AI & Claims Investigations: Must-Know Terms & Tools for Insurance Professionals

As the team that identifies and/or follows up on any “red flag” claims, our Special Investigations Unit (SIU) serves as the connective tissue between a lot of different service areas: claims, loss control, legal, etc. 

We’ve seen that a limited understanding of AI’s capabilities among these various industry players can also limit its impact. Equipping all industry professionals—not just claims handlers—with a basic knowledge of how AI tools work in claim investigations helps ensure everyone understands what’s possible. 

According to a study by the Coalition Against Insurance Fraud (CAIF), the estimated annual cost of insurance fraud in the U.S. for 2025 is over $308 billion annually. Healthcare fraud is estimated at $105 billion, property and casualty at $50 billion, and workers’ comp between $32-44 billion. Ultimately, these fraudulent insurance payouts contribute to rapidly rising premiums that impact both employers and employees. 

To help everyone in the insurance ecosystem identify and curtail these fraudulent claims, here’s a quick look at common AI terms and the related claims investigation tools available. 

Your Brief AI Glossary 

Machine Learning
Learns from data and improves performance over time without being explicitly programmed.
Sample Applications: Email filtering and fraud detection 

Deep Learning
Summarizes, translates, predicts, and generates human-like text to convey concepts and ideas.
Sample Applications: Image recognition and autonomous vehicles 

Natural Language Processing (NLP)
Understands, interprets and generates human language by combining linguistic rules with machine learning models.
Sample Applications: Chatbots, language translation, document summarization 

Computer Vision
Interprets and processes visual information (pictures and video).
Sample Applications: Facial and license plate recognition 

Generative AI (GenAI)
Creates new content based on learning patterns from existing data.
Sample Applications: Copy, code and images 

Predictive AI
Analyzes historical data for patterns, trends and correlations to make predictions about future outcomes.
Sample Applications: Forecasting claim severity, customer demand or litigation probability 

How AI Tools Assist with Claim Fraud: 

• Identifies Red Flags 

• Analyzes Links 

• Recognizes Deep Fakes 

• Reviews Documents & Audio Recordings 

• Researches Internet & Social Media Presence 

• Assists with Medical Canvassing 

• Generates Fraud Scoring 

• Makes Real-Time Recommendations 

Our AI Investigation Tools 

In tandem with traditional investigation methods, our team applies AI-assisted tools from claim intake to closure. Here are several examples of what we already actively use or are piloting. 

Risk & Fraud Scoring 

We use a custom Risk & Fraud Scoring tool to analyze a claim, identify any red flags, then provide the scoring output summary to adjusters or investigators for direction as to how to handle/investigate the claim. For adjusters in particular, we are also able to generate recommendations on how to handle the claim. 

Predictive Analytics 

We receive the scoring output—now what? This is where predictive analytics come in. It forecasts the claim costs and complexity, flags the ones most likely to end up in litigation, and also identifies any suspicious patterns/relationships between claimants, medical providers, attorneys or other associated parties.   

Open-Source Intelligence (OSINT) 

OSINT is a technical term for the internet/social media investigations that complement our human investigations when we need to take a closer look at a claim. We use it to scrub for related social content, preserve the content in a method ensuring court admissibility, and uncover and link accounts to the subject of the investigation (aliases, different usernames, etc.). 

Image & Video Forensics 

We use AI to help identify doctored forms, medical records, images, videos, deepfakes and other content that would be hard for a person to identify as fake. 

Recorded Statement Analysis 

We use Natural Language Processing (NLP) to detect anomalies in tone, hesitation, stress, or specific phrases used during recorded statements that may indicate deception. We also use it to analyze notes, statements, interviews, etc. to detect inconsistencies or red flag phrases.  

Now that you know the tools available: if you’re an insurance professional who could use help with deeper research and recommendations for your claims, find out more about Davies Special Investigations Unit and contact us.