Artificial Intelligence vs Fraud

What is the role of Artificial Intelligence?

Since the pandemic, we’ve been feeling the push of digitization but the insurance industry was it hard by the move to remote everything. According to McKinsey, investing in AI may equate to a potential annual uptick in value of $1.1 trillion for the industry. You can buy cars online, jewelry, why not insurance?

What AI can offer investigators is more speed and accuracy in verifying claims. AI can sift through incredible amounts of data far quicker than a person. Think about doing massive mathematical computations by hand, or just using a calculator. In the end, what it is helping you do is make decisions with information that comes faster to you. Ideally, it’s using existing information to find patterns, which can help defend against fraudulent activity – or at least help raise red flags for further exploration. 

Insurers can also use AI and advanced analytics capabilities to analyze Internet of Things (IoT) data to help better identify potential disasters before they happen and nudge policyholders to take corrective and preventive actions.

 

In 2017, respondents who say their organization has adopted AI in at least one business unit or function was 20%, whereas today this figure stands at 50% in 2022.

Tools of the Trade

Anomaly Detection

Here the user defines baselines in the software for key performance indicators (KPIs) associated with tasks or events, then sets thresholds. When a threshold for a particular measure is exceeded, then the event is reported. These are called anomalies, as they are outside the standard data set for acceptable parameters of claim validity. Outliers like these are used to identify existing, or new and previously unknown, fraud patterns. Something that would be extremely time consuming and tedious for a human to do but quick for a machine.

Network Analysis

Organized fraud involves more than one claim, and generally more than one type of business. It remains a pervasive and growing problem for the industry. The solutions are in more sophisticated technologies. New platforms now can analyze social networks in a rapid fashion looking for patterns or connections that would take a human an endless amount of time. Network link analysis is a proven method found to be effective at identifying organized fraud through modeling algorithms. These models look at relationships of the entities involved in multiple claims and even new business acquisition. Fraud investigators have been using this technique for some time, but AI’s ability to cross-reference massive amounts of data makes this technique a much easier and more effective way at connecting the dots. The sheer level of data analysis is at levels never before seen.

Natural Language Processing

This arm of AI helps computers understand human language. And more than just understand it, NLP helps them interpret and manipulate this language. Think of it as a mixed form of computer science and computational linguistics. Its goal is to close the distance between human communication inputs and the ways computers understand it.

As an example, when you ask Alexa who won the gold medal in downhill skiing at the 1956 Olympic games in Italy and she gives you a recipe for soup, the NLP isn’t yet working as well as it could.

Machine Learning (ML)

Machine learning does not use rules-based programming, rather it uses mathematical algorithms to learn from the data it processes. Each time the algorithm completes an iteration of review it becomes smarter and better. This in turn delivers more accurate results. There are two types of ML, known as supervised and unsupervised learning. Supervised uses rules with a known dataset to evaluate accuracy, whereas the unsupervised makes sense of unknown data by looking at the patterns and features on its own. The difference is the unsupervised is better at detecting activities that are potentially problematic when an investigator isn’t sure what specific data could flag a fraudulent claim.

In other words, supervised is used to predict future activity and unsupervised is used to build strategies or identify patterns or segments from an existing dataset you may be working.

Speech Recognition

A fascinating algorithm that helps an investigator by raising alarm bells they would be unlikely to notice during normal conversation. Sentiment analysis is used when communicating with a client. It analyzes their speech patterns and voice wavelengths. This tool is capable of tracking specific keywords during conversation that relate to various sentiments of the topic. It can then make an assessment that further investigation is necessary by showing indicators of fraud solely through how and what they are saying.

Image and Vision Analysis

Aptly named, this algorithm can scan the photo of the object of the claim and cross-reference available data to make sure the item in the picture is actually what is insured. This could be for a broken television, flooded basement, or cracked windshield or taillight. If it is something unrelated to the claim it can flag it for additional investigation. Another nice feature is its ability to quickly verify whether the same photo has been used before. It can happen where a fraudster uses stock images or photos from the internet or other previously successful claims, and the image analysis can pick this up. It can also tell if the photo has been manipulated in some way or altered to enhance the claim amount.

Web Crawling

Searching publicly available social media profiles is part of the standard kit these days. Imagine the benefit of having a software program that can crawl the web for you and find evidence of claims that raise fraud suspicions. Sign us up! Web crawling algorithms can mine this data and examine the text on all social media platforms to help speed up the discovery process.

AI Challenges

AI automation and reduction in fraud sounds great but before this can happen, the adoption rate must increase. There is certainly progress and many insurers are using some form of automation. The majority however are moving slowly. One reason for this is lack of legislative history. The development of regulatory provisions is still ongoing and will take some time. Insurers are naturally concerned to invest significant monetary and time resources into something that may end up being ruled as improper or even as a disqualifying measure in a court case. Particularly in cases where the claim is denied.

1. Over 50% of P&C companies surveyed by the CAIF claimed their biggest concerns to be data related – specifically the quality and potentially insufficient size to make it reliable.
2. Data privacy too has become a colossal national issue yet only 15% of companies surveyed thought it was a potential problem.
3. Upfront costs are the first thing perceived by companies. Although AI results in higher revenues in the long term, its a tool that needs to be implemented, employees have to be trained on it and there will be costs and mistakes along the way.

Bottom Line

Adoption of AI is rapidly growing. Its making everyone’s jobs more efficient although it may come with some upfront costs. Artificial Intelligence has been adopted by marketing, underwriting, and fraud detection. But this tool is only as effective as we train it to be. The insurance industry has the resources and motivations to make AI work. Reduction in fraud, improvement in data analysis and a larger data pool, are just some of the benefits AI brings to our industry. The use of this technology will never replace the tried-and-true capabilities of human investigators or analysts. But it does have the potential to help assist everyone from preventing and catching fraud as it occurs. Stay informed about the changes happening in our industry through subject matter experts and the latest content.