Finance departments are automating more of their work while simultaneously confronting a less obvious risk: employees becoming more tolerant of small financial misconduct. A new Medius survey of 2,386 finance executives across the U.S., U.K., Sweden and France suggests that accounts payable teams are caught between rising invoice fraud, increasingly sophisticated AI scams, persistent manual processes and growing adoption of agentic AI.

The biggest threat to a finance department may not always arrive as a sophisticated deepfake or a carefully engineered invoice.

Sometimes, it is a small expense claim that nobody questions.

According to the 2026 Financial Census from Medius, 87% of surveyed finance executives said they had ignored a small expense, reimbursement or claim they believed was fraudulent. Meanwhile, 67% said they would be likely to make a minor dishonest expense claim themselves if that behavior were common among colleagues.

Medius describes this low-value rule bending as “shallowfake” fraud—a category covering seemingly insignificant financial misconduct that can accumulate into material losses.

The terminology is new, but the underlying problem is familiar. Finance controls tend to receive the most attention when organizations are worried about large payments, invoice manipulation or executive impersonation. Small expenses can fall into the gaps between policy, automation and human oversight.

The Medius findings suggest that this cultural dimension of fraud deserves more attention.

Nearly three-quarters of respondents said minor forms of workplace fraud were already common. Another 64% said they would feel justified in committing a small dishonest financial act if they believed they were underpaid or undervalued, while 57% said they would round up an expense or mileage claim if they thought it would go unnoticed.

That creates a difficult environment for finance leaders. Automation can identify anomalies, but technology cannot by itself determine whether employees believe financial controls are legitimate or whether small violations have become socially acceptable.

At the other end of the spectrum, AI is making conventional fraud more difficult to detect.

Ninety-three percent of respondents said they were concerned about AI-generated fraud over the next year. Medius also reported that U.S. organizations experienced average annual invoice-fraud losses of $168,000 in the survey.

The company’s broader Financial Census research highlights how invoice fraud and deepfake attacks are already affecting finance teams. Its published research has found that 53% of finance professionals had been targeted by attempted deepfake scams and 43% said they had ultimately fallen victim to one.

The contrast is striking: finance leaders are preparing for increasingly sophisticated AI-assisted attacks while some organizations still struggle to control the everyday exceptions generated by their own employees.

Automation has not eliminated late payments

The second major finding concerns a problem automation was supposed to solve.

Eighty-five percent of finance teams surveyed said they have some level of accounts-payable automation. Yet nearly half of invoices still require manual intervention, according to Medius.

Respondents estimated that 28% of invoices are paid late in a typical month. The consequences extend beyond internal efficiency. Forty-five percent said suppliers had imposed stricter upfront payment terms because of late payments, while 43% reported suppliers ending relationships altogether.

Other respondents cited reduced supplier service quality or speed and escalation of disputes into legal action.

The issue highlights a distinction enterprise finance technology buyers increasingly need to make: having automation is not the same as having an automated process.

A company may have optical character recognition, invoice-routing software or automated approvals while still relying heavily on employees to resolve exceptions, chase approvals and validate supplier information.

Medius’ own published research similarly identifies manual intervention as a persistent AP problem, reporting that finance teams spend substantial time responding to invoice inquiries and that approval delays can contribute to supplier-payment problems.

IDC’s research reinforces the broader enterprise trend. The analyst firm says organizations are scaling agentic AI but still face challenges determining which use cases to prioritize, how to measure return on investment and what actually works in production.

That makes measurable outcomes increasingly important for finance-technology deployments.

Finance is adopting agentic AI—but keeping humans in the loop

Agentic AI is emerging as the next stage of finance automation.

Medius reports that 38% of surveyed finance executives already have agentic AI operating in at least some finance processes, while another 50% expect to deploy it within 12 months.

Yet adoption is accompanied by a significant governance constraint: 90% said there is always a financial or compliance threshold requiring human approval regardless of an AI system’s accuracy or track record.

That is a useful signal for enterprise technology leaders.

Agentic AI in finance refers to software systems capable of executing multi-step tasks or making recommendations with limited human intervention, rather than simply generating information in response to a prompt.

In accounts payable, potential applications include invoice classification, exception handling, supplier verification, fraud detection and payment recommendations. But financial workflows operate under regulatory, accounting and audit requirements that make unrestricted autonomy difficult.

The question therefore becomes one of controlled autonomy.

Medius says respondents were almost evenly divided over who should be accountable when an AI decision causes a financial loss or compliance problem. At the same time, 45% of finance leaders said their teams often act on AI-generated recommendations without human intervention.

That gap between operational reliance and formal accountability could become one of the defining governance challenges for autonomous finance.

IDC’s 2026 AI maturity research points to a similar problem across enterprises: only 3.1% of organizations in its global benchmark had reached its highest “optimized” AI maturity stage, while just 12.8% had reached either of the two most advanced stages.

For CFOs, that suggests AI adoption should not be measured solely by the number of automated processes. Governance, ownership, auditability and measurable financial outcomes need to become part of the deployment equation.

AI is changing finance careers, too

The technology shift is also changing the finance workforce.

Medius found that 55% of finance leaders now consider AI fluency a meaningful differentiator when hiring. Even more—86%—said employees’ use of AI is now considered in performance evaluations.

That creates a new contradiction.

AI is supposed to reduce repetitive work, but 75% of respondents said AI usage has increased worker fatigue or burnout.

The result is a warning against treating AI adoption as an automatic productivity gain. Employees may be expected to use new tools while simultaneously maintaining existing processes, validating machine-generated recommendations and learning new technical skills.

Finance teams are consequently moving toward a hybrid operating model in which people remain accountable for high-impact decisions while AI handles more of the underlying processing.

For accounts payable and finance leaders, the technology opportunity is significant. Automated workflows can potentially reduce manual work, improve fraud detection and accelerate payments. But the Medius findings suggest the harder problem is organizational: determining where automation should act, where humans must intervene and who ultimately owns the outcome.

The future of autonomous finance, in other words, may depend less on whether AI can make a decision than whether a company can trust—and govern—the decision after it is made.

Market Landscape

Finance technology is moving from rules-based automation toward AI-assisted and agentic workflows. The shift is occurring across accounts payable, procurement, expense management, fraud prevention and financial planning.

Medius is competing in a market that includes enterprise platforms from SAP, Oracle, Coupa, Microsoft and specialized AP automation providers. The competitive advantage is increasingly moving beyond invoice capture toward end-to-end workflow orchestration, fraud detection, supplier intelligence and AI-based decision support.

The market’s central challenge is execution. IDC’s 2026 research says 56.1% of organizations have agentic-AI initiatives underway, but enterprises continue to struggle with prioritization and measurable outcomes.

For CFOs and CIOs, that means the next generation of finance automation will be judged on business results: fewer fraudulent payments, faster invoice cycles, stronger supplier relationships, lower exception rates and clearer accountability.

Top Insights

• Small fraud is becoming normalized: Finance leaders report widespread tolerance for minor expense manipulation, creating cumulative losses that traditional high-value fraud controls may overlook.
• AI raises both risk and opportunity: Finance teams are adopting agentic AI while simultaneously preparing for AI-generated scams, creating new requirements for governance and human oversight.
• Automation has an execution gap: Despite widespread AP automation, manual invoice intervention and late payments remain common, damaging supplier relationships and increasing finance-team stress.
• AI accountability remains unresolved: Organizations increasingly act on AI recommendations, yet responsibility for financial or compliance failures remains divided across finance, IT and employees.
• Finance jobs are changing: AI fluency increasingly affects hiring and performance evaluations, while heavier technology demands are also contributing to reported fatigue and burnout.

Get in touch with our fintech expert