A customer data platform is frequently described as a marketing database, which understates the part that is difficult. Its central function is deciding which records belong to the same person.
Source systems each hold a partial view
A retailer's data sits in an ecommerce platform, an email tool, a point-of-sale system, a service desk and a web analytics product, each with its own identifiers.
None of these were designed to reference the others, so the same customer exists several times over under unrelated keys.
Any question spanning channels therefore cannot be answered from a single system, which is the gap the platform is bought to close.
Identity resolution is the core mechanism
The platform ingests records from each source and applies rules to determine which belong to one individual, using shared identifiers where they exist.
Deterministic matching relies on an exact shared value such as a hashed email address, and is reliable where that value is present.
Probabilistic matching infers a link from weaker circumstantial signals, which extends coverage at the cost of introducing errors that are hard to detect.
Persistence separates it from an analytics tool
Analytics products are built to aggregate behavior into reports, and many deliberately avoid retaining detailed individual histories over long periods.
A customer data platform retains the profile, accumulating events against it so the record grows over years rather than resetting by session.
That persistence is what allows segmentation on a full history, such as customers who bought once and never returned within a defined window.
Activation is what makes it operational
A profile store with no outbound connection is a reporting asset, and the platform's practical value comes from pushing segments into the tools that act on them.
Those connections carry audiences to email systems, advertising platforms and site personalization tools, usually as a hashed identifier rather than raw personal data.
Match rates at the destination are always below the segment size, since the receiving platform can only recognize the subset of people it already knows.
Governance is where implementations succeed or fail
Because the platform consolidates data from across a business, it concentrates obligations around consent, retention and deletion into one place.
A deletion request has to propagate through the profile and everything derived from it, which is only possible if the lineage of each field was recorded.
Projects that treat this as a later phase generally discover that retrofitting governance costs more than the original build, and that the platform cannot be used in the meantime.