How the Insurance Industry Can Use AI to Detect Fraud

May 30th, 2024

Artificial intelligence (AI), a hot topic in the news, promises to significantly impact the insurance industry in the years ahead. To give group captive member-companies a clearer idea of the future, we invited Mike Lundgren, director of marketing at Ethos Risk, to host a recent Captive Resources Risk Control Webinar focusing on a leading growth area for AI in insurance — fraudulent claim detection. Lundgren covered:

  • What AI is and its current industry uses and investment level.
  • AI methodologies fraud detection applications might employ.
  • Barriers to greater future adoption.
  • Priorities to focus on to overcome these barriers.

Read on for Lundgren’s insights, which can help guide your AI implementation strategy.

AI Defined and Current Industry Uses

Defining AI in a fraud detection context, Lundgren cited the Coalition Against Insurance Fraud: “A data technology solution that systematically screens for questionable claims using predictive analytics logic driven by advanced modeling techniques and statistical algorithms.” AI’s strength lies in its use of this logic, modeling, and algorithmic problem-solving combination to rapidly interpret large quantities of data beyond human capability.

Lundgren noted that the industry has primarily used AI in marketing, product development, underwriting automation, and claims processing. He added that although nearly 60% of insurers already use AI to combat fraud, perpetrators are changing tactics with their own use of AI, such as image manipulation. As a result, they likely need to invest in new tools to keep up.

Lundgren said that the coronavirus pandemic, which generated $55 billion worth of industry losses in 2020, according to Swiss Re, has motivated industry leaders to digitize their operations for cost efficiency — and AI adoption is part of this effort. He pointed out that industry investments in enterprise-level AI-enabled solutions will reach $500 billion in 2027, according to International Data Corp. (IDC).

Indicating that the audience should start becoming familiar with AI language, Lundgren also described several methodologies AI uses in various fraudulent claim detection applications.

AI Methodologies for Fraud Detection

Anomaly Detection

A claim with data exceeding a historically based threshold triggers an event that might warrant a fraud investigation.

Network Analysis

AI’s robust computing capability can detect relationships between individuals or groups based on similar data posted on social media networks and uncover organized crime ring activity, for example.

Natural Language Processing (NLP)

This capability can be used to gather text-based client information and compare it with other records, such as previous claims or criminal records.

Machine Learning

One of the most powerful AI capabilities, this uses algorithms to familiarize computers with data patterns, with or without preset thresholds, and makes them “smarter” in detecting fraudulent claim patterns over time.

Speech Recognition

This capability uses an algorithm that can perform sentiment analyses of conversations with the potential to indicate possible fraud.

Image and Vision Analysis

Technologies are available to detect images that are altered or duplicated from webpages and used in fraudulent or enhanced claims, particularly against auto collision or liability policies.

Web Crawling

With this capability, computer systems can quickly gather data from policyholders’ social media accounts and detect contradictory claim evidence such as imagery.

Barriers to Future Adoption

Lundgren noted that insurers face public policy challenges in increasing AI adoption. He predicted significant legislative and regulatory activity in consumer protection, especially regarding data privacy and preventing false positives among low-income groups.

Another barrier to adoption is simply a lack of awareness of how AI works, Lundgren said. Referring to the use of AI for image manipulation, for example, Lundgren recommended that industry leaders continuously study and adapt to increasingly high-tech threats. That said, he indicated that insurers are not yet fully leveraging AI's potential to address their organizations’ specific fraud detection needs effectively.

Priorities for Maximizing AI Effectiveness

To be used effectively, Lundgren said AI needs four qualities he calls “the 4 V’s”:

No. 1: Variety

AI requires a wide variety of data. Third-party data can provide industry-level depth and breadth to a company’s internal data.

No. 2: Value

Lundgren pointed out that data quality is critical. Insurers often have incomplete or miscoded data, he added. As a result, many insurance carriers are increasing investment in their data.

No. 3: Volume

The insurance industry collects large volumes of policyholder data and is well-positioned to use AI. In-depth data can help investigators uncover fraudulent claims more efficiently.

No. 4: Velocity

Time is typically of the essence in fraud detection, so AI systems need robust computer processing capacity. Cloud-based computing is a potential solution to a capacity deficit.

About the Webinar

This presentation was part of Captive Resources’ Risk Control Webinar Series — regular installments of webinars to educate the group captive members we work with on topics like workplace safety, organizational leadership, and company performance. The thoughts and opinions expressed in these webinars are those of the presenters and do not necessarily reflect Captive Resources’ positions on any of the above topics.

Share this article