Preventing invoice fraud in accounts payable

The email looks routine. A long-standing supplier writes to say it has changed banks and attaches a letter on familiar letterhead with the new account details. The next invoice arrives on time, matches an open purchase order and is approved without a second look. By the time the real supplier calls to ask why it hasn't been paid, the money has moved through several accounts and is unlikely to come back.

Scenarios like this are now among the most common threats facing finance teams. In the Association for Financial Professionals' 2026 Payments Fraud and Control Survey, 76% of US organizations reported attempted or actual payments fraud in 2025, and 74% were targeted by business email compromise, up from 63% a year earlier (AFP; Nacha). Fraud from inside the organization adds to the exposure: the Association of Certified Fraud Examiners estimates that organizations lose about 5% of annual revenue to occupational fraud, with a median loss of $104,000 per case across 2,402 cases studied in its 2026 Report to the Nations (Dean Dorton summary of ACFE 2026).

Accounts payable sits at the center of this risk. It is the function that turns a document into a payment, and every weakness in supplier data, approval policy or verification becomes a route to cash. Preventing invoice fraud is therefore less about a single tool than about designing the AP process so that a fraudulent request must defeat several independent controls at once.

How invoice fraud gets in

Invoice fraud takes four broad forms, and each exploits a different gap.

The first is the fictitious invoice. A fraudster sends an invoice for goods or services that were never delivered, often for modest amounts that fall under approval thresholds, to a company with a busy AP inbox. Generative AI has made these documents harder to spot. Convincing logos, plausible line items and correct tax formatting can now be produced in minutes, so a document's appearance is no longer a reliable signal.

The second is the duplicate invoice, which is not always malicious. Suppliers resend invoices when payments are slow, and the same invoice can arrive as a PDF by email and on paper by post. Deliberate duplicates are harder to catch because they are designed to look new: an invoice number with an added suffix, a slightly different date or an amount changed by a few cents. Basic ERP checks that compare only invoice numbers miss them.

The third, and often the most costly, is vendor impersonation and bank-detail change fraud, the scenario described above. Attackers compromise or spoof a supplier's email account, study real correspondence and request a change of payment details at the moment a large payment is due. Because the invoice itself is genuine, matching controls pass. The fraud sits entirely in the supplier master data.

The fourth is insider fraud. Employees with access to supplier records and payment runs can create shell suppliers, approve their own invoices or redirect payments. These schemes tend to run for months before discovery, and they thrive where one person can both maintain supplier data and release payments.

The controls that make the difference

The most effective defenses are not new. What has changed is that they now have to hold up at speed, across more channels and against more convincing attacks.

Treat supplier master data as a control point

Most serious AP fraud ends with a payment to an account the fraudster controls, so the supplier master file is where prevention matters most. Every change to bank details should be verified out of band. In practice, that means calling the supplier on a number already held on file, never one supplied in the change request. The change should be approved by someone other than the person who entered it, and the first payment to a changed account should be flagged for review. New suppliers deserve the same discipline: validate tax IDs and bank account ownership before the first invoice is paid, not after.

Separate duties and enforce approval policy

No single person should be able to create a supplier, approve its invoices and release payment. Segregation of duties sounds basic, but it erodes quietly during staff shortages, system migrations and year-end pressure. Approval rules should follow a documented delegation-of-authority matrix and be enforced by the system rather than by convention. Splitting a large invoice into several smaller ones to stay under a threshold is a familiar tactic, and one that controls can catch when they look across invoices rather than one at a time.

Match, then verify

Two- and three-way matching against purchase orders and goods receipts remains one of the strongest protections against fictitious invoices, because a fraudster rarely controls a purchase order and a receipt as well as the invoice. Non-PO invoices carry more risk and warrant tighter approval routing. Matching alone cannot stop bank-detail fraud, however, which is why it must be paired with payment verification.

Build a culture of verification

Controls fail when people feel pressure to skip them. Urgency is a fraudster's main tool: an executive who needs a payment today, a supplier threatening to stop deliveries. AP teams should know they will be supported, not criticized, for pausing an unusual request to verify it. Regular training using real examples, and a simple route for reporting suspicious requests, turn the AP team into a detection layer in their own right. The ACFE finds that nearly half of occupational fraud cases are detected through tips, which shows how much depends on people noticing and speaking up.

Where automation and AI help

Manual processes are poorly suited to fraud prevention. A reviewer who handles hundreds of invoices a day cannot reliably spot a near-duplicate from three months ago or notice that a supplier's bank details changed last week. Software can do both on every invoice, every time.

Automation helps first by removing the conditions fraud depends on. When invoices are captured into one controlled workflow, there is no side channel of emailed PDFs approved outside the system. Rules for matching, tolerances and approvals are applied consistently, and every action is recorded in an audit trail that internal and external auditors can follow. Suppliers who submit invoices and check payment status through a portal also have less reason to email AP directly, which narrows the channel that impersonation attacks rely on.

AI adds a second layer: pattern recognition across the whole invoice history. Machine learning models can flag invoices that are similar but not identical to earlier ones, amounts that break a supplier's usual pattern, unusual timing or new combinations of supplier and approver. Those are the signals that rule-based checks and tired reviewers miss. Adoption still lags the threat: only 17% of organizations in the AFP survey use AI to combat payments fraud.

The best results come when AI flags risk and people decide. A system that blocks payments without explanation erodes trust, and one that explains why an invoice was flagged lets AP staff resolve genuine cases quickly and escalate the rest.

This is the approach behind Tungsten InvoiceAgility+. It captures invoices from every channel, including paper, email, portals, EDI and Peppol, into a single validated workflow, with two- and three-way PO matching, configurable business rules and auditable AI decisions. Specialized AI agents check invoices for suspicious patterns, document inconsistencies and duplicates before payment, and low-confidence or flagged cases route to people for review. Through the Supplier+ portal, suppliers can submit invoices, track invoice and payment status, and resolve issues in one place.

A governance discipline, not a project

Invoice fraud adapts. As one route closes, attackers move to the next, and the arrival of AI-generated documents and voices means verification methods that worked last year may not work next year. Organizations that keep losses low treat fraud prevention as a standing part of AP governance. They review supplier master changes monthly, test segregation of duties at every reorganization, track flagged and confirmed cases as a metric, and revisit controls whenever the process or the technology changes.

The principle underneath is simple. No single control will stop every fraud, but a process in which a fraudulent request has to pass verified supplier data, enforced approvals, document matching and intelligent anomaly detection, all at once, makes fraud expensive, slow and visible. For most attackers, that is enough to move on.