AI vs AI: Lessons from Aviva on Fighting Fraud in the Age of Generative AI

TH

For years, insurance claims have relied on one assumption: evidence represents the truth. Photos, repair estimates, invoices, and supporting documents have been the foundation for verifying claims.

However, Generative AI is challenging this approach. Today, realistic images, documents, and fake evidence can be created within minutes, making fraud more sophisticated and harder to detect through traditional methods.

A fake accident image, repair quotation, or claim document can now be generated with widely accessible AI tools. The challenge is no longer just the rise of fraud cases, but that fraudsters are using the same technologies organizations rely on.

One example is Aviva, which used AI to detect and prevent over £230 million in fraudulent claims across 18,400+ suspicious cases by analyzing large-scale data, identifying abnormal patterns, and prioritizing high-risk cases for investigation.

When AI Makes Fraud Easier, Fraud Detection Must Become Smarter

In the past, creating fake evidence for insurance fraud required specialized skills. Manipulating images required advanced editing tools, creating convincing documents required design expertise, and fraudulent claims often depended on complex coordination between multiple parties.

Generative AI has significantly lowered these barriers. Tasks that previously required hours or days can now be completed within minutes through simple prompts, producing results that appear realistic enough to pass initial human review.

Common examples include:

• Generating fake accident images: AI can create realistic images of vehicle damage, including dents, scratches, lighting, and shadows, making it difficult to distinguish between real and AI-generated evidence.

• Creating or modifying claim documents: Repair estimates, invoices, medical reports, and supporting documents can be generated or altered to appear legitimate, increasing the complexity of verification.

• Claims inflation: AI can also be used to manipulate costs by generating inflated repair estimates, increasing parts prices, or creating exaggerated damage descriptions to maximize claim payouts.

As fraud techniques become more advanced, organizations need a new approach that goes beyond traditional manual reviews.

How Aviva Uses AI to Detect Fraud

Instead of simply increasing the number of investigators, Aviva uses AI to analyze massive amounts of data and identify patterns that may indicate fraudulent behavior.

The role of AI is not to automatically approve or reject claims. Instead, it acts as an intelligent screening layer that identifies high-risk cases and prioritizes them for further review by human experts.

Aviva’s approach highlights three key capabilities

1. Connecting Data Across Multiple Sources: AI analyzes relationships between claimants, vehicles, repair shops, payment accounts, and historical claims to identify hidden patterns, repeated behaviors, and potential fraud networks that may not be visible when reviewing cases individually.

2. Validating the Consistency of Evidence: AI checks whether claim details, images, and documents align with each other, such as whether vehicle damage matches the reported accident, timelines are consistent, and supporting evidence is reliable.

3. Detecting Abnormal Claim Costs: AI compares repair costs, replacement parts, and claim amounts against historical data and market benchmarks to identify unusually high claims and reduce losses from inflated payouts.

AI Supports Decisions, But Humans Remain in Control

Although AI plays an important role in fraud detection, Aviva does not replace human judgment with automation. The company follows a Human-in-the-Loop approach, where AI analyzes information, identifies risk levels, and supports decision-making, while final claim decisions remain with experienced investigators.

This combination allows AI and humans to focus on what they do best. AI helps process large amounts of data and detect hidden patterns at speed, while human experts apply judgment, context, and experience when handling complex cases.

This approach also helps reduce False Positives, ensuring that legitimate customers are not incorrectly identified as fraudulent and protecting overall customer trust.

Key Lessons for Organizations Beyond Insurance

The case of Aviva demonstrates that AI-powered fraud is not limited to the insurance industry. Any organization that handles large volumes of documents, transactions, or sensitive data faces similar challenges.

Banks, financial institutions, e-commerce platforms, healthcare providers, and government organizations are all facing a future where fraud attempts can become faster, more convincing, and harder to detect.

Key lessons include:

• Analyzing individual transactions is no longer enough: Modern fraud often leaves small signals across multiple sources. Connecting data points provides a clearer picture than reviewing each case separately.

• AI should assist, not replace human decisions: AI can improve efficiency by identifying risks and prioritizing cases, while experts remain responsible for final decisions.

• Fraud detection systems must continuously evolve: As fraud techniques change, AI models must continuously learn and adapt to detect new patterns effectively.

Building the Next Generation of Fraud Detection with AI

As AI becomes accessible to both organizations and fraudsters, the ability to detect abnormal patterns quickly and continuously will become a critical capability for businesses across industries.

At Sertis, we help organizations develop AI Solutions for fraud detection and risk management, including Anomaly Detection, Risk Scoring, Network Analysis, and AI models that continuously adapt to changing fraud patterns.

By combining advanced AI capabilities with human expertise, organizations can build more resilient systems, reduce risks, and make better decisions in an increasingly complex digital environment.