Welcome to this week’s issue of Risk, Bytes and Beyond - your bi-weekly roundup of the latest insights at the intersection of data analytics, artificial intelligence and insurance. Delivered to your inbox every second Friday! In each edition, you’ll find:
Curated content from experts in data and insurance
An insight brief on a changing topic
The latest headlines on data and AI in insurance
Thanks for reading - let’s dive in!
💡 Noteworthy Content
This article explores how AI’s current use in insurance tends to optimize back-end efficiency rather than fundamentally disrupting business models. It argues that while generative AI and LLMs are gaining buzz at conferences, few insurers are truly taking the plunge into reinventing customer engagement or product innovation.
🔗 DigFin
An interesting podcast featuring Clyde & Co partner Darryl Smith, discussing the potential risk that traditional insurance policies - such as professional indemnity insurance - may be required to cover AI-related losses, despite not having priced in this emerging exposure.
In this article, WTW’s Pardeep Bassi explores how insurers can harness AI-driven tools to enhance pricing, portfolio management, underwriting, and claims handling. He emphasizes the need for insurers to strategically integrate AI to remain competitive and improve efficiency in a rapidly evolving technological landscape.
💭 Insight Brief
Fighting Insurance Fraud with Data: From Gut Feeling to Algorithms
Insurance fraud is a costly problem. It’s estimated that fraudulent claims cost the global insurance industry hundreds of billions of dollars annually. These losses drive up premiums for honest customers and erode trust in the system.
Traditionally, fraud detection in insurance has relied heavily on manual review and the intuition of experienced claims handlers. While human judgment is valuable, it has clear limitations, especially in today’s high-volume, digital-first environment.
That’s where data comes in. Modern insurers are increasingly turning to data science and machine learning to detect and prevent fraud more efficiently and at scale.
Here’s how data is changing the game:
1. Pattern Recognition in Claims Data
Machine learning models can be trained on historical claims data to recognize patterns and red flags associated with fraudulent activity. For example:
Unusual claim frequency (e.g. multiple claims from the same customer or address)
Suspicious timing (e.g. claims filed shortly after policy inception)
Inconsistent or improbable claim details (e.g. damages that don’t match the event described)
Such models do not (yet) replace human judgment, but they help focus investigative resources on the most suspicious cases.
2. Network Analysis: Uncovering Hidden Connections
Some types of fraud involve organized groups, such as repair shops, doctors, and claimants working together to game the system. These schemes can be hard to detect if you only look at claims individually.
That’s why insurers are increasingly using graph analytics and network detection. By connecting data points across policies, claims, and third-party vendors, they can reveal hidden relationships:
Claimants using the same repair shops or legal advisors repeatedly
Groups of individuals involved in each other’s accidents
Unusual clusters of activity in a specific region
Graph algorithms can detect these patterns far more efficiently than manual reviews.
3. External Data Sources: Enriching the Picture
Fraud detection becomes even more powerful when internal claims data is combined with external information:
Telematics data (e.g. vehicle speed and location) can help validate or challenge the narrative in a motor claim.
Social media and public records can reveal inconsistencies or evidence contradicting a reported loss.
Weather data or satellite imagery can confirm whether a storm actually occurred in the claimed location.
By integrating these sources, insurers can build a more complete and realistic picture of each claim.
4. Real-Time Alerts and Automated Triage
With the right infrastructure, fraud detection can happen not just after the fact, but in real-time. Modern fraud systems can automatically flag high-risk claims for further review as soon as they are submitted.
This allows for:
Faster processing of clean claims
Earlier intervention on suspicious ones
Better allocation of investigative resources
Challenges and Considerations
Despite the promise of data-driven fraud detection, there are still challenges:
Data quality and availability are foundational. Poor or missing data weakens the models.
Bias and fairness must be carefully managed to ensure models don’t unfairly target certain groups.
Interpretability is key: Investigators and regulators need to understand why a model flagged a claim.
Still, when done responsibly, data-driven fraud detection offers a powerful advantage.
📰 Industry Headlines
Five U.S. states recently embedded web trackers in their public health insurance exchange sites, inadvertently sending sensitive user data - such as prescriptions, provider details, pregnancy status, and disability - to third-party platforms such as Google, LinkedIn, and Snapchat. This raises major concerns around data privacy, regulatory compliance, and consumer trust for insurance professionals and state-run health portals.
Scania has confirmed that a cyberattack in May 2025 on its third-party managed insurance portal exposed some 34,000 insurance claim documents, potentially including personal, health, and financial data. The breach was discovered after extortion attempts by hackers, prompting Scania to take the affected subdomain offline, launch an internal investigation, and alert authorities and potentially impacted individuals.
Quandri, an AI-driven platform looking to enhance automated policy reviews, renewal insights, requoting, and proactive client outreach for insurance brokerages and agencies, has secured a $12 million investment led by Framework Venture Partners alongside Intact Ventures.
Was this email forwarded to you? Sign up here!