Success story
Alliance Medical: Automating Clinical Data Validation with Rule-Based Flagging
How a leading medical company replaced manual clinical data validation with an automated, rule-based system that laboratory staff could trust and manage themselves.
fewer false positives
saved in claim costs
operations
through automated case routing
built into every decision
Managing millions of policyholders across multiple insurance lines, UnipolSai faced a relentless challenge: detecting increasingly sophisticated fraud schemes without compromising the speed and reliability of claim handling. With Rulex, the company identified a strategic opportunity to move beyond traditional methods, integrating AI-driven analysis with greater transparency and operational control.
Although the tools in place were technically advanced and able to flag potentially suspicious claims, investigators lacked visibility into the reasoning behind each alert. Like many black-box machine learning systems, it surfaced cases requiring attention without explaining why – making it harder to assess them quickly, filter out legitimate claims, and maintain a smooth customer experience.
UnipolSai is one of Italy’s largest insurers, serving millions across motor, health, property, and life insurance. With a nationwide presence, it supports individuals and businesses with comprehensive risk-management solutions, combining broad coverage with ongoing investment in innovation and digital transformation.
What UnipolSai needed was a fraud detection solution that went beyond identifying suspicious claims, by clearly explaining the reasoning behind every alert: a system with built-in accountability.
Improving insurance fraud detection with explainable AI
To redesign the UnipolSai fraud detection process, Rulex introduced an explainable AI solution built to operate in real-world insurance conditions, where fraud cases are relatively rare and high-quality labeled data is limited.
The solution combined clear logic with statistical insights to identify unusual patterns. For each flagged case, the model provided a transparent explanation of the reasoning behind the alert – for example, a claim filed unusually close to a policy’s expiration or suspect relationships between involved parties – offering claim handlers intuitive, traceable justifications they can use in their assessments.
The transparency of the approach fostered trust among the users, who could now see precisely which behaviors or inconsistencies prompted an alert. This explainability also ensured alignment with privacy and regulatory requirements such as the GDPR.
The insurer’s innovative approach was recognized publicly, receiving the 2019 Italy Insurance Forum Award for Best Anti-Fraud Solution.
A smoother experience for business and customers
Once Rulex Platform was integrated via APIs into the company’s core claims system, the explainable AI technology began supporting teams in several practical ways:
A 10% reduction in false positives reduced unnecessary claim reviews and minimized delays for legitimate customers.
More than €50 million in claim costs were saved through more accurate fraud detection.
Automated triage improved the routing of cases to the appropriate departments, reducing manual workload and accelerating handling times.
Tech point
Take a closer look at the core capabilities of Rulex Platform behind this insurance fraud detection solution
01. Explainable AI architecture
The system used Rulex’s proprietary explainable AI technology to extract clear IF-THEN rules from UnipolSai’s historical claim data. Unlike traditional fraud detection software or black-box machine learning, every prediction included traceable rules. These rules showed exactly which conditions contributed to a “fraud”, “suspicious”, or “legitimate” classification. Rulex’s approach ensured interpretability for audit purposes and maintained compliance with internal and regulatory standards.
02. Sparse-fraud optimization
Fraud is rare by nature, which leaves most traditional models with too few confirmed examples to learn from effectively. Rulex’s proprietary algorithm is designed to work like the human brain, which can learn from a minimal number of positive examples.
03. Anomaly categorization
The system automatically categorized different kinds of anomalies, including inconsistent reporting, suspicious timing patterns, deviations from a claimant’s historical behavior, geographical patterns, and network-linked irregularities involving claimants or providers.
04. System integration
The fraud engine was deployed via REST APIs, integrating directly into the insurer’s core system for fraud detection. Users can upload data and inspect results through interactive custom UIs.
05. Governance & Compliance
For insurers handling insurance fraud detection at scale, regulatory alignment isn’t optional. The explainability built into the system ensured alignment with GDPR transparency requirements, internal audit frameworks, and fair-AI assessment standards.
Why Rulex for insurance fraud detection software
Most fraud detection software forces a trade-off: accuracy at the cost of transparency, or simplicity at the cost of precision. Rulex Platform is built to eliminate this trade-off, combining explainable AI, operational integration, and decision-level control so that insurers can reduce risk and run leaner investigations without compromise.
Explore more financial services case studies powered by Rulex Platform
Early warning system
Timely identifying signs of potential risk in open banking data.
NPL management
Improving NPL strategy and allocation across collectors and territories.
Wealth management
Automating the creation of tailored portfolios.
Client retention
Timely identifying clients at risk of churn to implement corrective actions.
Dynamic Pricing
Optimizing pricing with explainable, human-supervised automation.
Insurance fraud detection
Accurately identifying fraudolent claims with limited historical data.
Frequently asked questions
What is explainable AI (XAI) in insurance fraud detection?
Explainable AI in insurance fraud detection is a machine learning approach that produces human-readable justifications alongside each prediction. Rather than outputting a fraud score with no context, an XAI system reveals the specific conditions that triggered the alert – such as a claim filed days before policy expiration, or overlapping party networks – so investigators can evaluate and act on results with confidence.
How does Rulex's XAI platform detect fraudulent insurance claims?
Rulex Platform extracts if-then rules from historical claims data and applies them to incoming cases in real time. Each flagged claim is classified as fraudulent, suspicious, or legitimate, with a full trace of the rules that led to that classification. The system also categorizes anomaly types – timing irregularities, behavioral deviations, geographic patterns, and network-linked irregularities – so investigators know not just that something is wrong, but what kind of wrong it is.
Why is explainability important in AI-based fraud detection?
Explainability matters for three reasons. First, it builds operational trust: investigators are more likely to act on an alert they understand. Second, it supports compliance – GDPR and fair-AI frameworks require that automated decisions affecting individuals be explainable and auditable. Third, it reduces errors: when the reasoning is visible, it’s easier to catch and correct model mistakes before they affect customers.
How does Rulex handle fraud detection when labeled fraud data is scarce?
Rulex is optimized for sparse-fraud environments, where confirmed fraud cases make up a small fraction of overall claims. The system is designed to learn from minimal positive examples, identifying anomalies relative to established behavioral baselines rather than relying on large labeled datasets.
How is Rulex's approach to fraud detection different from traditional machine learning?
Traditional machine learning models in fraud detection are often black boxes: they produce a risk score but cannot explain which factors drove it. Rulex uses a rule-extraction approach that generates explicit, auditable logic from the same underlying data. The result is comparable or superior accuracy with the added benefit of full interpretability – meaning every prediction can be reviewed, challenged, and justified by a human investigator.