archive, boxes, shelf, folders, documents, data, storage, rack, row, stock, business, warehouse, depth of field, archive, data, storage, warehouse, warehouse, warehouse, warehouse, warehouse. How to choose customer data platforms: review criteria
Photo by Pexels on Pixabay

Operations

Part of Customer data platforms: durable identity models anchor every build

How to choose customer data platforms: review criteria

How to choose customer data platforms by assessing need, architecture, identity tests, supplier diligence, cost units, pilot evidence, and exit terms.

What to take away

  • A CDP earns its place only when one customer decision needs identity resolved across systems no single tool already owns.
  • Identity graph tests use your own messy recordsshared emails, duplicate accounts, anonymous-to-known transitions, false merges and splits.
  • Architecture class decides your operating cost. Packaged, suite-native, warehouse-native, composable and hybrid models fail in different places.
  • Price the volumes you actually runprofiles, events, credits, identity runs, streaming, audience refresh, egress and non-production environments.
  • Export profiles, consent records, segment definitions and logs before you sign. Deletion and transition terms belong in the contract, not the roadmap.

What the CDP is for

How to choose customer data platforms starts with the decision that currently fails. If a connector, a warehouse table or a CRM field fixes it, you do not need a CDP.

A CDP is warranted when one customer view must feed several destinations and no existing system owns that view. Write down which system stays authoritative for each entity before you take a demo.

Pick the architecture class

The five classes behave differently under load and under audit.

CDP architecture classes compared

Packaged

Identity lives
Vendor platform
Suits
Fast activation

Suite-native

Identity lives
Marketing/CRM suite
Suits
Suite standardised

Warehouse-native

Identity lives
Your warehouse
Suits
Strong data engineering

Composable

Identity lives
Split across tools
Suits
Willing to assemble

Where identity lives

Packaged
Vendor platform
Suite-native
Inside a marketing or CRM suite
Warehouse-native
Your warehouse, vendor computes on top
Composable
Split across storage, identity and activation tools
Hybrid
Vendor platform plus warehouse sync

Suits

Packaged
Teams without warehouse skills who need activation fast
Suite-native
Buyers already standardised on that suite
Warehouse-native
Teams with strong data engineering and strict data-location rules
Composable
Teams willing to assemble and operate the parts
Hybrid
Groups running both a central warehouse and a marketing suite

None of these is the modern choice by default. The class follows where your data already sits, who can operate it, and what your governance rules permit.

Test the identity graph

Ask every finalist to resolve the same controlled record set. Include shared household emails, duplicate accounts, anonymous-to-known transitions, and job changes. Also include conflicting attributes, opt-outs, late-arriving events, and deletion requests.

Identity graph test record set

  • Shared household emails
  • Duplicate accounts
  • Anonymous-to-known transitions
  • Job changes
  • Conflicting attributes
  • Opt-outs and late-arriving events
  • Deletion requests

Then check the merge itself. Can you see why two profiles joined, split a false merge, reverse it, and correct the history downstream? A profile screen in a demo shows none of this.

Run supplier due diligence

NIST SP 1326 frames supplier due diligence as research into pertinent information about an ICT supplier or product, assessed around ownership or influence, provenance, resilience, foundational cyber practice and supply-chain tiers. Use the NIST SP 1326 publication in proportion to what the CDP touches.

The NCSC's SaaS security guidance covers these areas:

  • purpose
  • access
  • service identities
  • data protection
  • logging
  • monitoring
  • incident investigation Apply its operating questions to the CDP you shortlist. Your DPO and the relevant supervisory authority set the requirements that bind you; this is not legal advice.

Model the cost at your volumes

Vendors price on different units. Some charge per profile, some per event, some per credit, some per row scanned.

Build three scenarios in a spreadsheet: ordinary month, peak month, and a month with a failed identity run you must reprocess. Include these on your side:

  • streaming
  • audience refresh
  • destinations
  • data egress
  • extra environments
  • support
  • cloud processing Get the vendor's written definition of each billable unit before you compare totals.

Pilot, then plan the exit

The GOV.UK technology selection guidance recommends adaptable choices, data control, security review and ownership-cost analysis. Those public-service questions transfer to a commercial CDP review.

Run a bounded case with real source owners, real permissions, monitoring and a recovery drill. Measure identity accuracy, activation latency, task effort and defects. Before signing, require:

  • export profiles, identities, consent, events, segments, configurations and logs
  • confirm deletion obligations in writing

The CISA software acquisition fact sheet adds development practice, supply-chain exposure, deployment and vulnerability management questions. Treat it as a checklist, not an approval.

Selection gate

GateEvidenceDecision
NeedThe failing decision and the responsibility boundaryCDP, alternative, or no change
ProductIdentity and activation results on your recordsShortlist or reject
SupplierDue diligence, assurance, service terms, contractAccept, mitigate, or reject
Operation and exitNamed owner, modelled cost, export and deletion proofPilot, revise, or stop

Teams that cannot name the record owner usually start by mapping customer data platforms use cases before they sign anything. If the architecture question is still open, compare the classes on customer data platforms against your own data location and skills.

Common questions

How many vendors should reach the pilot?

Two or three. Once your mandatory identity, security, cost and exit requirements are written down, most of a long list removes itself.

Can we trust a vendor's identity graph demo?

No. Controlled records with known correct answers are the only test worth running. Ask to see a false merge split and reversed.

What should stop a purchase?

Unclear record ownership, an identity test that fails on your data, a billable unit the vendor will not define, or export and deletion terms that stay verbal.

Do we need a warehouse-native CDP?

Only if your warehouse is already the trusted source and your team can operate identity and activation there. Otherwise the operating burden lands on people who did not ask for it.

More in Operations

Latest from Analysis Desk