One column leaves evidence. Two don’t

by Raul Liive, Director of Product, Marketplaces. Veriff

Share this post

Fraud prevention must account for lost customers and operational costs alongside fraud losses, with shared oversight of verification friction and risk.

The argument in 30 seconds:

  • Fraud loss is counted. Friction loss and review cost are inferred or buried — so the sector optimises the one column that generates paperwork.
  • 68% of European consumers have abandoned a financial application mid-onboarding. No fraud model ever saw them.
  • Verification drop-off doesn’t just lose customers. It inflates the CAC of every customer you do acquire.
  • Agentic commerce protocols solved authorisation. They did not solve identity assurance. A hijacked account signs perfectly valid mandates.

In 2025, 4.18% of identity verification attempts were fraudulent, roughly one in 25, and in financial services, above 5.5%. That is Veriff’s data, and the kind of number our sector produces well. Ask the same firm what its stack cost in lost customers last year, and the answer gets vague.

You will get an accurate answer from any UK payments firm, one column at a time. The fraud team reports a loss rate. Growth reports conversion. Operations reports queue depth. Every number is correct and optimised. Nobody owns the relationship between them.

This is usually diagnosed as a measurement gap. It isn’t. Approval-rate discipline in payments is mature, and Adyen’s 2026 fraud report already states that the cost of stopping fraud is increasingly measured in lost lifetime value rather than prevented loss. The sector knows.

What stops it from acting is asymmetric observability. A fraud loss arrives with paperwork: a chargeback, a write-off, a reimbursement claim, a named owner. A lost good customer arrives with nothing. The applicant who gave up at document upload never tells you they were legitimate. One column generates evidence. The other two generate absence. Organisations optimise what leaves a trace.

Three columns, not one

Column one is fraud loss. Column two is friction loss: everyone who abandoned, was falsely declined, or churned afterwards. Column three is operating cost: review, re-verification, support, latency.

Independent industry data points in the same direction. Sumsub’s global fraud rate fell from 2.6% to 2.2%, even as multi-step attacks rose to 28% of identity fraud. Entrust, analysing more than a billion verification events, likewise found that deepfakes and injection attacks are becoming more prevalent. The signal is not simply more fraud. It is fewer, more sophisticated attacks. In our own data, physical document fraud dropped 13% while digitally presented media became 300% more likely to be AI-generated or altered. Fewer attacks, better ones.

Column two is routinely mis-specified as false positives alone. That captures only the users your system decided about and got wrong. The larger loss is the users it never decided about. Signicat’s Battle to Onboard research puts financial application abandonment at 68% of consumers, up from 40% in 2016. Almost none of that is your fraud model’s accuracy. It is credential availability and retry loops: asking a user to re-upload a document makes abandonment three times more likely.

This is where it gets expensive, because acquisition cost is calculated against funded accounts, not applications. Every applicant who abandons at document upload was paid for and delivered nothing. That cost is redistributed across those who did complete, inflating the CAC of every one of them. Marketing is then asked to explain a number a risk threshold created. Multiply your verification-step abandonment rate by your own CAC. That figure appears on nobody’s fraud dashboard.

None of this data is missing. Step-level analytics show where applicants drop, and the fixes are dull: inline image-quality validation, specific failure reasons, passive liveness instead of active. The numbers exist, but no one who attends the fraud committee owns them.

Why the seam is about to matter more

Industrialised AI fraud, first. Entrust’s 2026 report, drawn from over a billion verification events, found deepfakes now account for one in five biometric fraud attempts, with injection attacks up 40% year on year. The instinct is to send more cases to human review. But most people cannot reliably distinguish deepfakes. That response grows column three without shrinking column one.

Then agentic commerce. AP2’s signed mandates give merchants a cryptographic record of what the user authorised, and networks can deny anything outside the mandate’s scope. That solved authorisation. It did not solve identity assurance. A mandate proves a credential holder consented. It does not prove that the holder is who they claimed at onboarding, or still controls the account. Impersonation accounts for more than 85% of fraud attempts in our data, and Sumsub expects autonomous agents running end-to-end attacks to overwhelm systems built on isolated checks. A hijacked account signs valid mandates all day.

What the sector should do

UK context sharpens this. Liability for APP scams is now shared 50/50 between sending and receiving PSPs, and the PSR moves to formal consultation in December 2026. The cost of a miss went up, which is exactly why the columns cannot be run separately.

Raul Liive, director of product, marketplaces. Veriff

Put all three columns in one paper, not three dashboards and three committees. One page: fraud loss, abandonment, false-decline rate, review cost per check, p95 latency.

Give column two a named owner with authority over thresholds. It will always be partly an estimate. That is an argument for accountability, not for ignoring it.

Review identity assurance at high-risk moments. Re-authenticate at moments of consequence: payout changes and password resets.

The firms that win will not have the least fraud, or the least friction. They will be the ones who see both numbers on one page and know who to call when they diverge.

Logo-1024-x-576
Article by Veriff

Membership

Merchant Community Membership

Are you a member of The Payments Association?

Member benefits include free tickets, discounts to more tickets, elevated brand visibility and more. Sign in to book tickets and find out more.