Fraud-proof by design: AI's role in fighting economic crime

Growing up in the village, AI was widely understood to mean artificial insemination and was naturally associated with local veterinary officers.

Those days are gone: today, even in remote regions, AI means artificial intelligence. Some argue that AI is overhyped and more bark than bite.

Yet in the financial sector, AI has many practical applications, with fraud risk management leading the development of AI use cases.

It is unsurprising that fraud risk management leads in AI use cases. Long before AI agents, agentic AI, or generative AI (gen AI), financial crime fighters used big data and predictive analytics to detect and stop crime.

These capabilities grew from rule-based transaction monitoring systems used since digital payments became widespread.

Today, gen AI and AI agents improve transaction monitoring, reduce false positives, investigate suspicious activity, and draft reports.

They already have a significant impact in institutions with the right data and AI governance. In these applications, AI makes traditional fraud detection and investigation more effective and efficient.

But can we use the AI revolution to transform the fight against financial crime beyond these traditional measures?

Can AI deal a decisive blow to economic crime more broadly? Economic crime includes corruption, tax evasion, exploitation of vulnerable groups, and insider trading, alongside financial crimes such as fraud, money laundering, and bribery.

To answer this question, we must consider why people commit economic crime. The simplest and most widely cited explanation is the fraud triangle: people offend when they face pressure or motivation, can rationalize the intended act, and have the opportunity to commit it.

The fraud diamond adds capability; the fraud pentagon adds arrogance. My favourite, rational choice theory, holds that rational offenders commit crime when the expected benefit exceeds the expected consequences.

Agency theory is also important, particularly for corruption. It holds that economic crime occurs when agents, such as employees or government officials, pursue their interests at the expense of principals, such as shareholders or citizens. Other theories emphasize the roles of strain, learned social norms, denial, and greed.

AI’s computing power, influence, reach, and imminent disruption of modern life offer an opportunity to design an economic system that prevents crime rather than merely improving its detection and response.

Consider opportunity and the agency problem. AI can reduce the control and discretion that give human agents opportunities to offend.

Automation began addressing this problem before AI by removing unnecessary human intervention. For example, automatically queuing supplier invoices for payment removes accounts payable staff’s discretion and their opportunity to solicit bribes from suppliers. As AI reshapes processes, designers should deliberately remove opportunities for abuse.

On rationalisation and rational choice, AI can reduce crime’s allure while increasing its consequences. Current applications already improve detection rates, investigation speed, and evidence quality. Wider adoption can shift the balance from “crime pays” to “crime does not pay".

To deliver consequences faster and more consistently, however, internal administrative processes, including disciplinary hearings, as well as litigation, prosecution, judicial processes, and asset recovery measures must be modernized to support swift and more certain justice. There is scope to use AI more extensively in these areas.

AI also has applications in education, where it can influence social norms. I cannot, however, contemplate a use case that would reduce greed or arrogance—but chances are that AI can recommend one!

Even so, as we continue to adopt AI, we must avoid inadvertently increasing economic crime by, for instance, intensifying pressure and strain.

Workers who fear losing their jobs to AI may, for example, feel driven to secure their futures through criminal activity.

Proper AI governance is also important to ensure that, as we adopt AI, its outputs remain explainable and auditable, with an accountable human being ultimately responsible.

Overall, AI’s role in fighting economic crime should extend beyond detection and investigation. People and units responsible for fraud risk management should be involved when systems and processes are redesigned.

Their expertise, and the need to manage economic crime sustainably, should be embedded in products, processes, and systems to achieve fraud-proof outcomes.

Fraud risk management—and the fight against economic crime more generally—needs to shift from being considered the preserve of specialist units in risk, compliance, forensics, or law enforcement. It needs to become a strategic business imperative and part of a wider societal conversation. It needs to move beyond internal controls to encompass culture and broader operating models. Only then will the full potential of AI as a crime fighter be achieved.

- The author is a Partner and the Forensics Services Lead at PwC Kenya