Customer Data Platform

A retailer has purchase history in its checkout system, email choices in a marketing tool and support conversations elsewhere, and it wants to know whether a campaign reached existing buyers or new prospects. A customer data platform can combine selected records when identity and permissions allow.

Define the use case before buying a platform, since a single view for service agents, a customer-retention analysis and real-time ad targeting need different data and controls. Map sources and identifiers, because an email address, account ID, browser identifier and phone number may each refer to a person but can change or be shared.

Adobe's Real-Time Customer Profile documentation describes merging fragments from online, offline and CRM sources using identity mapping and merge policies, which shows that a profile is assembled from source records and conflicts require a decision rule, while Salesforce describes a CDP as bringing customer data together for segmentation and use across channels. Distinguish deterministic from probabilistic matching, since a signed-in account link may support a stronger match than similar browsing patterns, so state what evidence was used and avoid treating a low-confidence connection as a fact for sensitive decisions.

Decide which source wins when values conflict, because a customer might update their contact choice through a privacy portal while an older CRM import says something else, and a merge policy should not let stale data override a newer valid preference. A CDP often creates audiences, such as recent buyers who have opted into a particular channel, but that is different from granting permission to message them, so apply the appropriate consent, marketing and regional rules to the actual activation.

Design events carefully, since an "order placed" event is not the same as a completed paid order and a return can reverse an earlier signal, and if definitions shift between systems a segment may include customers who never actually bought. Keep timestamps and source lineage, because a support issue from two years ago may be irrelevant to today's message while a newly withdrawn preference is urgent, and analysts should be able to trace why a person was included in an audience.

Respect deletion and access requests, since information copied into profiles, segments and downstream destinations can be difficult to remove, so build a process that finds relevant identifiers and sends changes to connected systems. Limit access by role, because marketing may need an eligible audience count while support needs a customer's recent orders, and neither necessarily needs all health, payment or identity details held by the organisation.

Measure match quality by checking a sample of linked profiles for false merges and missed links, especially across shared family accounts or business contacts. An illustrative profile-link rate is profiles with at least two verified source links divided by eligible profiles, so if 7,000 of 10,000 qualify the rate is 70%, which does not show whether those links are correct or whether activation is permitted.

Separate analytics from activation, since a business may study aggregate behaviour without sending a message or exporting identifiers to an ad network, and document each use and its access path, not merely the central storage location. Consider integration costs, because data connectors, event schemas, identity rules, privacy requests and ongoing maintenance can be more demanding than the licence fee, and a small business may meet its first use case with well-governed CRM and reporting data.

Test an end-to-end scenario with real constraints by changing a preference, creating an order, issuing a return and seeing how the profile updates. For owners, a CDP is useful when a reliable joined view supports a specific customer decision, more data is not automatically a better relationship, and the test is whether the platform can make a useful view while honouring source accuracy and customer choices.