Customer service interactions hold a wealth of information that can drive improvements in service quality and customer retention. Every year, AT&T receives 40 million calls from customers seeking assistance with various tasks or reporting issues. While transcription has been automated for some time, the challenge lies in efficiently categorizing these calls to extract actionable insights. Initially, this process involved manually sorting millions of call summaries into 80 distinct categories, an effort aimed at reducing customer churn.

A breakthrough came when AT&T began leveraging advanced artificial intelligence solutions. Initially, they employed ChatGPT to automate the classification of call summaries. Although effective, the cost and reliance on external systems posed significant challenges. To address these concerns, Hien Lam, a senior data scientist at AT&T, collaborated with Ryan Chesler of H2O.ai to develop a more efficient and cost-effective solution. By integrating multiple open-source AI models, they created a flexible system capable of handling varying levels of complexity in call data while ensuring privacy.

The adoption of open-source models represents a significant advancement in how businesses can harness AI technology responsibly and economically. The team's innovative approach not only reduced costs by 65% but also improved processing speed, cutting the time required to analyze daily call summaries from 15 hours to less than five. This progress underscores the potential of open-source collaboration to empower organizations, fostering innovation and efficiency. As technology continues to evolve, embracing such solutions will enable companies to provide better services and strengthen customer relationships through data-driven insights.