Fighting Fraud in Typhoon Season

Natural disasters create an environment for scam artists to exploit desperate communities. AI can help insurers detect and counteract these nefarious schemes before they cause societal harm.

Every year, natural disasters devastate countries around the world, causing untold destruction. In the wake of these disasters, as people pick up the pieces, assess the damage, and file insurance claims, unscrupulous scam artists seek to exploit the situation. These bad actors exploit desperate people by ensnaring them in elaborate schemes with extortionate fees when they are most vulnerable. 

Counting the Cost of Tropical Storms

Although natural disasters occur worldwide, some regions are particularly susceptible due to their geographical location, and these areas pay a heavy toll each year. Typhoons and other tropical weather events are some of the most consistently devastating natural disasters. In 2019, Typhoons Hagibis, Faxai, and Lekima caused more than $34 billion in combined damages in Asia. In the US, hurricanes and tropical storms caused economic losses of nearly $500 billion in 2024 alone, with Hurricane Helene causing an estimated $20 billion in property damage and economic disruption, with total losses projected to surpass $100 billion. 

Since 1980, tropical cyclones, including typhoons, have caused over $1.3 trillion in damage, with an average cost of $22.8 billion per event. The extraordinary damage to people’s properties and belongings means that every disaster leads to enormous volumes of insurance claims. One of the places on Earth that experiences this pattern most acutely is the Asia-Pacific (APAC) region, in particular, Japan.

Steve Roberts, Mind Foundry’s Co-Founder and Chief Science Officer, notes that “Because of its geographic location, Japan is no stranger to natural disasters. Earthquakes, landslides, typhoons, and other extreme weather events are yearly recurrences that are only worsening with climate change. These natural disasters are always a source of non-life insurance payouts, many of which can be huge.” For context, in both 2018 and 2019, non-life insurance companies in Japan paid out over ¥1 trillion. When the sums in play are this big, it’s no surprise that criminal-minded individuals seek opportunities to get in on the action.

How Scam Artists Defraud Victims and Insurers

In recent years, scam artist activity in Japan has been on an alarming upward trend. According to the National Consumer Affairs Center of Japan, the number of complaints about scam artists surged from 111 in 2010 to 5,359 in 2020, representing an increase of 4,728%. The way these scam artists operate is cunning, and the consequences for disaster victims can be severe.

Watch the full video here.

Scam artists position themselves as “Insurance claims support companies” and try to mediate between the policyholder and the insurance company to estimate any property damage and arrange repairs. They also support the policyholder in the filing of their insurance claim, as it’s illegal for a “proxy” to file a claim on behalf of someone else. However, scam artists target vulnerable individuals by posing as these companies, and this activity is especially prevalent in Japan during and after typhoon season.

A common scenario is that an individual will be targeted, either on social media, by phone, or in person at home. They will offer to repair any property damage and pay for any repairs using the money from an insurance payout and to support the policyholder in filing their claim. Sometimes, they will lure people into scams by suggesting, “Why not pay for your home repairs using your insurance?”. These scam artists will often use aggressive tactics to try and pressure desperate or unwitting individuals to sign contracts that charge exorbitant and often hidden fees. Sometimes, they will encourage the victim to file false insurance claims with exaggerated damage reports with the promise of a larger payout, and in some cases, they have been known to damage properties themselves to dupe insurance companies. 

If the insurance company rejects the claim, or if the claimant attempts to cancel the contract, the scam artists can charge a cancellation fee that ultimately leaves the claimant significantly out of pocket. Furthermore, these scam artists often undertake shoddy repair work at great cost to the victim and react aggressively to any attempts to complain or renege on the contract. Finally, if the claimant unknowingly files a false insurance claim at the behest of these scam artists and the insurance company finds out, then they can be liable for legal action for having undertaken criminal activity. 

Because these scam artists usually encourage policyholders to file claims themselves, it can be incredibly difficult for insurance companies to determine which ones are fraudulent. As noted by Kiyoshi Hashimoto, Manager of Aioi Nissay Dowa Insurance (ANDI) Claims Administration Division, “Malicious vendors exploiting natural disasters raises the risk that customers might unwittingly become complicit in this fraud, making it a critical issue from both customer protection and fraud prevention perspectives”. This is where the Aioi R&D Lab - Oxford’s fraud detection solution is having an impact.

Fighting Back Against the Scammers

The Aioi R&D Lab - Oxford has developed a solution to detect malicious vendors early, eliminate fraudulent claims, and protect policyholders in cases previously unnoticed by humans by utilising an advanced AI model. This model was the product of extensive collaboration between Mind Foundry and ANDI in the Lab. It began with initial workshops and discussions with ANDI to establish the requirements for the solution, such as data availability and speed of delivery. This included incorporating feedback from the claims handlers themselves, as it was critical for the solution to complement their work effectively whilst harnessing their domain expertise. 

Ensuring that the system design was consistently aligned with the end user's needs increased the chances of a performant solution once deployed. This collaboration is a fundamental aspect of the operation of the Aioi R&D Lab - Oxford. In this instance, we were able to understand the true nature of the problem and thus ensure that the most valuable features needed to get to the heart of the problem were the ones that made it into the solution.

To train the model to detect fraud effectively, Dr Owen Parsons, a Lead Machine Learning Engineer at Mind Foundry, highlights that “We do this in a supervised way, so you have a lot of historical data that the model can learn from. We take data points that are fraudulent claims and data points that are non-fraudulent claims, and we train the model to detect the difference between them.” This way, the model can differentiate between a new claim that is suspected to be fraudulent and one that isn’t. Furthermore, the model could be seamlessly integrated into the claims department to add value and positively impact their work immediately.

Fraud, however, is not a static phenomenon, and fraudsters are constantly looking for new ways to take advantage of customers and insurers. This is why, as Dr Parsons highlights, “Continuous learning is really important for a model like this. The ways that people try to commit fraud will evolve over time, and the model needs to be able to keep up with this and learn the new patterns over time.” This was achieved by incorporating a feedback loop into the solution so that, as new types of fraud arise, a human can label them, and the model learns from these labels, improves its detection capabilities, and captures more fraud from that point on. Ultimately, this means that this model will continuously stay on track as the market shifts, and at the same time, it can help claims handlers about these new fraud types to make them more effective at their job.

Watch the full video here.

The Power of Human · AI Collaboration

One of the most impressive aspects of this collaboration was the speed at which a deployed model was achieved. The timeframe for this project was critical, and the team needed to have an impact straight away. From the initial conception of this project until the first prototype, was just four months. Then it took only another two months for a final model to be ready. 

Human-AI collaboration is at the heart of the solution’s success. It enabled human fraud expertise to be directly encoded into the AI model to maximise its effectiveness. As Kiyoshi Hashimoto goes on to say, “With the help of Mind Foundry, we were able to quickly develop a high-precision AI model. By combining AI's multidimensional analysis capabilities with human intelligence, we can now quickly detect malicious vendors and protect policyholders.”

The ever-changing nature of fraud and the adaptability of fraudsters and their techniques mean that combatting home insurance fraud is an ongoing battle. Only by using the latest in AI technology and combining it effectively with human intuition and domain expertise can we protect our society, businesses, and individuals from this pervasive threat. This is just one of many projects being undertaken by the Aioi R&D Lab - Oxford, where the power of Mind Foundry’s AI is being harnessed in partnership with ANDI’s insurance expertise to enact lasting and positive change in the world.

You can watch the full video on preventing home repair fraud with AI here 👇