Fraud Vision 

How severe a problem is fraud for the financial industry overall?

In the financial industry, document fraud is the main issue in credits and loans and also in insurance. The two elements are quite different, in credit, loans and factoring, companies are lending or giving money to individuals or other companies and base their ratings on documents. In France it represents a loss of 3% of revenue for these companies. Some of them went into bankruptcy due to fake documents. Urica is a London-based fintech that went bankrupt because they financed a company that did fake invoices.

In insurance, the problem is different because you can create fake documents—if your house burns and you want to send invoices to the insurance companies, you can send fake invoices, you can increase the amount of the invoices, and this represents a big problem for these companies because it’s from 5% to 7% of their revenues. It is an enormous amount of money. They have no tools to detect it. Today they analyze these kinds of documents with the naked eye, so they need tools and information to analyze everything. Today, 7% to 15% of insurance files include fraudulent documents.

In France, there was an 86% increase in fraud between 2020 and 2021, which is huge, and we think that will continue. Just in France, fraud related to documents represents more than €40 billion. Globally, it increases every year because companies don’t have any solutions and it’s very easy to do. According to FiVerity’s Cyber Fraud Network, synthetic identity losses at U.S. financial institutions have reached $20 billion.

The type of insurance that has the most fraud each year is life insurance, with annual losses of nearly $75 billion. Medicare fraud is a close second, costing the federal government approximately $68 billion annually. Workers compensation fraud results in $34 billion in annual losses, much of it concentrated in claims fraud.

Where does fraud most commonly arise in the insurance process?

There are two parts—during the onboarding, because you can send fake driver’s licenses, or you send fake information about you or your company. The second part is when there is conflict or a reimbursement due, because you can send fake invoices or fake police statements. So 80% of what we analyze is on the second part, when there are reimbursements, and 20% is for the onboarding of the client.

Does it differ by lines?

The frauds are kind of similar because most of the time fraudsters falsify documents. The kinds of documents you are trying to falsify depends on the insurance company, but the fraud remains the same. The fraud in insurance differs in relation to the size of the clients of the insurer because individuals do fraud more often than companies. An insurance company that works with individuals in B2C has more fraud than insurance companies that work in B2B.

Where is fraud most common at the commercial level?

It occurs most often at the time of subscription or registration. In the case of the banking sector, this will be when applying for a loan at the bank. In the case of insurance, this will happen more at the time of the claim with a modification of the invoice in order to obtain a higher compensation. Since the bulk of the fraud is on the compensation side, if the brokers are on the front line of the claim, they will be the ones most likely to intercept the fraud.

How much of a risk do you expect from AI-generated documents?

Today it is not possible to generate fake documents with AI. We are working on addressing it, however, because it is a big topic. You already have websites that follow you to generate fake documents, and you also have data on the dark web that is used to generate fake documents. That doesn’t use AI, but still they are very powerful, and the documents are very well done. Today there are no fake documents generated by AI, but we must be aware because we need to develop a solution. Our vision is that in two years, we will build AI that will fight other AI that generates fake documents.

How does technology address document fraud?

We analyze every part of the document. First, we analyze anomalies in the document, if there are variations of the document. When you create a document from invoice software or pay-slip (payroll) software, the documents are always the same, and if you modify it, you change an element on it; the structure of the document will be different; the code of the document will be different. We analyze this code to tell you if there is something that’s not fine and if there are false elements. We also do the same thing with digital analysis.

We analyze the documents using computer vision algorithms, and we can say here the element is not fine because the color black is wrong, because the font is not right, or you have a strange space between words. We analyze everything. We do various analyses on the consistency of the document. If you want to modify a pay slip, you have some elements that have to be normal. We verify that the address works. We verify if the amounts are fine, if the rates are good.

We analyze if the documents come from the dark web or a website that follows you to generate fake documents. We do a very complete analysis in real time. It takes about five to 10 seconds per document, and—after—we give reports to our client and say this document is fine or this document has a higher risk because some of the elements have been changed.

You’ve begun in France, but you’re planning to expand. Can you please tell us about that?

Today we work mainly in France, but we are opening in new countries, for example, French-speaking countries, part of Canada, and Switzerland and Belgium and North Africa. We are also opening in new countries in Europe, like Italy and Spain and maybe Germany. We are thinking about going into the U.S. next year or the year after. First of all, the U.S. market is huge, and it’s a very good economic opportunity for us.

The solution is really applicable in the U.S., which is why we partnered with Sunlight Solutions, which is a cloud-based suite for insurance carriers supporting all insurance products, because they can work with us in the U.S. today. The Finovox solution is applicable to every language in every country. We analyze documents in 200 languages for clients. If the solution is already applicable to every country, we can just expand and scale.

How do you keep ahead of fraudsters?

The risk is different in every country and also in every sector. What we do is organize fraud contests. We will ask 1,000 U.S. citizens to falsify documents so we will understand the fraud, the kinds of documents and how it works in the country. It is very important to do that to stay the best. For example, the next fraud may be to generate fake documents using generative artificial intelligence, and on our next contest users will do it and that’s how it will work. We adapt to fraud in different countries, but what we see is that the fraud is mainly the same. You have some differences that now occur in France, Spain, the U.K. and in the U.S. It’s kind of similar: you have an original document, and you change something with your computer, you use Photoshop or any software, and you send it to the insurer, to the bank or the real estate company.