Deploying artificial intelligence in a counter-fraud context sounds straightforward in theory. In practice, the barriers that prevent effective use are rarely technical. That was the clear message from Akber Datoo, founder and CEO of D2 Legal Technologies, during a recent webinar on AI and fraud in the public sector.
Akber, who works extensively on technology procurement and responsible AI adoption, said the barriers to effective deployment are primarily about governance, assurance, and trust - not about the capability of the technology itself. He set out six foundations that organisations need to get right.
1. Use Case Clarity
Akber described vague ambitions as a significant problem. Saying "I want to use AI to reduce fraud" is, in his words, far too vague. A proper use case, he suggested, looks more like: "We want to use AI to triage payment anomalies for human review." The more specific the use case, the more possible it becomes to design appropriate controls around it.
2. Data Quality
Akber described data quality as a huge issue, and one that AI will not fix on its own. If the data going in is incomplete, biased, or messy, he warned that AI will not simply clean it up - it will industrialise the problem. "Rubbish in isn't just rubbish out," he said. "It's exponentially increasing the amount of rubbish we're then going to have to deal with." Jennifer Evans from the Public Sector Fraud Authority echoed this point, noting that having organised, cross-departmental data available - with the right governance frameworks for sharing it responsibly - is the very first foundation, and one that has little to do with AI itself.
3. Data Protection and Confidentiality
In counter fraud work, the data involved will often include personal information, case material, operational intelligence, and information about vulnerable people. Akber said the position on public generative AI tools should be clear: they are not the right place for sensitive data. He noted that government guidance to civil servants is explicit on this point. However, he also cautioned that telling people not to use public tools is only half of the answer. Organisations also need to provide a proper alternative - a way for people to use these capabilities appropriately within a secure environment.
4. Challengeability
Public bodies need to be able to clearly explain why a person, supplier, transaction, or claim has been flagged. Akber was direct on this point: saying "the model said so" is not going to be sufficient. Any AI-assisted decision that affects an individual needs to be explainable in terms that can be understood and challenged.
5. Meaningful Human Accountability
Akber described the human-in-the-loop requirement as one that cannot be ceremonial. A human approver who simply agrees with whatever the system produces is not providing meaningful oversight. He highlighted the risk of what he called automation bias - the tendency to go along with a machine output because it feels authoritative and well-structured, and to quieten the internal voice that might otherwise raise a concern. The human approver, he said, needs to be trained, empowered, and genuinely able to disagree with the machine.
6. Procurement Discipline
Akber argued that too many organisations procure an AI model and consider the job done. What they should actually be procuring, he said, is a governed workflow - one that includes data controls, audit logs, testing, monitoring, escalation procedures, redress mechanisms, supplier obligations, and exit rights. The model itself is only one part of what needs to be in place.
The AI Literacy Problem
Alongside these six foundations, Akber identified AI literacy as a cross-cutting issue. The need for literacy does not stop with data scientists and vendors. It extends to investigators, lawyers, procurement teams, senior responsible owners, and the public-facing staff who will need to explain these systems to the people they serve.
And it does not stop there either. Akber noted that in order to secure public trust in the use of AI, the general public also needs to be able to understand, at least at a broad level, why these tools are being used and what safeguards are in place.
Jennifer Evans added that bringing everyone along is essential. "It doesn't work if you have a couple of people who are pushing out a product and actually the entire department doesn't really know what it's used for," she said. You need full buy-in across the board.
Founder and CEO of D2 Legal Technologies and Professor of Law, Technology and AI at the University of Surrey
Jennifer Evans, Project Manager at the Public Sector Fraud Authority (PSFA)
This post is based on a webinar on "AI and fraud prevention in the public sector", featuring speakers from the Public Sector Fraud Authority, the Hertfordshire Shared Anti-Fraud Service, NUIX, and D2 Legal Technologies. Listen to the whole thing for free here >> https://register.govnet.co.uk/webinar-ai-vs-fraud
Jessica Kimbell, GovNet
