top of page
Website Quantum Payments (1).png

The $231 Billion Approval Problem: Why AI Fraud Detection Is Now a Revenue Strategy

Writer: Quantum Payments
Quantum Payments
Sep 11
8 min read

SEO title: AI Fraud Detection and the $231 Billion False-Decline Problem Meta description: False declines cost merchants far more than card fraud. Discover how AI fraud detection, intelligent routing and better approval metrics can lift payment success rates. Suggested slug: false-declines-ai-fraud-detection-payment-success-rate

Fraud prevention has traditionally been treated as a cost centre: reduce fraud losses, contain chargebacks and satisfy scheme requirements.

That optimisation problem is now incomplete.

In 2026, merchants are projected to lose more than US$231 billion to false declines : legitimate transactions wrongly rejected by payment or fraud systems : compared with approximately US$39.6 billion in actual card fraud losses. The gap is roughly six to one. Datos Insights also estimates that false declines cost the industry US$213 billion in 2025 and could reach US$297 billion by 2029.[1]

Fraud detection is no longer only about stopping bad transactions. It is about approving good ones accurately. For merchants, issuers and acquirers, precision has become a revenue strategy.

The six-to-one gap: fraud is visible, false declines are not

A fraudulent transaction is easy to classify through fraud reports, disputes, investigations, chargebacks or scheme notifications. A false decline usually disappears into an error message, an abandoned basket or a switch to a competitor, with no single system recording the revenue lost.

The global average false-decline rate is estimated at around 1.51 per cent of e-commerce sales. At scale, that becomes a material commercial leak and can depress future spend and trust.

BPC’s modelling makes the issuer-side economics equally clear. For a mid-sized issuer processing 10 million debit transactions per month, a deliberately conservative false-decline rate of 0.5 percentage points represents an average of US$160,000 in lost interchange revenue each year.[2]

On legacy, rules-based systems, BPC says an issuer can wrongly decline roughly 30 good transactions for every fraudulent one caught. Best-in-class real-time platforms can hold false declines below 0.5 per cent, while static-rule issuers can exceed 5 per cent during peak periods or when customers travel.

That is a revenue-performance variance, not just a risk-control variance.

Abstract neon ledger showing measurable fraud signals while legitimate payment value disappears into an untracked shadow

Why false declines stay invisible: the measurement gap

Banks have spent decades building frameworks to measure fraud. False declines do not produce the same visibility.

Fraud loss is usually reported as a defined financial outcome. False declines are scattered across customer service, conversion, interchange, repeat purchasing and payment operations. The result is an asymmetrical scorecard:

  • Fraud teams are judged on fraud prevented or fraud losses incurred.

  • Finance teams see completed sales, not blocked legitimate demand.

  • E-commerce teams see conversion rates without always knowing whether the payment layer caused the abandonment.

  • Acquirers may report approval performance, but not the true value of transactions their controls rejected.

  • Boards receive fraud and chargeback figures without an equivalent approval-accuracy measure.

This makes blunt controls appear successful. If more transactions are blocked, observed fraud may fall, but legitimate revenue may fall faster.

The right question is not, “How many suspicious transactions did we stop?” It is, “How accurately did we distinguish fraud from legitimate commerce?”

That means reporting fraud loss and false-decline rates together as joint performance indicators and attaching a revenue value to blocked legitimate spend.

The pincer movement from schemes and regulators

Payment businesses are being pushed from both directions.

On one side, schemes and acquirers are demanding tighter control of fraud and disputes. Visa’s Acquirer Monitoring Program, or VAMP, tightened the merchant fraud-and-dispute ratio threshold from 2.20 per cent to 1.50 per cent in the US, Canada, the EU and APAC from 1 April 2026.[5]

The ratio combines fraud and dispute counts and divides them by total settled transactions. That denominator matters. Blocking more traffic can reduce the number of transactions that settle. If fraud and dispute events do not fall proportionately, the denominator becomes smaller and the VAMP ratio can worsen.

On the other side, regulators are scrutinising payment processors that allow fraudulent merchants to operate.

On 4 September 2026, the US Federal Trade Commission announced a US$4.85 million settlement with Nuvei over allegations that the processor opened and maintained accounts for merchants it knew or should have known were engaged in deceptive activity, including tech-support scams. The proposed order includes enhanced merchant screening and monitoring obligations.[3]

On 8 September, the FTC sued 5967 Ventures, doing business as Humboldt Merchant Services. The company agreed to pay US$12 million and accept permanent restrictions under a proposed settlement after the agency alleged it helped more than 1,000 sham merchants. Some accounts reportedly exceeded 7 per cent chargebacks, compared with network monitoring thresholds of approximately 0.9 per cent to 1.5 per cent.[4]

These matters involve allegations and proposed or agreed settlements. The parties did not admit wrongdoing where applicable, and the Humboldt order remained subject to court approval at the time of reporting.

The broader signal is clear: stronger merchant screening is essential, but so is more precise decisioning at the transaction level.

Why rules decay and context is everything

Static rules are attractive because they are understandable, but legitimate behaviour changes faster than rules.

Travel, large purchases, first-time online transactions, subscription renewals and rapid repeat spending can all look suspicious without context. A customer who normally spends $50 may suddenly purchase a $500 flight. A returning subscriber may use a different device while overseas. An enterprise buyer may place a large order after months of research.

This is why AI fraud detection in payments must be adaptive rather than simply more aggressive. A model should assess the relationship between the customer, merchant, device, location, payment method, transaction history and behavioural pattern : not rely on one isolated trigger.

It should also learn from outcomes. A successful step-up authentication, an approved repeat purchase or a confirmed customer relationship should improve future decisioning.

What AI fraud detection and intelligent routing change in practice

The best AI payments strategies do not treat fraud prevention, authorisation and routing as separate functions. They connect them into one decision layer.

Real-time decisioning signals for device, behaviour, authentication and issuer context converging into one approval decision

The agentic wildcard

The false-decline problem will become more complex as AI agents begin initiating legitimate transactions.

On 9 September 2026, BLIK, PayU and Juo announced Poland’s first fully agent-executed purchase: a PLN 19.99 hand cream bought within pre-authorised consent limits. The transaction was approved using a six-digit BLIK code confirmed in the customer’s banking application.[6]

Yet a fraud model that treats “non-human” traffic as inherently suspicious could classify that transaction as abnormal and increase false declines. Agentic payments require payment systems to recognise the difference between unauthorised automation and authorised automation.

The industry is beginning to address this at network level. On 10 September, Ant International, Mastercard and Visa announced a Know-Your-Agent interoperability collaboration through MAS-convened BuildFin.ai. The initiative is intended to support agent identity, operator traceability, shared certification requirements and continuous transaction monitoring.[7]

For merchants, agent traffic should become a distinct, legitimate traffic class with its own identity, consent, behavioural and monitoring signals.

The emerging model is not “trust the bot”. It is “verify the agent, the operator, the customer’s intent and the boundaries of authority”.

What merchants should do now

The commercial priority is to govern approval accuracy with the same seriousness applied to fraud loss.

Merchant checklist

  • Measure approval accuracy alongside fraud loss: Make both joint board-level metrics, with clear owners and targets.

  • Segment false declines: Analyse rates by issuer, country, channel, device, customer cohort and payment method.

  • Quantify blocked legitimate spend: Attach revenue, margin and potential customer lifetime value to every meaningful false-decline segment.

  • Move beyond static rules: Introduce real-time adaptive decisioning that can assess customer and transaction context.

  • Enrich authorisation data: Use 3DS 2.0, device intelligence, behavioural signals, customer history and relevant transaction metadata.

  • Test smart retries and intelligent routing: Apply recovery logic to soft declines without repeatedly retrying hard declines.

  • Monitor scheme ratios and denominator risk: Track fraud-and-dispute ratios against VAMP thresholds and understand how blocking traffic changes settled transaction volumes.

  • Classify agent-initiated payments properly: Treat authorised agents as a legitimate traffic class with consent, identity, spending limits and continuous monitoring controls.

  • Review processor screening and oversight: Reassess onboarding, merchant monitoring and escalation obligations in light of the recent FTC enforcement actions.

  • Create one operating view: Connect checkout, fraud, authorisation, routing, disputes and reconciliation data so decision quality can be improved continuously.

As businesses evaluate this operating model, AI-driven payment operations and touchless finance provide a useful reference point. Merchants operating in more complex sectors should also consider how risk classification and scheme pressure affect high-risk industries.

The next generation of payment performance will be won by the provider that can distinguish suspicious from simply unfamiliar : and do so quickly enough to approve legitimate commerce.

Fraud prevention is becoming a precision discipline. The commercial prize is a higher payment success rate, stronger customer trust and revenue that would otherwise never reach settlement.

Authoritative sources

Daily handover to Sonny

The post is scheduled in Wix for 07:30am AEST on 11 September 2026, so both social posts go out after it is live.

LinkedIn post 1: 8:08am AEST

ANGLE: Data-led contrarian thesis : “the fraud paradox”

Copy:

Merchants are losing roughly six times more to false declines than to actual card fraud.

That should change how fraud teams define success.

A legitimate customer wrongly blocked at checkout is not a “safe” transaction. It is lost revenue, weaker customer trust and potentially a permanent shift in where that customer spends.

Yet many fraud programmes still optimise for the wrong problem: reduce fraudulent approvals at almost any cost.

At the same time, schemes and acquirers are tightening pressure on merchants to reduce fraud and disputes. Visa’s VAMP threshold is now 1.50 per cent in the US, Canada, the EU and APAC.

Regulators are applying pressure in the opposite direction too : scrutinising processors that allegedly allowed fraudulent merchants and sham businesses to operate. The recent FTC matters involving Nuvei and Humboldt show that loose merchant screening and oversight carry serious consequences.

The answer is not to block more traffic.

It is to make better decisions.

Measure fraud loss and false-decline rates together. Enrich authorisation data. Use adaptive AI decisioning, intelligent routing and smart retries to separate genuine risk from unfamiliar but legitimate behaviour.

Fraud prevention is now a precision problem, not a volume problem.

Relevant tags: Datos Insights, Visa, FTC, merchants, acquirers

Visual concept: Hero image : neon dashboard contrast of blocked legitimate traffic versus concentrated fraud losses.

LinkedIn post 2: 3:23pm AEST

ANGLE: Operator/practitioner checklist : “how to govern approval accuracy”

Copy:

Approval accuracy needs to become an operating metric, not a hidden payment-system outcome.

For payments, risk and finance leaders, the practical starting point is instrumentation:

  • Make approval accuracy a board metric alongside fraud loss.

  • Segment false declines by issuer, country, channel, device and payment method.

  • Quantify the revenue attached to legitimate spend being blocked.

  • Report fraud and false-decline rates together.

  • Track VAMP performance and understand how blocking traffic can shrink the settled-transaction denominator.

  • Use real-time adaptive decisioning instead of relying only on static rules.

  • Test smart retries and intelligent routing for soft declines.

  • Treat agent-initiated payments as a legitimate traffic class, with consent, identity and monitoring controls.

The objective is not simply more approvals. It is more accurate approvals: protecting the payment ecosystem without turning legitimate customers away.

Relevant tags: Quantum Payments, risk leaders, e-commerce, retail, payment orchestration

Visual concept: Inline image showing real-time decisioning signals converging into a single approval decision.

 
 
bottom of page