Key Takeaways
- Veriff: U.S. consumer losses from identity fraud and scams reached $38 billion in 2025, increasing pressure on banks to combine face matching, liveness detection, and device intelligence.
- A composite regional bank evaluating verification solutions replaced document-only review with an API-based flow that routed high-risk applications to same-day analyst review.
- Biometric evidence operates within a layered authentication process; NIST SP 800-63B-4 treats biometrics as one factor used with a physical authenticator, not as a standalone replacement for multi-factor authentication.
Biometric verification confirms that an applicant's physical characteristics match a claimed identity. For financial-account onboarding, platforms such as Veriff combine face matching, liveness detection, document checks, and device intelligence to identify suspicious remote applications.
The Challenge
A regional bank sees a familiar pattern: a new customer uploads a convincing driver's license, passes basic data checks, and opens an account remotely. The document looks legitimate, but the applicant is a synthetic identity, a fraudulent profile combining real and fabricated personal information, assembled from a genuine Social Security number, invented contact details, and an AI-generated portrait.
This composite scenario reflects a broader change in financial fraud. According to Javelin Strategy & Research's 2026 study, U.S. consumers lost $38 billion to identity fraud and scams in 2025, while new-account-fraud losses increased 13% year over year to $7 billion. Separately, the Sumsub Identity Fraud Report 2025-2026 recorded a 2.7% identity-fraud rate in financial services in 2025, and synthetic identities represented 21% of identified first-party fraud during the reporting period. These figures highlight different scopes of the problem, from total consumer financial impact to specific verification failure rates.
Document validation alone has trouble resolving that combination. Optical character recognition, or OCR, the conversion of text in an image into machine-readable data, can extract a name, birth date, and document number from a JPEG or HEIC image, but extraction does not prove that the person holding the phone owns the identity. Deepfake video, digital injection, printed face masks, and replayed recordings create additional problems.
The bank's existing workflow illustrates the operational cost. Its mobile application sends identity images to a document-checking service through a REST API, an interface that lets software exchange requests and responses over HTTP, then writes the result to a PostgreSQL onboarding database. Applications with mismatched fields enter a manual queue in the case-management system. Analysts switch among the document image, core banking records, and device logs, sometimes leaving an application unresolved for several business days.
Coverage from Help Net Security describes a similar erosion of trust at the identity layer, where realistic digital artifacts make visual inspection less dependable. The issue is not merely better counterfeit documents. It is the ability to coordinate forged evidence across several stages of an onboarding session.
The Approach
The bank designs an illustrative verification flow around layered evidence. Its evaluation includes platforms from Jumio, Socure, Onfido, and Veriff, with testing focused on API behavior, supported identity documents, presentation-attack detection, and regional data-processing options.
During account opening, the customer photographs an identity document and records a short selfie sequence. The service checks document security features, compares the portrait with the live face, and applies liveness analysis, a process designed to determine whether biometric input comes from a present person rather than an artifact, to identify screen replays, masks, image injection, or manipulated video. Device intelligence, meaning risk signals collected from the applicant's hardware and network context, adds IP reputation, emulator indicators, operating-system details, and application-velocity patterns.
Risk rules then determine what happens next. A low-risk applicant with consistent document, facial, and device signals can continue to account funding. A mismatch sends the case to an analyst. Higher-risk activity, such as repeated applications from the same device with different identities, can trigger additional evidence collection or rejection according to the bank's policy.
That said, biometrics are not treated as a universal answer. NIST SP 800-63B-4 positions biometric comparison within multi-factor authentication and pairs it with a physical authenticator. For an existing customer changing a payment destination, that could mean facial verification plus a registered FIDO2 security key or device-bound passkey.
The Implementation
The composite rollout spans several months and uses an illustrative team of six: a product manager, two application engineers, a fraud analyst, an information-security specialist, and a compliance lead. Specific implementation metrics for this scenario were not disclosed, so the staffing model and workflow represent a practical example rather than a single reported customer deployment.
During initial integration, the engineers connect the mobile onboarding application to the vendor through REST endpoints and signed webhooks. A webhook is an automated message one system sends to another when an event occurs. Verification decisions are stored in PostgreSQL, while images and audit artifacts receive encrypted object-storage references rather than being copied into the core banking database.
Midway through implementation, testing exposes a problem with webhook retries. Duplicate status events create multiple review cases for one application. The team adds idempotency keys, identifiers that prevent a repeated request from creating a second transaction, along with event timestamps and a unique verification-session identifier before forwarding decisions to the case-management queue.
The evaluation also included comparative testing for passive liveness, document coverage, and response-payload detail. Passive liveness evaluates a capture without requiring the applicant to perform a prompted movement. The fraud team maps vendor reason codes, such as face mismatch or suspected presentation attack, to internal case categories instead of relying on a single pass-or-fail field.
Using ISO/IEC 30107-3 concepts, the bank evaluates spoof resistance through presentation-attack detection testing. A presentation attack is an attempt to fool a biometric system with an artifact or manipulated biometric sample. A Biometric Update report on NIST's mobile driver's license work also points to a future in which cryptographically signed identity credentials supplement photographed documents.
The Results
Following the controlled rollout, the organization reported within the composite scenario that straightforward applications moved through the automated path with fewer analyst touches. Reviewers also received consolidated evidence, including face-match results, liveness signals, device attributes, and document findings, rather than opening several systems to reconstruct a session.
Exception handling shifted from an open-ended document review to a reason-coded queue. A suspected replay attack could go directly to the fraud team, while an unreadable barcode could prompt the applicant to recapture the document. That distinction reduced back-and-forth for cases caused by image quality rather than suspected fraud.
The organization did not disclose verified conversion, fraud-loss, or false-rejection figures. Its observable change in the scenario was procedural: analysts could resolve many flagged applications on the same business day because the case record included the relevant evidence and session history.
Meanwhile, Infosecurity Magazine has documented the sharp rise of AI-assisted and deepfake fraud. That trend makes false-acceptance rates (the share of fraudulent attempts incorrectly approved) important, but false rejections, which occur when legitimate applicants are incorrectly denied, deserve equal attention. A system that blocks legitimate customers with older cameras or limited mobility can shift cost from fraud operations to call centers and branch staff.
Lessons Learned
Duplicate webhooks became the most concrete integration lesson. Without idempotency controls, one verification session generated several analyst cases, which distorted queue volume and complicated audit records.
Reason codes also proved more useful than a binary score. By mapping document tampering, face mismatch, replay suspicion, and device anomalies separately, the bank could apply different review procedures rather than treating every exception as identity fraud.
Finally, camera testing affected accessibility. Testing only recent flagship phones missed glare, focus, and low-light problems common on older devices. Expanding the device matrix helped the product team adjust capture instructions and offer an assisted-review path when automated liveness could not reach a reliable decision.
Broader Applicability
Credit unions, payment providers, brokerages, and insurers can adapt the same pattern by connecting biometric evidence to existing know-your-customer controls, or KYC, case-management platforms, and authentication systems. Smaller teams may start with remote onboarding, then extend step-up verification to account recovery or high-risk payment changes after reviewing false-rejection data.
How long does a biometric verification implementation take?
A focused onboarding integration often spans several months, depending on mobile software development kit, or SDK, work, document coverage, and compliance review. Buyers should reserve time for webhook testing, device-matrix validation, reason-code mapping, and production monitoring rather than treating the REST API connection as the full project.
What is the difference between face matching and liveness detection?
Face matching compares a selfie or video frame with the portrait on an identity document. Liveness detection seeks to confirm that a capture features a live individual, counteracting threats such as printed photos, screen replays, masks, or injected videos; ISO/IEC 30107-3 provides a testing basis for this form of presentation-attack detection.
Is biometric verification suitable for a small financial team?
It can be, particularly when a vendor provides hosted capture, configurable rules, and structured webhook responses. A small team should begin with one workflow, such as new-account opening, and track manual-review volume, false rejections, abandonment, and decision latency before expanding to account recovery or transaction approval.
⬇️