Veriff

More than a thousand signals per session across document authenticity, facial matching and liveness feed a verdict in around six seconds. Without the models, a capture form and database lookups are all that remain.

Automated AI decisions come back on the Essential plan, Plus and Premium add review by human specialists, and Enterprise offers configurable flows. No threshold says which sessions reach a specialist, and no reconsideration route is described for a genuine user who fails.

Accuracy near 99.6 percent is a falsifiable figure a buyer can test against its own funnel, and customers get insight into how the decision engine reaches a verdict. Methodology behind the accuracy figure, independent validation and drift monitoring are all unpublished.

Monzo, Bolt and Webull are among named customers, but none is tied to a figure, and the BB Finance case study describes stronger fraud protection without one.

End user data is used to develop and improve the service, including machine learning, under legitimate interest or contract, with an individual right to object. Cross session intelligence matches signals across encounters, and no published boundary keeps one customer's users out of models serving others.

The April 16, 2026 privacy notice covers verified end users with Veriff as processor, transfers under standard contractual clauses and rights including a request for human intervention in an automated decision. Retention is only as long as necessary under the customer agreement, with extended retention sold as a paid add on.

ISO/IEC 27001:2022, 27017 and 27018, a SOC 2 Type II attestation and Cyber Essentials are named on the May 2026 security page, with AES 256 encryption at rest and a trust center. Report access in the trust center is not stated, and no accredited presentation attack detection result is named.

Its identity verification products are certified under the UK Digital Identity and Attributes Trust Framework, a government scheme with a public register. Membership of a global card network's vetted fintech program adds a second admission, though neither covers the AI models themselves.

If 95 percent of genuine users pass on a first attempt, the remaining 5 percent across 12,500 document formats is unlikely to be spread evenly by skin tone, age or capture conditions. A March 31, 2022 blog post describes an in house face model, session quality review and a specimen database, but no breakdown by population or independent demographic evaluation is published.

A genuine user who fails is not told which signal produced the rejection, and no accuracy guarantee or remediation term is published. Published accuracy figures give a buyer something to measure against, but nothing published stands behind them in contract.

Proprietary, integrated technology is claimed, with device and network analytics as first party signal sources and a public subprocessor list linked from the privacy notice. Model providers, database verification sources and watchlist suppliers go unidentified in the material.

Delivery runs through an API and SDKs with public developer documentation and a hosted verification page, and a card network partner program puts it in front of issuers. Decisioning and case management platforms are not named as integration partners.

Daily encrypted backups run on Amazon Web Services with 90 day default retention, but no hosting region, residency option or tenancy model is published. Identity documents and biometrics move internationally, which makes where sessions are processed a material question.

Essential costs 80 cents per verification with a $49 monthly minimum, Plus $1.39 and Premium $1.89, with screening at 64 cents per session and a 15 day free trial. Enterprise pricing above 5,000 verifications a month is quoted.

Banking, fintech and lending, along with trading and crypto, each get their own regulatory framing and material, across more than 230 countries in 48 languages and dialects. Business verification arrives through the Vespia integration.