The SmartRisk approach to efficacy testing, anonymisation, and explainability

Contacts: Dr Karim Derrick , Joe Cunningham , Harvey Maddocks , Xi (Lisa) L.

Part 6 of the Kennedys IQ SmartRisk Series

In our journey through AI and professional judgment, we’ve explored the power of Large Language Models (LLMs), structured decision-making, and the role of neuro-symbolic AI in claims handling and underwriting. But how do we ensure AI-assisted decisions are accurate, fair, and explainable? In this final installment of the Kennedys IQ SmartRisk Series, we examine how efficacy testing, anonymization, and explainability make SmartRisk a trusted AI solution for insurers and legal professionals.

The need for rigorous efficacy testing

One of the most pressing challenges with AI in professional services is maintaining accuracy and consistency. Unlike traditional software, AI models evolve over time, meaning they require continuous efficacy testing to ensure optimal performance.

How SmartRisk ensures high performance:

1. Gold-standard data calibration – Experts curate a benchmark dataset for testing AI performance against human decisions.
2. The Akhil Test – Named after the pioneering work of Insurance Data Science expert Akhil Sivinand, this method ensures at least 95% accuracy in attribute extraction and decision modeling.
3. Continuous model refinement – SmartRisk adapts and improves through real-world case assessments, ensuring AI-assisted decisions align with expert-level performance.

With foundational AI models constantly evolving, this robust testing process ensures that SmartRisk remains reliable, auditable, and legally defensible.

Anonymization: protecting sensitive information

For AI to be effective in claims handling and underwriting, it must process large volumes of insurance policies, legal documents, and client records—often containing personal and sensitive information. SmartRisk’s anonymization process ensures compliance with GDPR, data protection laws, and ethical AI standards.

How SmartRisk protects privacy:

• Structured data filtering – Directly identifies and redacts specific personal identifiers such as claimant names and policyholder information.
• AI-powered text anonymization – Uses TextWash, an advanced natural language anonymization model, to detect and mask identity-revealing data.
• Cardinality preservation – Ensures redacted entities retain consistent references (e.g., “NAME1” and “NAME2”) to maintain contextual meaning.

This approach enables highly secure AI decision-making, allowing insurers to benefit from AI-driven insights without compromising privacy.

Explainability: ensuring AI is transparent and accountable

A core requirement for AI in professional decision-making is explainability—the ability to understand and justify why a decision was made.

Unlike black-box AI models, SmartRisk provides:

• Step-by-step decision justifications – Every decision is broken down into logical components, explaining how individual attributes contributed to the outcome.
• Evidential reasoning mapping – Ensures traceable, rule-based AI decisioning that can be audited and challenged when necessary.
• Bias monitoring and adjustment – Regular screenings by qualified data scientists ensure that AI does not reinforce prejudices or systemic biases.

By embedding explainability at its core, SmartRisk builds trust and ensures compliance with legal and regulatory frameworks in insurance and legal decision-making.

The future of AI in professional services starts now

Over this six-part series, we have outlined how Kennedys IQ SmartRisk is pioneering the next generation of AI-assisted professional judgment—combining LLMs, structured decision-making, and rigorous testing to deliver an AI system that is accurate, explainable, and legally defensible.

Last week, Kennedys IQ will unveiled SmartRisk, the first AI system specifically designed for professional services. Be part of the revolution in risk assessment and claims decisioning.

Contacts us and experience the future of AI in insurance!