Customer Data Platform Selection: When a CDP Solves a Problem You Actually Have
A CDP does not fix bad data. It centralises it faster.
Customer data platforms are bought for one of two reasons. The good reason: marketing cannot act on customer data because it lives in six systems with no common identity. The bad reason: the current martech stack underperforms and a new platform feels like progress.
The distinction matters, because a CDP centralises data faster — it does not improve data that is inaccurate, inconsistently defined, or collected without a usable consent basis.
Identify the problem precisely
- Fragmented identity. The same person exists as separate records in commerce, email, support, and app analytics with no link — a genuine CDP problem.
- Slow activation. Building a segment takes an engineering ticket and three weeks — a genuine CDP problem.
- Poor data quality. Records are wrong or stale — not a CDP problem; it will propagate the errors more efficiently.
- Unclear strategy. Nobody agrees which segments matter — not a CDP problem, and buying one will not produce agreement.
Identity resolution is the actual work
| Decision | Consideration |
|---|---|
| Deterministic matching | High confidence on email, phone, or account ID — start here |
| Probabilistic matching | Higher reach, real error rate; unsuitable for sensitive use cases |
| Household versus individual | Affects targeting appropriateness and privacy expectations |
| Consent linkage | Consent must travel with the profile, per channel and purpose |
| Deletion propagation | Erasure requests must reach every downstream destination |
Consent and deletion are frequently deferred and should not be. A CDP that cannot honour an erasure request across every activation destination is a compliance liability sitting at the centre of your customer data.
Packaged, warehouse-native, or composable
Packaged CDPs offer fast time to value, prebuilt connectors, and marketer-friendly interfaces, at the cost of duplicating your customer data into another system with its own governance surface.
Warehouse-native and composable approaches build profiles and segments where your data already lives and add an activation layer on top. If you already operate a mature warehouse with reliable pipelines, this is usually the better economic and governance answer — it avoids a second copy of customer data and reuses existing access controls.
Choose packaged when data engineering capacity is the constraint; choose composable when governance and duplication are.
Define the activation use cases first
Write down the three to five things you will do on day one that you cannot do today — suppress paid advertising for existing customers, trigger a service outreach on a behaviour signal, personalise a category page by observed affinity, unify support context across channels.
Then verify the platform can activate into those specific destinations with the required latency. Real-time activation is expensive and only some use cases need it; batch is adequate for most segmentation, and paying for streaming you do not use is a common cost mistake.
Avoid rebuilding a CRM
Scope creep in CDP programmes tends toward becoming the master system for customer records, which is a multi-year data governance project in disguise. Keep the CDP as a profile and activation layer that reads from systems of record and writes segments outward. Where a field needs to be authoritative, it belongs in the CRM, ERP, or commerce platform — not in the CDP.
Measure it on activation outcomes
Track time to build and activate a new segment, the share of campaigns using unified profiles, match rate and confidence, suppression accuracy, and incremental revenue or cost saving from the day-one use cases. Profile counts and data volume prove nothing about value, and they are what vendor dashboards emphasise.
Frequently asked questions
What does a CDP do?
Ingests customer data from many sources, resolves it into unified profiles, builds segments and attributes, and activates them into marketing, service, and advertising channels.
Do we need one if we have a warehouse?
Often not a packaged one — warehouse-native or composable approaches deliver the same outcomes without duplicating data and governance.
What is the hardest part?
Identity resolution: which identifiers may be joined, at what confidence, and under what consent basis.
Is real-time activation necessary?
For a minority of use cases. Batch activation is sufficient for most segmentation and materially cheaper.