Careers
Part of Customer data platforms: durable identity models anchor every build
Customer data platforms features: a buyer's checklist
Use this customer data platforms features checklist to connect collection, identity, governance, activation, and operations with use cases and acceptance tests.
What to take away
- Turn every CDP feature into a requirement tied to an approved use case, owner, edition, dependency, acceptance scenario, and complete cost.
- Test data collection, models, identity, permission, profiles, audiences, destinations, access, monitoring, correction, deletion, and recovery as a connected service.
- Remove impressive features that serve no approved decision, and distinguish generally available functions from add-ons, configuration, partners, preview, and roadmap.
Customer data platforms features matter only when they support a defined customer or business decision under clear ownership. Build requirements from representative records and failure paths, then verify which features are native, configured, custom, partner-delivered, limited by edition, or still on a roadmap.
Collection and connection
The CDP Institute's Use Case Generator asks teams to specify goals, systems, data types, task categories, functions, participating groups, and KPIs. Use that vendor-neutral structure to connect features to a declared use case, while reviewing every generated or prefilled field before relying on it.
Ingestion and connection checks
- Batch and streaming ingestion
- SDKs, APIs, files, databases
- Warehouses, federation, zero-copy access
- Change capture, event forwarding, reverse ETL
- Schema discovery, validation, retry, backfill
- Deletion and monitoring
Evaluate batch and streaming ingestion, SDKs, APIs, files, databases, warehouses, federation, zero-copy access, change capture, event forwarding, reverse ETL, destinations, custom connectors, schema discovery, validation, retry, backfill, deletion, and monitoring. Test the actual sources, volumes, credentials, latency, and error behavior.
Data model and lineage
Require entities, relationships, events, attributes, timestamps, source precedence, transformations, calculated fields, schema versioning, lineage, quality rules, and owners. Check B2C, B2B, household, account, device, product, and order structures needed by the use case. A predefined model can accelerate work or force costly translation.
Identity and persistent profiles
Test deterministic and probabilistic matching, anonymous-to-known transition, account linking, merge, split, confidence, conflict, source history, stewardship, and reversal. Require persistent unified records and clear identity responsibility, then confirm how the platform preserves changes and exposes them downstream.
Privacy, consent, and governance
NIST's Data Governance and Management Profile project is developing a joint framework resource for data-governance priorities across privacy and cybersecurity, with the initial public draft still pending. Use the project status and published materials as evolving planning context, not finished guidance, certification, or legal advice.
Evaluate permission and preference records, usage policies, labels, role-based access, sensitive-data controls, regions, retention, correction, deletion, audit history, service accounts, encryption, incident evidence, and downstream enforcement. Structure risk work while obtaining qualified review and testing the local implementation.
Audiences and activation
Check segment logic, size estimates, refresh, real-time evaluation, exclusions, versioning, expiry, approvals, destinations, identity mapping, hashing, suppression, delivery confirmation, match reporting, failure, replay, and deletion. Verify the receiving system's interpretation. An audience marked delivered may still be incomplete or joined incorrectly.
Operations and economics
Require environments, testing, approvals, deployment, logs, alerts, health dashboards, usage and cost monitoring, APIs, backup, recovery, support, status communication, documentation, export, and termination. Add marketer usability, administrator effort, data-team workload, accessibility, and implementation skill. Feature breadth without operability creates shelfware.
Score every feature with an acceptance scenario, evidence, owner, importance, edition, dependency, and full cost. Remove requirements that serve no approved use case. Pilot the highest-consequence identity, permission, destination, and recovery paths before selection and procurement.
Feature acceptance record
| Feature area | Representative test | Evidence to keep |
|---|---|---|
| Collection and model | Late, duplicate, invalid, changed schema | Input, transform, lineage, error, recovery |
| Identity and profile | Shared ID, merge, split, correction, history | Rule, confidence, steward, downstream effect |
| Audience and activation | Exclusion, permission, failure, retry, deletion | Membership, receipt, suppression, reconciliation |
| Operation | Role, change, incident, volume, export | Approval, log, cost, support, exit |
Assign the practice
The NIST Privacy Framework starting guide outlines a voluntary process for identifying and managing privacy risk. Use it to assign data and response owners for customer data platforms features; it is not legal clearance.
The CISA business-system logging guidance explains how event records support security review. Preserve appropriate customer data platforms features access, change, failure, and correction records without claiming that logging validates a metric.
Common questions
Which CDP features are mandatory?
Only those required to deliver approved use cases under the company's identity, data, governance, service, accessibility, security, cost, and recovery constraints.
Is real-time processing always better?
No. Use the slowest service level that meets the customer and business decision, then test freshness, latency, failure, cost, and recovery end to end.
How should a feature request be scored?
Record the use case, decision, evidence, owner, importance, edition, dependency, implementation, complete cost, acceptance test, and stop condition.
