Improving Fraudulent Claims Detection With AI

Sompo Holdings (Asia) was the recipient of the AI Award for general insurance during Singapore Business Review’s Technology Excellence Awards last year. A prime mover of the initiatives leading up to the award is Dr. Patrick Chin, the Head of Global Digital, Branding and AI Practice, who led the team working with AI Singapore in the 100E Programme to implement an AI-driven fraudulent claims detection system. I had a good chat with him on how the project came to be and the future plans for Sompo on the horizon.

The conversation was recorded in May, 2021.

Below is a transcript of the conversation [*].



Hi, Patrick. Nice to have you with us.

Hello, everyone. Nice to meet you.

Patrick, three years ago, in mid-2018, you joined Sompo with big plans as Head of Digital for Asia. The following year, 2019, AI Singapore became part of that journey with the execution of a 100E project. First off, tell us more about Sompo and the services it provides to customers.

Sure. So, Sompo originated from Japan. We have more than a hundred years of experience worldwide in the general insurance space. Over at Japan, we cover not only insurance. We do even nursing homes down to their health and wellness. Outside of Japan we cover all the way to Turkey and Brazil, focusing on providing insurance, reinsurance and even underwriting services to our partners and stakeholders. So, over at Asia, we focus mainly on retail for general insurance.

Bringing the focus onto technology, we know the insurance industry has been around for a very long time. I would say that it is one of those industries that have always been data-driven. How do you see the evolution of the industry along with progress in digital and data technology?

As you have rightly pointed out, insurance is actually a traditional industry, that focuses a lot on manual processes. Over at general insurance space, if you think about auto insurance like motor insurance, when someone buys insurance, whether they buy from agencies or from other platforms, the processing at the backend is actually quite manual, which means when someone submits a request, it will be sent to the backend where there will be an underwriter looking at it, there will be an administrator looking at it before it is sent back to customers as covered. So, that process is actually quite manual and we have seen a huge change in terms of how a lot of insurance companies have adopted digital from end-to-end, whether it is from the buying process to the claims process. So, I think that part is coming quite strongly, especially now with the current COVID situation, more and more customers will expect all deliveries to be in digital, at least along the digital touch points and especially for the backend processing, like claims or even policy servicing. They will expect 24-by-7. They will expect as and when they need someone else to help them, to be able to service them. So, I foresee that change coming and also digital will play a very important role. In fact, it is already playing, but there are more disruptions coming in terms of how we propel ourselves faster into that space.

So, technology can really play a role in every part, from end to end. There’s one particular area every insurer has to grapple with and that is fraud. In fact, that was the subject of collaboration with AI Singapore. What is the impact of fraud on the insurance business? What does the process of fraud detection typically involve?

In a typical insurance company, within the claims department, they will have a department called SIU – Special Investigation Unit. Not every insurance company has it, but for companies who have it, they invest maybe a headcount of one-two people whose main role is to spend time identifying the potential fraud cases that can be highlighted to the management or within the processing. Normally, they do it by sampling, around four to seven in a week or two weeks. How it works is, it depends a lot on personal experience and judgment to determine whether this is a potential fraud and there’re no written rules. I call it tacit knowledge because it depends on the person who is working on it and how well he communicates within the team to follow a certain process that they have to pick up the fraud. I think this is traditionally how it is done in a lot of companies, so that’s why I do agree with you, getting the fraud detection mechanism in place is so important, especially within an insurance company, specifically for the claims department.

And that actually brings me to my next point. Among the many services that Sompo provides, travel and personal accident were picked for the 100E collaboration with AI Singapore. Was it a very easy decision to make to undertake this venture?

For sure. So, what happened is, within Sompo we do have different business lines. We have motor as our biggest portfolio, then we have travel, PA, home or even commercial insurance. The reason why we picked travel and PA is because of the maturity of how they were already handled online. Before COVID, we knew that people liked travel and most of the purchases or even the processing were through online. That means you can buy from Expedia or even from SIA. All you need to do is a couple of clicks of the button and all this will be parked in place. It’s quite important for us to select the right business or the business line that’s ready to try out for digital because for motor, it is not so developed yet, which means a lot of the business is still going through the dealer or the bank or the agency, so that’s why we picked these two – travel and PA for a minimum viable product test to see whether it works because they are more ready for digital trials, especially for AI fraud.

And, yes, AI Singapore worked about seven months on this project and delivered a minimum viable product at the end of it. The team here who worked on the project had very good things to say about the technology lead in Sompo, who had a very clear understanding of the business case, as well as the technical team in general which provided great database support.

Thank you very much for that! I think it’s actually both ways. Besides your team having good experience working with us, I heard feedback from my team members that they had a very good working relationship with AISG as well. So, I think it’s both ways and we learned a lot as well.

That’s great to hear. So, how was the outcome of the project? What impact were you able to achieve among the stakeholders in tangible terms?

When we did a POC study with AISG building that particular model. During the process itself, we picked, as you will know, PA and travel with close to three thousand over cases to try to build a model, the model itself was performing quite well. I think we have achieved more than 92% prediction accuracy which is good. Then after that was done, we presented to our stakeholder who belongs to Sompo Singapore head of claims and the department, in the end we put into production. So to date, the benefits that we have seen, especially for Sompo Singapore is as follows. First, definitely the customer experience has improved tremendously because AI fraud itself is not just to detect fraud, but the key thing is the manual process of checking actually slows down the whole process. By having something like the AI fraud (detection) itself, we can screen through 100% of all the cases, like thirty thousand cases that we had before COVID. When it goes through that way, customers who submit claims will also experience a faster turnaround time, because we don’t have to spend time trying to investigate everything using manual way – that is one plus point. The second benefit is also cost savings that the department has seen. The predicted cost savings that they foresee by the third year is about two hundred and close to three hundred thousand dollars – that’s only cost saving for having to go through scanning each of the policies. The third one which we have yet to establish is quite difficult to measure because these are the potential fraud that may come in which we can pick up by the system. So, I think these are the three impacts that we see within Sompo Singapore. The last one, which is more critical for me because I run Digital now globally, is the potential knowledge that I can bring not only for Sompo Singapore, but to other countries like Sompo Indonesia or even Sompo Vietnam, that we can learn within the Sompo group how we can scale this across the Sompo group. Yes, these are the four benefits I would say.

Yes, in fact the AI-powered fraudulent claims detection system there was deployed won the AI award for general insurance in Singapore Business Review’s Technology Excellence Awards sometime in the middle of last year so I must congratulate you guys on that.

Thank you very much for that. Thank you AISG for being a partner as well.

So, moving forward, what’s in the pipeline for Sompo?

We have an AI roadmap that we have drawn up for the whole Sompo group. For the next two, three years definitely we are not just doing AI fraud, we have launched an AI chatbot in the Philippines, we’re also implementing other AI capabilities within our processes whether it’s claims automation or backend or even some telematics. So as regard to that, we do have an AI roadmap that we want to implement for the next three years. Specifically for AI fraudulent claims detection we have finished and implemented the rollout in Sompo Singapore. In fact, just this month (May 2021) we had also finished the same model in Sompo Indonesia with good results as well. The predictive accuracy is close to 85 to 90 percent and that one is already being done, also for travel and PA. With the two countries done, it is actually a very strong testament to the team that the countries are buying and adopting it and they see some value in this AI fraudulent detection. The next step for us is to move to another line of business. The next line of business we are targeting is our motor line of business, which is the biggest line of business for us across the whole Asia. Right now, my team has already set up the proper infrastructure and also starting to talk to our countries, especially Indonesia, to see how we can bring this to expand it for the motor business, because for motor business, it is slightly more complicated. Motor fraud is not as straightforward as travel or PA where you submit a claim you can analyze from there. Motor fraud can happen in many areas. You can start all the way from the whole value chain, so the model will be a bit more complicated and you need more resources and tacit knowledge to run it. So that’s our plan for the next year at least.

Very exciting times to be a Sompo customer. Now that you guys have been involved in a 100E which is an initiative at the national level, what do you think about our national AI strategy?

I think our national AI strategy is going in the right direction for sure, which means we are trying to bring our AI capability for the whole Singapore starting from the schools, education, even the industry level .. I think the direction is definitely good. If you ask me what feedback do I have, maybe one part is really to identity potential use cases by leading industries that can be shared across different industries. One thing I advocate is design thinking which I always believe that whatever that we learnt from one industry, the same problem may have occurred already in some other industries. So there’s no need for us to wait. If the national AI strategy is able to bring all these together, which means, for example, the airlines might have used AI lot within their whole processes like the customer service, the terminals and so on. If such use cases can be shared and adopted to another industry, that will help to speed up our AI adoption.

You mentioned design thinking. In fact, we do have design thinking as part of the mandatory training for our engineers. So, hopefully, we at AI Singapore are on the right track for that. Yes, thanks, Patrick, for sharing the insights from the insurance industry and letting us know about the big plans at Sompo. On behalf of AI Singapore, I wish you success in your future endeavours.

Thank you very much. Thank you, Basil and the team.



[*] This conversation was transcribed using Speech Lab. The transcript has been edited for length and clarity.

Related Story

Author