Choosing an identity verification provider is a security, privacy, compliance, and operational decision—not just a software purchase. This reusable due diligence checklist helps security, procurement, engineering, fraud, and compliance teams compare vendors, request meaningful evidence, identify contract gaps, and document a defensible risk decision before signing.
Overview
An identity verification vendor may process identity documents, facial images, biometric templates, device data, contact details, and decisions that affect whether a person can open or access an account. The review should therefore examine the complete service: data collection, OCR, document verification, face verification, liveness detection, manual review, fraud signals, APIs, dashboards, storage, support, and deletion.
Start by documenting your intended use case. Specify the countries and document types in scope, expected transaction volumes, user journeys, review requirements, retention needs, integration points, and consequences of a false acceptance or false rejection. A vendor that works well for low-risk account registration may not satisfy a high-risk KYC onboarding process or a regulated workflow.
Use evidence rather than assurances. Ask the provider to distinguish between controls that are implemented, controls that are planned, and capabilities that depend on configuration or a subcontractor. Certifications and audit reports can support a review, but they do not replace questions about your particular deployment, data flows, model performance, or contractual protections.
For broader requirements, compare this checklist with KYC and AML software requirements and the practical considerations in build versus buy identity verification.
A simple risk-rating method
Score each control area from 0 to 3: 0 means no answer or unacceptable evidence, 1 means a material gap, 2 means acceptable with conditions, and 3 means strong evidence and clear ownership. Weight areas that could affect safety, regulatory obligations, or service continuity more heavily than convenience features. Record the evidence, reviewer, date, open action, and decision owner—not just the score.
Checklist by scenario
Security and infrastructure review
- What data is collected at each step, and where is it processed, stored, cached, or logged?
- Is data encrypted in transit and at rest? Who manages the keys, and how are privileged key-management actions controlled?
- How are production environments separated from development, testing, demonstrations, and support tools?
- What identity and access controls protect administrative interfaces, APIs, support systems, and internal data stores?
- Is multifactor authentication required for privileged access? Are access reviews, service-account controls, and emergency access documented?
- How are vulnerabilities discovered, prioritized, remediated, and communicated to customers?
- What independent security assessments, penetration tests, or control reports can the vendor provide under appropriate confidentiality terms?
- How are API credentials issued, rotated, scoped, monitored, and revoked?
Biometric and computer-vision review
- Which biometric data is created or derived, including face images, templates, embeddings, or liveness signals?
- Is biometric processing performed only for the stated verification purpose, or can data be used for product improvement, training, testing, or other purposes?
- How do passive, active, or behavioral liveness controls work at a high level, and what attack classes are tested?
- How does the provider address presentation attacks, replay attempts, injection, deepfakes, manipulated documents, and unusual capture conditions?
- What accuracy, failure, demographic, device, and document-type testing is available for the proposed workflow?
- Can your organization configure thresholds, route uncertain cases to review, and monitor changes after model updates?
- How are model versions governed, tested, approved, rolled back, and communicated?
Do not treat a single accuracy figure as a complete assessment. Review false acceptance, false rejection, abandonment, latency, manual-review rates, and performance across the populations and devices relevant to your service. The face verification testing guide and liveness evaluation checklist provide useful questions for a technical comparison.
Privacy, compliance, and data-governance review
- Who is responsible for each processing activity, and does the contract clearly describe the parties' roles?
- What legal, operational, and configurable controls support the customer’s privacy and KYC obligations?
- Where are data and backups located, and can the vendor identify all relevant processing regions?
- Which subprocessors, hosting providers, analytics services, review teams, or support systems can access the data?
- How are new subprocessors announced, assessed, challenged, or rejected?
- What are the retention periods for source documents, images, biometric outputs, decision records, logs, and backups?
- Can deletion requests be verified across primary systems, replicas, caches, and backups, subject to documented exceptions?
- What audit logs demonstrate consent or notice presentation, verification events, administrative access, decisions, changes, and deletion?
- How does the service support data-subject requests, investigations, complaints, and regulator or auditor inquiries?
Biometric data requirements vary by jurisdiction and use case. Have privacy counsel or the responsible compliance team confirm the applicable analysis rather than relying on a vendor’s general statement about GDPR biometric data compliance.
Resilience and incident-response review
- What uptime target, service credits, support coverage, maintenance process, and status communication apply to the purchased service?
- What recovery time and recovery point objectives are defined for the API, decisioning systems, document stores, and administrative console?
- How often are backups restored and disaster-recovery procedures tested?
- Does the vendor have a documented incident-response plan, named escalation contacts, and a process for customer notification?
- What happens if a critical model, cloud region, document source, or subcontractor becomes unavailable?
- Can your workflow fail safely by pausing, routing to manual review, or applying a preapproved alternative?
Commercial and exit review
- Are identity verification pricing assumptions clear for retries, manual reviews, document types, countries, support, storage, and overages?
- Which features are included in the contracted tier, and which are subject to future packaging or separate fees?
- Can the vendor provide complete exports of required records in a usable format at termination?
- What is the deletion timetable after termination, and what evidence will be supplied?
- Are assistance, notice periods, migration support, and transition fees defined?
- Can the vendor change subprocessors, processing regions, models, or material controls without meaningful notice?
What to double-check
Request an evidence pack that matches the proposed architecture. Useful items may include a current security report or certification, penetration-test summary, data-flow diagram, subprocessor list, privacy and retention schedule, business-continuity summary, incident-response procedure, access-control description, model-change policy, sample audit logs, and service-level terms. Mark each item as received, relevant, current, and contractually binding where appropriate.
Compare vendors using the same test protocol. Give each provider identical representative documents, capture conditions, fraud scenarios, and integration requirements where permitted. Measure not only whether a transaction passes, but also the reason for failure, time to decision, manual-review path, available evidence, and behavior when dependencies are unavailable. For workflows involving manual review, define quality assurance, escalation, reviewer access, and audit requirements; see the KYC manual review operations guide.
Review the contract against the technical answer. If a sales response promises configurable retention, regional processing, deletion, or model controls, ensure those commitments appear in the agreement, data-processing terms, security exhibit, service description, or an incorporated policy. Ask what happens when documentation conflicts, and identify which document controls.
Suggested comparison table
| Area | Evidence to request | Decision question |
|---|---|---|
| Data protection | Data-flow diagram, encryption description, retention schedule | Can we limit collection and verify deletion? |
| Biometrics | Processing description, model governance, test summary | Can we understand and control biometric risk? |
| Resilience | Service terms, recovery objectives, test summary | Can onboarding continue safely during an outage? |
| Subcontractors | Current list and change process | Can access and geographic exposure be monitored? |
| Exit | Export format, deletion procedure, transition terms | Can we migrate without losing evidence or control? |
Contract-redline prompts
Ask counsel to consider explicit limits on data use, defined retention and deletion, approved processing locations, subprocessor notice and objection rights, security-control obligations, incident-notification timing, audit cooperation, confidentiality, liability allocation, service-level remedies, model-change notice, export assistance, and post-termination support. The exact wording depends on the parties, jurisdictions, and risk profile; the goal is to turn operational assumptions into verifiable obligations.
Common mistakes
- Choosing a familiar brand without reviewing the deployment. A provider’s general reputation does not answer where your data goes, who can access it, or how your configuration behaves.
- Comparing feature lists instead of outcomes. “OCR,” “face matching,” or “liveness” labels do not reveal performance, review paths, evidence quality, or attack resistance.
- Accepting a certification as a complete review. Ask what scope, systems, dates, and control limitations apply to the report.
- Ignoring manual-review operations. A failed automated check still requires a secure, consistent, auditable resolution process.
- Leaving retention to default settings. Confirm defaults, configurable limits, backup behavior, and deletion evidence before production use.
- Failing to plan for vendor change. Models, subprocessors, cloud regions, pricing, and APIs can change. Define notice, review, rollback, and migration options.
- Scoring without recording exceptions. A high average score can hide one unacceptable issue in biometric processing, incident response, or exit support.
When to revisit
Re-run the review before each major procurement or renewal cycle, after a material workflow or tool change, and before expanding into new countries, document types, user populations, or risk categories. Also revisit it after a provider incident, significant model or API change, new subprocessor, processing-region change, regulatory development, or change in your retention policy.
At minimum, maintain a living record with the last review date, accountable owner, approved use case, risk rating, unresolved actions, evidence expiry dates, and next review trigger. During an annual supplier review, confirm that the production configuration still matches the evaluated design: endpoints, regions, subprocessors, retention, thresholds, access groups, logging, and fallback procedures.
For a practical next step, assign one owner from security, privacy, compliance, procurement, engineering, and fraud operations. Send the vendor the same questionnaire, request dated evidence, run a controlled comparison, document exceptions, and obtain written approval for residual risk. Keep the completed assessment with the contract and implementation records so it can be reused during renewal, audit, incident response, or migration planning. Related implementation guidance is available in the digital identity verification workflow and identity verification audit-trail checklist.