EXL and NVIDIA collaborate on specialised language model for insurance

Data analytics and digital solutions provider EXL has announced the launch of the EXL Insurance Large Language Model (LLM), the first industry-specific LLM designed to enhance the efficiency and accuracy of insurance-related tasks. This development builds on EXL’s recent collaboration with NVIDIA AI Enterprise, according to a news release.

The EXL Insurance LLM intends to address the insurance sector’s unique challenges, which often arise from the inadequacies of general-purpose LLMs that lack the fine-tuning needed for the industry’s complex processes.

Traditional LLMs have been criticised for their inability to effectively handle tasks, such as claims reconciliation, data extraction, question-answering, anomaly detection, and chronology summarisation. EXL said that this oversight has led to inefficiencies, increased costs, and prolonged settlement times for insurance companies.

According to internal studies conducted by EXL, the new LLM is expected to outperform leading pre-trained models, including OpenAI’s GPT-4 and others, achieving a 30% improvement in accuracy across various insurance tasks.

Key features of the new model

Gartner has predicted that by 2027, over 50% of generative AI models will be tailored to specific industries or business functions, a substantial increase from approximately 1% in 2023.

The news release revealed that the EXL Insurance LLM was developed by EXL AI Labs, and the team utilised the NVIDIA NeMo platform for the model’s training, customisation, and deployment, employing advanced techniques such as low-rank adaptation (LoRA) and supervised fine-tuning (SFT) to optimise its capabilities. The model’s performance was enhanced through rigorous testing in both single and multi-node configurations, maximising the use of NVIDIA’s H100 GPUs.

EXL also incorporated the NVIDIA Triton Inference Server to enhance GPU efficiency and enable robust handling of extensive datasets. The model employs retrieval-augmented generation (RAG) powered by NVIDIA NeMo Retriever microservices, enabling effective processing of long documents for comprehensive question-and-answer capabilities.

“With 25 years of expertise in processing medical records data for bodily injury, workers’ compensation, and general liability claims, EXL has developed curated data sets with domain-specific tagging, labelling, and question and answer pair creation for claims adjudication to fine-tune our models,” said Anand Logani, EXL’s executive vice president and chief digital officer.

“The EXL Insurance LLM offers 30% greater accuracy and 30% lower costs than generic LLMs while ensuring full regulatory compliance.”

The EXL Insurance LLM supports a range of functions, including:

1. Data ingestion: The model aggregates and reconciles extensive data sets, including medical records and claims histories.
2. Contextual classification: It categorises extracted data for core functions like claims adjudication and customer service.
3. Insights and conversations: The model provides actionable insights, facilitating improved negotiations and real-time customer interactions.

Established in 1999 and headquartered in New York, EXL employs over 55,000 professionals worldwide, delivering innovative solutions across industries, including insurance, healthcare, and finance.

What are your thoughts on the role of AI in the industry? Feel free to share your comments below.