AI identity verification isn’t one-size-fits-all

Artificial intelligence has become the default answer to online identity verification. Machine learning models analyse behavioural patterns in real time, biometric systems match faces in milliseconds, and automated engines assess risk at scale. The industry narrative is compelling: faster onboarding, lower fraud and frictionless digital journeys.

However, at checkout, identity is increasingly inferred rather than confirmed. Systems assess patterns and assign a likelihood score instead of directly verifying the person making the payment. When those assessments are wrong, legitimate customers are declined, which affects approval rates, customer trust and ultimately revenue.

When Probability Determines Identity

At checkout, inconsistency translates directly into friction and AI-driven verification systems are only as reliable as the data on which they are trained and the assumptions embedded within their design. When legitimate customers are flagged, asked to retry or subjected to additional checks, conversion rates suffer and confidence erodes. If training datasets are incomplete or historically skewed, the resulting models risk reproducing those imbalances at scale with limited recourse to self-correct. 

An error rate that appears negligible in testing can become material when applied across millions of transactions. Research has also demonstrated that these errors are not evenly distributed. Facial recognition systems have been shown to produce significantly different outcomes across demographic groups.

The Commercial Impact of False Rejection

False declines and verification failures carry real economic costs. For consumers, repeated rejection creates frustration and distrust, particularly when the reason for the failure is opaque. Many do not persist through additional authentication layers. Instead, they abandon the transaction or switch providers.

For merchants and payment providers, the implications extend further. Automated systems that generate uncertainty often trigger manual reviews or secondary checks, increasing operational overhead while slowing transaction flows. In response, organisations frequently layer additional tools on top of existing systems in an attempt to balance fraud risk and acceptance rates. Complexity increases, but certainty does not.

At scale, even incremental declines in authorisation rates translate into measurable revenue loss. Identity verification is therefore not simply a compliance function. It is a core driver of conversion, dispute management and customer lifetime value.

Reintroducing Determinism into Digital Environments

In physical retail environments, card-present transactions using chip and PIN dramatically reduced fraud by confirming both possession of the card and cardholder intent. The process is deterministic. The individual enters a PIN linked directly to a regulated card, creating confirmation rather than inference.

Online transactions remain predominantly card-not-present, where neither physical possession nor intent is validated in the same way. This structural distinction helps explain why card-not-present fraud remains persistent and why digital identity decisions rely so heavily on behavioural modelling and biometric interpretation.

Introducing card-present verification into online channels applies proven security principles to digital commerce. By enabling consumers to authenticate transactions using their physical card and PIN, merchants can replicate the certainty of in-store transactions within e-commerce journeys. The outcome does not depend on facial interpretation, lighting conditions or image quality. It confirms possession and intent directly.

In practical deployments, introducing this deterministic layer into online flows has been associated with improvements in authorisation rates of between 5% and 10%, alongside reductions in disputes and chargebacks. More importantly, it reduces reliance on probabilistic identity scoring at the most commercially sensitive moment of the transaction.

AI As Part of the Stack

Artificial intelligence remains a powerful tool within modern identity infrastructure. Behavioural analytics and anomaly detection can identify suspicious activity earlier in the transaction lifecycle and allow risk teams to focus resources where they are most needed.

The challenge arises when AI becomes the sole arbiter of identity at checkout. No predictive model, regardless of sophistication, eliminates uncertainty entirely. A resilient identity framework acknowledges this and incorporates mechanisms capable of delivering certainty when probabilistic systems reach ambiguity.

A hybrid approach offers a more balanced path. AI can assess behavioural signals and detect anomalies, while deterministic verification methods can provide clarity when needed. This combination reduces false rejections, protects revenue and preserves consumer trust without introducing demographic inconsistency.

Security Without Exclusion

As digital commerce continues to expand, identity verification systems must deliver security without exclusion and performance without bias. Systems that work well for some but inconsistently for others undermine both fairness and commercial outcomes.

AI is a powerful component of the identity ecosystem, but it is not universally sufficient. By pairing intelligent risk detection with proven, intent-based verification standards, the payments industry can move beyond probability alone and build identity systems that deliver certainty, consistency and trust at scale.