Case Study: AT&T cuts fraud and AI costs with H2O.ai

AT&T cuts iPhone fraud 80% and LLM costs 90% with H2O.ai

AT&T, a major telecommunications company, sought to democratize AI across its business units and accelerate the delivery of AI-driven solutions, moving beyond its initial reliance on open-source tools. Their challenge was to scale AI capabilities to address issues like iPhone fraud and high operational costs while also finding a cost-effective way to leverage generative AI for call analytics and database queries. They partnered with the vendor H2O.ai to utilize their H2O AI Cloud platform, including products like H2O Driverless AI and H2O LLM Studio.

The solution implemented by H2O.ai involved building a comprehensive AI-as-a-Service platform and creating fine-tuned small language models to replace expensive large ones. This resulted in an 80% reduction in iPhone fraud, saving millions annually on fleet operations, and cutting generative AI costs by up to 90% for call center analytics. Additionally, their fine-tuned text-to-SQL model achieved top global benchmark rankings while reducing the cost per query to nearly zero. H2O.ai enabled AT&T to deploy projects faster and scale AI exponentially across the organization.

AT&T