TELECOM
World’s leading telco leverages H2O LLM Studio to finetune models and enhance query generation with reduced costs


Significant
Cost reduction
Higher ROI
Higher
Customization
Optimized performance
Maximized
Accuracy and efficiency
Faster decision making
Overview
In today’s competitive landscape, collecting actionable insights from databases is critical for business units looking to enhance their campaigns and drive growth. Recognizing this need, the CDO Office team at AT&T aimed to create a user-friendly tool that empowered non-technical users to interact with their database. The solution would allow users to use plain English and get results back in numbers, enabling them to uncover valuable insights, such as identifying key individuals in specific regions to target for campaigns and market expansion
Challenge/Solution
The business unit needed a solution that not only identified prospective customers but also analyzed current customers to uncover untapped opportunities—services the company provides but the customers hadn’t purchased yet. Precision was critical, as the insights derived from the solution would drive key decisions, leaving no room for error.
While a large language model provided the required analytical capabilities, its high operational costs posed a challenge. To address this, the team explored how to fine-tune a smaller language model, aiming to achieve comparable performance at a significantly lower cost, ensuring both accuracy and efficiency, using H2O LLM Studio.
Why Small Language Models and H2O LLM Studio?
By leveraging the power of semantic similarity search, data curation, and data profiling, the AT&T CDO team was able to fine-tune a small language model to achieve comparable performance to a large language model like GPT4.
The results show that fine-tuning a small language model with H2O LLM Studio can significantly reduce costs, with the Llama SQL coder 8 billion parameter model being effectively free to use, compared to the two cents per call cost of the previous solution.
Furthermore, fine-tuning small language models with LLM Studio provides a more tailored solution that can be customized to meet the specific needs of a business or organization. By using data curation and data profiling, the model can be trained to understand the terminology and keywords used within a specific domain, leading to more accurate and relevant results.