Jamie Dimon’s much-circulated remark in October 2025 comparing fraud to cockroaches resurfaced at a moment when several trade finance failures made headlines. His comment drew attention to an underappreciated corner of the financial system: supply-chain financing, particularly factoring.
According to Resistant AI, despite its reputation as a niche activity, factoring supports the flow of goods behind almost everything consumers purchase, yet the mechanics and risks behind it rarely enter the mainstream conversation.
At its simplest, factoring allows a seller to unlock near-instant liquidity by selling an invoice—normally payable within 30 to 120 days—to a specialist financier at a discount. The factor pays upfront and later retrieves the full amount from the end customer. As relationships mature, documentation often becomes lighter, moving from bundles of proofs to little more than lists of invoice numbers and amounts, or in extreme cases, only totals.
That convenience, however, opens a door to disputes, credit risk and, frequently, fraud. Industry practitioners quietly suggest that around 5% of financed invoices are, at best, “administratively deficient”, meaning they may be fabricated or manipulated.
A recent example came from First Brands, an automotive parts supplier that collapsed after being caught inflating and duplicating invoices. As Bloomberg Opinion columnist Matt Levine detailed: “First, in many instances, the amount set forth on a factored invoice did not accurately reflect a customer’s order, without any apparent reason for the discrepancy… Second… purported invoices representing customer orders were created… even though the Debtors’ books and records… do not reflect such customer invoices existed. Third… the same invoice was factored more than once to different third-party factors.”
Consortium-style initiatives aimed at solving duplicate financing have existed for years, but their effectiveness is limited by low participation and data-sharing challenges. The deeper issue is that detecting fraud requires more than checking an invoice—it requires understanding the business relationship that supposedly produced it. Invoices are created by the party seeking financing and can be replicated or manipulated with relative ease. Criminals who intend to deceive will simply embed fake invoices into the same workflow as legitimate ones.
The real insight often lies in third-party documentation such as purchase orders, shipping paperwork and delivery confirmations. These artefacts are harder to forge consistently because each originates from a different actor in the supply chain. Historically, reviewing all this material was too labour-intensive to scale, which is why many factors defaulted to financing spreadsheets rather than scrutinising full documentation packs. But AI has shifted this cost equation dramatically, allowing risk teams to analyse large bundles of semi-structured documents in minutes.
Fraud in this space is rarely sophisticated; it is repetitive and becomes bolder over time. As Levine notes, “after you’ve been doing it for a while, everyone will get comfortable, and eventually you’ll just send mass emails like ‘hey everyone, time to fake the invoices again.’” This repetitiveness makes patterns detectable—if firms are willing to inspect beyond the invoice.
Modern AI systems now allow factors to request full documentation rather than simple lists of numbers. That requirement alone discourages manipulation. Once the bundle arrives, tools such as Resistant AI can classify documents, identify the issuing systems, and confirm whether a purchase order genuinely originates from the claimed customer. Any reasonable LLM can extract invoice data, cross-match identifiers and highlight mismatches or omissions.
Once the documentation is sorted and verified, AI can run deeper integrity checks, examining fields such as PO numbers, dates, quantities, line items, issuer details and pricing. Even subtle modifications—altered digits, inconsistent metadata, mismatched shipping confirmations—become detectable. Fraudsters may change invoice values, tweak PO numbers, or attempt to mask reuse, but forged artefacts tend to diverge from historic patterns. Conversely, countless legitimate edits occur within real trade documentation. Modern systems can recognise these and minimise false positives, avoiding unnecessary strain on risk teams.
With verified operational evidence in hand, a factor can make a more informed risk assessment. They can compare new transactions with established patterns and evaluate whether the amounts, customers or delivery rhythms align with the seller’s history. This approach does not need to eliminate fraud entirely—no system can—but it does make long-term manipulation significantly harder to sustain. Quarterly reviews and light-touch audits cannot offer the same level of defence.
AI, applied with careful scepticism, offers a practical means to illuminate areas of trade finance that have long operated on trust. It helps risk teams validate substance over surface-level detail and raises the barriers for those attempting repeated deception. In doing so, technology increases pressure on the “cockroaches” of invoice fraud—at least until they evolve, and the cycle continues. But the industry cannot accept the old notion that the optimal amount of fraud is anything other than zero.
Find more on RegTech Analyst
Copyright © 2026 FinTech Global
