Identity verification in financial services was built around a simple assumption: the criminal was a stranger. Checks at onboarding established that a new customer was who they claimed to be, and once that gate had been passed, the relationship was treated as trusted. Synthetic identities, cloned voices and forged documents undermine that assumption. The threat is no longer a fraudster using someone else's credentials; it is a convincing imitation of the genuine customer, produced at the moment an institution is asked to act.

That shift changes where identity proofing must sit in the customer lifecycle, how it is evidenced, and who absorbs the loss when a payment was technically authorised but not genuinely intended. For the Global Finance Awards 2026, the distinction matters: the category rewards operating controls that can be demonstrated and measured, not tooling described in policy language.

The risk has moved down the lifecycle

Onboarding remains an important control. It is where an institution first establishes a customer's identity, screens for sanctions and politically exposed persons, and sets the baseline against which later activity is judged. But the economics of impersonation have changed. A criminal who can convincingly imitate an existing customer does not need a new account; they need only reach a moment where the institution will move money, change a beneficiary, reset credentials, update contact details or settle a claim.

Those moments are high value, often time-sensitive, and the customer expects them to be quick. They are also where re-verification is thinnest. A customer authenticated at login may be able to add a payee, amend a standing instruction or recover a locked account with little further challenge. Contact-detail changes are especially sensitive: an altered phone number or email address can redirect every later out-of-band confirmation to the fraudster.

Identity proofing should therefore be a continuous control applied at the moment of a high-value instruction, not a one-off gate. That does not mean interrogating customers at every click; it means deciding deliberately which instructions carry enough risk to justify a fresh check, and designing that check to withstand a determined impersonator.

Single-modality checks are being gamed

Each familiar verification method has known weaknesses. Liveness detection can be challenged with injected video or presentation attacks. Document verification depends on the quality of the underlying document and the issuer's own controls. Voice biometrics can be tested with cloned or synthesised speech. None of these methods is useless, but each is a single layer, and a single layer is a single point of failure.

Institutions are therefore layering signals. Device reputation can indicate whether an instruction originates from hardware previously associated with the customer or from an unfamiliar environment. Behavioural patterns, such as the rhythm of typing or the sequence in which screens are used, can flag divergence from an established profile. Transaction context — amount, beneficiary, channel and time of day — shows whether an instruction fits the relationship.

Independent out-of-band confirmation carries particular weight. A callback to a number held on file, rather than one supplied during the transaction, or confirmation through a separate channel the customer has previously registered, is harder to subvert than a check performed inside the session the fraudster controls. Layering matters less for any individual signal than for the difficulty of defeating several independent ones at once.

Liability and dispute resolution are the practical test

When a customer is manipulated into authorising a payment, the institution faces a question that is legal and operational at once. The instruction may have been technically authenticated and still not genuinely intended. Under evolving reimbursement expectations in several markets, the outcome increasingly turns on what the institution can show about the checks it applied, when and how.

That evidence requirement changes control design. A control that cannot be reconstructed after the event is hard to rely on in a complaint or a supervisory review. Institutions need records of which signals were assessed, what they indicated, what challenge was issued and how the customer responded — and, where no step-up check was triggered, why the instruction proceeded.

This is where the distinction between authentication and authorisation becomes commercially significant. Showing that a password was entered or a biometric matched is not the same as showing that the institution took reasonable steps to establish that the person giving the instruction was the customer. The second is a higher bar, and it determines who absorbs the loss.

Fraud operations and the contact centre are control environments

Many impersonation attempts do not begin with a payment instruction. They begin with a conversation. A caller claiming to be a customer asks for a password reset, a change of address or an urgent transfer, and the interaction is handled by a member of staff under time pressure. The contact centre is a control environment, not merely a service function.

Step-up authentication scripts, callback procedures and escalation routes need the same design discipline as any other control. They should specify what staff must verify, through which channel, and what they must not accept from the caller. They should also be tested against attempted social engineering, not only against system failure: rehearsing how a script holds up when a caller is insistent or emotionally manipulative reveals weaknesses a procedural review will miss. Fraud operations and contact centre leadership should own that testing jointly.

Vendor dependency deserves scrutiny

Layering is only as strong as the independence of the layers. Where several checks draw on the same provider, the same data source or the same infrastructure, a single outage or compromised dataset can disable multiple defences at once. An institution may believe it has four controls when in practice it has one, replicated.

Assessing concentration in identity tooling is therefore part of control design: mapping which checks depend on which providers, understanding the fallback when a provider is unavailable, and testing what happens when a layer is removed. That exercise exposes how much of a control stack's apparent depth rests on shared foundations, but it is the only way to know whether the layering is real.

What the award category rewards

In the Global Finance Awards 2026, the test is not which institution has bought the most sophisticated tooling. It is whether the controls operate. The differentiator is evidence: measured false-positive and false-negative performance, documented challenge testing, and audit trails that survive both supervisory review and a customer complaint.

That standard favours institutions that can describe their controls in operational terms. Which instructions trigger re-verification? What signals are assessed, and how are they weighted? How often are the checks tested against attempted impersonation, and with what results? When a step-up challenge fails, what happens next? When a customer disputes an instruction, what record can be produced, and how quickly?

Policy language rarely answers those questions. A statement that an institution applies layered identity verification says little about whether the layers are independent, whether they are applied at the right moments, or whether anyone has measured how they perform. The category rewards the harder evidence: that the control was designed, tested, and can be reconstructed after the event.

The broader point is that identity proofing has become a lifecycle discipline. Onboarding establishes the baseline, but the decisive tests now occur later, when money moves and access changes. Institutions that treat those moments as control points, and can prove what they did when it mattered, will be better placed to prevent impersonation and to defend their position when a coerced instruction is disputed.

← PreviousBack to Articles