ongoing operation and people.
• Plan the move to production in waves, starting with the use case of highest value and lowest risk.

LGPD: consent and legal bases in a CDP

A CDP concentrates personal data from many sources, which increases both its value and the responsibility. The LGPD (Law 13.709/2018), Brazil's General Data Protection Law, requires a legal basis for each processing activity and lists the grounds in Article 7, among them consent, performance of a contract and legitimate interest. Unifying a profile does not change the legal basis of each piece of data. The purpose stated at collection still applies, and the law does not allow further processing that is incompatible with it.

Three points deserve attention. The first is proof: the controller must demonstrate that consent was obtained in accordance with the law, and generic authorizations are void. The second is sharing: a controller that obtained consent and needs to communicate or share the data with other controllers must obtain specific consent for that purpose, except where the law waives it. This can affect the flow of profiles between companies in the same group that are separate controllers. The third is profiling: data subjects may request a review of decisions made solely on the basis of automated processing that affect their interests, including those that define their consumer profile.

For media, the guidance on cookies from Brazil's National Data Protection Agency (ANPD) states that legitimate interest will hardly be the most appropriate ground when data collected through cookies is used for advertising, and points to consent as the most appropriate ground in those cases, subject to the specific circumstances. In a CDP, consent should be a profile attribute, recorded by purpose and checked before each activation. Each use should be validated with the legal team and the data protection officer (DPO).

How to measure the value of a CDP

The value of a CDP is not in the number of unified profiles. It is in the additional results of the use cases it makes possible. That is why measurement should be incremental: compare a group that received the CDP-based action with a randomly selected control group that did not, as described in the article on media incrementality.

Consider a hypothetical retailer that uses the CDP to reactivate customers who have not purchased in six months. The numbers are illustrative. Of the 200,000 eligible customers, 180,000 receive the campaign and 20,000 are randomly assigned to the control group. Within 60 days, 4.5% of the treated group and 3% of the control group purchase again. The 1.5 percentage point difference across 180,000 customers corresponds to 2,700 customers reactivated by the action. With an average contribution margin of R$ 120 on the return purchase, the incremental result is R$ 324,000, to be compared with the campaign cost and with the share of the CDP cost attributable to the use case.

Operational indicators, such as the base identification rate and activation latency, show whether the platform works. Incrementality shows whether it pays for itself.

Common mistakes in CDP projects

• Buying the platform before defining the use cases and the value of each one.
• Assuming the CDP fixes bad data. It unifies what it receives, including duplicates and incomplete records.
• Treating identity resolution as a technical setting, with no business owner for the unification rules.
• Ignoring consent by purpose and activating audiences for uses the data subject did not authorize.
• Measuring success by unified profiles or audiences created, rather than by incremental results.
• Underestimating ongoing operation. Integrations, rules and audiences need maintenance after implementation.

Frequently asked questions about CDPs

What is a CDP in marketing?

CDP stands for customer data platform. It is a system that brings together first-party data from multiple sources, identifies when different records belong to the same person and maintains unified profiles that marketing segments and activates across media, email, website and customer service.

What is the difference between a CDP and a CRM?

A CRM records the sales and service relationship, with accounts, contacts and opportunities operated by people. A CDP unifies data from many sources, including website and app behavior, into profiles built for segmentation and activation. The two can coexist.

What is a composable CDP?

It is the approach in which the company's data warehouse serves as the foundation of the CDP, with no separate copy of the data. Specific tools handle identity, segmentation and delivery to channels. It reduces copies, but requires an organized data warehouse and the technical capacity to maintain the identity layer.

Does every large company need a CDP?

No. A CDP makes sense when there are many touchpoints, reliable identifiers, activation use cases with an estimated value and frequent demand for new audiences. When the CRM concentrates the relationship or the use cases run in batches, CRM, data warehouse and server-side tracking can solve the problem at lower cost.

What is first-party data?

It is the data a company collects directly through its relationship with the customer, such as registrations, purchases, website and app interactions and contacts with customer service. It differs from third-party data, which is collected by other companies. Even though it belongs to the company, it remains subject to the LGPD, with a legal basis, a defined purpose and respect for the data subject's rights.

To bring this into planning, explore our Data practice and the discussion on marketing data governance.