Wink

The biometric layer is entirely model work, combining face, palm, voice and device recognition into sub second identification of a person drawn from a population rather than confirmation of a claimed identity, and none of that exists without the models.

But the removal test does not empty the company, because it separately operates a payment gateway and network certified at the highest card industry tier, connected to major acquirers globally and processing hundreds of millions of transactions a year, adopted by independent sales organisations and fintechs in its own right. Two businesses sit here and only one is a model business, which is what holds this below the top grade.

Authentication completes in under a second with no human involved, which is inherent to the use case since a checkout queue cannot accommodate review. Real robustness is engineered in through multiple factors, combining face, palm, voice and device signals so a single weak reading does not determine the outcome, and that is a genuine design control rather than a claim.

What is absent is the failure path: nothing describes what happens to a customer the system cannot recognise, whether a fallback to card is automatic or awkward, or what recourse exists at the till when a match is wrong in a queue of waiting people.

No false match rate, false rejection rate, presentation attack detection result or independent benchmark was located, and the published figures measure speed rather than correctness, with sub second verification describing latency and nothing describing accuracy. The analyst recognition as best in class is a market assessment rather than a technical evaluation.

That gap matters more here than in onboarding verification because a one to many system fails in a direction that transfers money between strangers, and the rate at which it does so is the number a merchant and an acquirer both need.

Six enterprise partnerships are named and dated across a single year, and their calibre is the point. A major listed payments and financial technology group has integrated the biometric layer into its point of sale platform across every device in the estate without requiring new hardware, with rollout announced at the industry's largest retail conference.

A global terminal manufacturer has integrated through its own point of sale application, a second terminal maker is deploying on device processing across a worldwide Android fleet with a major chipset partner, a kiosk manufacturer covers self service, and an identity verification partner extends the work into agent driven commerce. The gateway business independently handles hundreds of millions of transactions annually. A research firm named the company best in class for biometric authentication delivered as a service in late 2025, and it won a leading industry showcase in 2023.

One question sits above all others here and nothing published answers it. Identification from a population only works against an enrolment set, and the commercial value of the model rises sharply if a customer enrolled once at one merchant is recognised at every other merchant on the network.

Whether enrolment is contained to the merchant that captured it or shared across the estate of a major point of sale platform is the difference between a convenience feature and a national biometric identity register assembled through checkout, and it is not addressed anywhere in published material.

The most consequential biometric architecture in this index and the least described. Identifying a customer from a population rather than verifying a claimed identity requires an enrolment database of face and palm templates searched on every transaction, and palm and facial geometry are permanent physical characteristics that cannot be reissued after a breach.

Privacy is asserted as a design principle and one deployment pattern genuinely supports it, with processing performed on the device at the edge in the terminal partnerships rather than in a central service. Against that, no retention schedule, deletion route or consent description was located, and no biometric privacy statute is named despite the company being headquartered in a state with its own biometric identifier law and selling into others with private rights of action attached.

Two named certifications are published and one of them is demanding. Certification at the highest card industry security level is the most rigorous tier available for an entity handling card data, assessed annually on site against a prescriptive requirement set, and a service organisation control attestation accompanies it. That combination places this ahead of most of the index.

What holds it below the top grade is the absence of anything covering the biometric half of the business, with no accredited presentation attack certification and no biometric specific standard named, so the assurance published addresses the payments rails rather than the face and palm data.

One genuine admission process has been passed and it is the operative one for the payments half of the business. Certification at the highest card industry security tier requires annual assessment by a qualified external assessor against a prescriptive control set, and it is the standard an acquirer will insist on before allowing a gateway near its rails.

A service organisation control attestation sits alongside it, and age verification is offered as a named use case, which is itself a regulated function. What is missing is the regime governing the other half: biometric capture is separately legislated in several of the markets served, with consent, notice, retention and deletion requirements and in places a private right of action, and no such statute is named.

Face recognition carries the best documented demographic error differentials in applied machine learning, and identifying someone from a population is a materially harder problem than confirming a claimed identity, so error rates that are tolerable in one to one verification compound in one to many matching. Palm recognition is far less studied across populations, and voice varies with accent, age and speech difference.

The consequence profile is unusual and cuts both ways: a customer who cannot be recognised is inconvenienced rather than excluded, since a card still works, but a false match in one to many identification means charging the wrong person's account, which is a harm with no equivalent in onboarding verification. No demographic performance data, accredited presentation attack testing or government evaluation result was located.

No guarantee, indemnity or falsifiable accuracy commitment was located. The consumer is in an unusual position: enrolment is voluntary and alternatives remain available, so nobody is excluded from paying by a failed match, which places this well above the identity vendors whose errors deny bank accounts.

But the person whose biometric is falsely matched to another customer's payment credential has no described route at all, and would in most cases discover it only through a disputed charge. Nothing describes correction, deletion on request, or how a person removes an enrolment captured at a merchant they never return to.

The hardware and distribution chain is named at every level, covering the acquirer whose platform carries the technology, two terminal manufacturers, a kiosk maker, a semiconductor partner supplying the on device processing platform and an identity verification partner for agent driven commerce, so a buyer can see exactly whose equipment and software the biometric layer runs inside. Models appear to be the company's own and no external provider is named.

What is not disclosed is the acquirer set behind the gateway, described only as major acquirers globally, and no subprocessor list or hosting arrangement was located for centrally processed deployments.

Integration reaches further down the stack than anything else in this index. The technology is embedded in a major acquirer's point of sale platform across its full device range with no additional hardware required, in a terminal manufacturer's own operating software through semi integration, in a second manufacturer's global Android fleet with processing performed on the device, in self service kiosks and customer facing displays, and at chipset platform level through a semiconductor partner.

Alongside that the company's own gateway connects to major acquirers worldwide. Being present in the terminal, the operating software and the silicon is a structurally different distribution position from selling an interface.

A meaningful architectural choice is published rather than left implicit. In the terminal partnerships biometric authentication is processed on the device at the edge rather than sent to a central service, which for face and palm data is the strongest privacy position available and materially changes what leaves the store.

Held at B because it is described for specific partnerships rather than as a platform wide guarantee, and no hosting provider, region selection or residency commitment is published for the deployments that do process centrally or for the gateway business.

No pricing, packaging or basis of charge is published for the biometric platform. One structural commercial fact is stated and it is unusual: the payment gateway can be owned outright by a merchant or rented, which is a real choice about who holds the infrastructure and is rarely offered either way.

That concerns the gateway rather than the biometric layer, and nothing indicates whether biometric authentication is charged per enrolment, per verification, per terminal or through the distribution partner.

Distribution runs through payments infrastructure rather than direct sales, so the buyers are acquirers and processors, terminal manufacturers, a chipset platform, kiosk makers, independent sales organisations and fintechs, and the reach that follows is very wide because each of those carries the technology to its own installed base. End environments named include retail, hospitality, quick service restaurants, venues and stadiums.

The limit is that the direct customer is infrastructure rather than a financial institution making its own decision, which is the Forter position, and no bank is named as deploying the technology for its own customers.