The future of insurance fraud investigations: AI, intelligence and the importance of human judgment

Sedgwick is a proud supporter of the ACFE’s International Fraud Awareness Week, taking place 15-21 November – a global effort to minimize the impact of fraud by promoting anti-fraud awareness and education.

Insurance fraud is becoming increasingly sophisticated, with AI enabling new forms of deception at greater speed and scale, while criminal activity continues to cross international borders. For those of us working to detect and investigate fraud, these developments raise important questions about how we strengthen our capabilities, share intelligence and protect customers.

These challenges were front and centre at the eighth annual Global Insurance Fraud Summit (GIFS) in Panama, which brought together anti-fraud specialists, regulators and law enforcement from around the world.

This was my third GIFS, and one thing that struck me was how much the power of this global collective is maturing. There is growing recognition that tackling insurance fraud effectively requires stronger partnerships, better intelligence and a willingness to explore emerging technology responsibly.

Three themes stood out to me, with one debate in particular raising some fascinating questions about the future of fraud investigations.

Three trends shaping the fight against insurance fraud

1. AI: Insurance fraud, whether committed by organised criminal groups or presented by individual consumers, is increasingly AI enabled, allowing fraud to be carried out more quickly and at greater scale. I’m in no doubt that AI is enhancing the capabilities of would-be fraudsters, enabling more people to commit increasingly sophisticated fraud. 

But our mindset has to be one that seeks to turn this negative into a positive by not treating AI as our arch enemy – the same technology can strengthen our spotting of unusual claims, identify patterns and support our disruption activity. The challenge is to ensure that our capability evolves as quickly as the threat in a safe, proportionate way. 

2. Cross-border: Whilst many of the threats we face are similar, each jurisdiction brings its own legislation / regulation, emerging risks and culture. Bringing these perspectives together helps us better understand the challenges, identify common priorities and coordinate a more effective global response. 

With more than 70% of fraud in the UK alone either originating from outside the UK or having an international element, creating new partnerships and pinpointing emerging cross-border challenges never felt more important. Stronger connections between our industry, law enforcement, governments, and the wider public and private sectors will make a difference, fostering a truly global response. 

3. Intelligence: It’s imperative that we have valuable intelligence – and we do. There are also mechanisms for distributing intelligence across organisations and within individual jurisdictions – but all too often there aren’t the cross-border connections to reveal the bigger picture. We must find better ways to safely and compliantly share and connect data across borders, turning isolated insights into actionable intelligence that can support a more coordinated response to fraud. 

The hottest topic in town: could Agentic AI investigate fraud?

Head and shoulders the hottest topic at the summit was, unsurprisingly, AI – and one debate in particular stood out: to what extent might we use agentic AI to investigate suspect claims, rather than using AI to spot suspect claims? 

Our industry is seeing change, at some pace, so for me it has to be a topic worth exploring because it moves the debate from using AI as an assistant to using agentic AI as an investigator. My view in a nutshell? I can’t see agentic AI – ‘an agent’ – replacing the role of an investigator; but I can certainly see an agent has the potential to perform the work of an exceptionally diligent investigative analyst. 

Moving from prediction to judgement

Spotting and flagging a suspicious claim is one thing. Deciding what happens next, what evidence to gather, what lines of enquiry to pursue, and when to escalate requires context and proportionality. Could an agent do that? For me, the real question isn’t whether AI can identify risk, because we know it can. It’s whether an agent can exercise investigative judgement fairly, consistently and transparently. 

Imagine AI identifying a suspect auto claim for a road traffic accident. I can see a day when the next step is for an agent to review policy history, examine claim documents, look for network links and supplier connections, and consider telematics data. I can also see it building an evidence pack – and it might even test both suspect and innocent explanations and even recommend escalation if it senses a case is high risk and warrants additional scrutiny.  But that’s where it ends for me – for the moment anyway. Then you bring in the (human) investigator to review the findings, challenge the reasoning by testing the agent and, fundamentally, remain accountable for the decision to investigate – and ultimately the final case outcome. A compelling model? 

Governance

A robust governance framework would be essential to ensure agentic investigations are safe, compliant and maintain public trust. For me, governance comes down to five things: explainability, accountability, proportionality, fairness and oversight. Every action should be auditable, outcomes should be tested for bias, and the higher the potential impact on the customer, the greater the level of human oversight. Trust comes from control, not technology.  

Scalable investigations

Let’s assume we deliver on the governance – could we make agentic investigations truly scalable, game-changing even? Such a breakthrough wouldn’t come from better models alone – it’d come from connected data, strong investigative workflows and seamless human-AI collaboration. But there’s real value to be had if an agent can gather evidence, test hypotheses, create an auditable case file and present it to an investigator in minutes, because that can be scaled up – in turn, investigators spend less time searching and more time applying judgement. 

The bottom line

Insurance fraud investigations aren’t like product recommendations or marketing campaigns. Decisions can have profound consequences, and getting fraud wrong isn’t a simple technical error. It’s what I sometimes refer to as a customer harm event.

That’s why I believe we must be careful not to scale autonomy faster than we can scale governance. Trust is earned by demonstrating fairness, transparency and accountability as capability grows. 

Get the governance right and we’re in the ballgame. AI can identify suspicion, and agentic AI has the potential to triage cases and build evidence packs that allow investigators more time to investigate. But that’s where it should stop. It’s for humans to determine dishonesty. 

Agentic investigations: yes or no? 

I’m enthusiastic about agentic AI’s ability to transform fraud investigation. The volume and complexity of modern insurance fraud mean investigators need better tools. Used responsibly, agentic AI can make investigators better. Used carelessly, it can make mistakes faster and at scale. Our challenge is not deciding whether agentic AI has a role but deciding how we introduce it in a way that protects customers while improving outcomes. 

The organisations that succeed won’t necessarily have the smartest algorithms. They’ll have the best data, the strongest governance, and the clearest partnership between human expertise and machine capability.

So it’s a cautious ‘yes’.

What this means for the future of fraud prevention

For me, the conversations at GIFS reinforced just how important collaboration, intelligence and innovation will be in tackling the next generation of insurance fraud.

The threats we face are increasingly sophisticated and international in nature, and our response has to keep evolving. That means making better use of the intelligence available to us, strengthening connections across borders and exploring how emerging technologies can support the people responsible for investigating fraud.

At Sedgwick, these are priorities that shape how we approach fraud prevention and investigation. We recognise the value of advanced technology and predictive analytics in identifying suspicious activity, alongside the investigative expertise, sound judgment and appropriate oversight needed to act on those insights. We’re equally committed to working with industry partners to share knowledge, understand emerging threats and strengthen our collective response.

There is also an important balance to maintain. As the industry works to make claims experiences simpler and faster, effective fraud controls must remain part of that process. Protecting genuine customers, identifying dishonesty and maintaining trust all matter.

Insurance fraud has real consequences for insurers, businesses and the people they serve. It drives up costs, disrupts operations and can damage reputations. That’s why I believe our responsibility is to keep challenging how we work, remain open to new approaches and make sure that every advancement ultimately strengthens our ability to protect customers.

If there’s one thing I’ve taken from my third Global Insurance Fraud Summit, it’s that we’re better equipped to meet these challenges when we tackle them together. At Sedgwick, we’ll continue to bring our expertise, technology and partnerships to that effort, helping protect our clients, their customers and the integrity of the insurance industry.