EXL unveils insurance LLM for improved claims accuracy

EXL has launched the EXL Insurance LLM, a specialised Large Language Model specifically designed for the insurance industry, aiming to improve the accuracy of claims and underwriting-related tasks.

EXL's new model, which builds on its recent collaboration with NVIDIA AI Enterprise, addresses several challenges the insurance sector faces with general Large Language Models. These generic models often lack fine-tuning with private insurance data and fail to fully understand the intricate business processes involved in claims adjudication.

The company states that the EXL Insurance LLM can significantly enhance the accuracy of industry-specific tasks, helping reduce costs and improving consistency in AI applications specific to the insurance sector. According to research by Gartner, the trend towards industry-specific AI models is expected to grow, with more than 50% of generative AI models anticipated to be industry or function-specific by 2027, up from about 1% today.

Internally, the EXL Insurance LLM demonstrated a 30% improvement in task accuracy over existing pre-trained models such as OpenAI GPT4, Claude, and Gemini. This achievement is credited to EXL AI Labs, which used the full-stack NVIDIA AI platform for its development.

Customisation of the LLM was achieved using the NVIDIA NeMo platform, part of the NVIDIA AI Enterprise Software Platform, which facilitated training, customisation, and deployment. Techniques such as low-rank adaptation and supervised fine-tuning were employed to enhance performance, and advanced parallel processing methods optimised the model's performance on H100 GPUs.

EXL also utilised the NVIDIA Triton Inference Server to harness GPU capabilities, incorporating retrieval-augmented generation to manage extensive documentation needs effectively. Input and output management were improved using NVIDIA NeMo Guardrails to ensure robust user interaction.

"LLMs have the potential to transform Insurance but generic models have not been trained on contextual data and nuanced processes," said Anand "Andy" Logani, Executive Vice President and Chief Digital Officer at EXL. "With over 25 years of expertise in leveraging medical record data for claims processing and underwriting Life policies in the Asia Pacific, EXL developed curated datasets to fine-tune a LLM using our Nvidia-powered gen AI platform."

He added, "The EXL Insurance LLM delivers 30% greater accuracy and 30% lower costs than generic LLMs, by not only accelerating claims adjudication [...] but also reducing indemnity spends, while ensuring full regulatory compliance to the new AI laws across markets."

The EXL Insurance LLM supports various tasks, including the ingestion of structured and unstructured data. It can aggregate and reconcile claims-related information such as medical records and handwritten notes. Further, it enables contextual classification and triaging, automatically categorising extracted data for use in core insurance functions.

The model facilitates enhanced conversations and insights from data, providing more accurate negotiations, anomaly assessments, and real-time customer interactions. This development results from the collaborative efforts of EXL AI Labs, with a focused team of specialists aiming to expand the model's utility across the insurance value chain, including underwriting and finance.