How AT&T’s AI tool is fast-tracking productivity for 100,000 employees
AT&T’s AI tool generates the equivalent of about 1 billion words a day.
AT&T, the telecommunications and technology giant headquartered in Dallas, serves more than 100 million individuals and 2.5 million business customers in the US.
AT&T offers individuals phone plans, internet services, and more. Business products and services include wireless, internet, cybersecurity, networking, cloud connectivity, and Internet of Things, among others.
Situation analysis: What problem was the company trying to solve?
AT&T already uses traditional artificial intelligence across its organization. The company employs the technology to recognize and block fraud to protect the organization and its customers and to optimize field technicians’ driving routes so they can handle more repairs with less fuel.
Recently, AT&T expanded its AI initiatives to generative AI and large language models.
“Generative AI was that next-level jump,” Andy Markus, AT&T’s chief data officer, told B-17. “We saw the potential of generative AI at the enterprise scale to really drive value for our company and our customers.”
In 2023, the company launched Ask AT&T — an internal tool using gen AI and LLMs — to help software developers write and refine code at a faster pace. The company has been expanding its capabilities, and customer-service teams have begun using the tool to summarize phone calls and documents.
Andy Markus is AT&T’s chief data officer.
Key staff and partners
Markus said AT&T worked with Microsoft to develop Ask AT&T; the tool was originally built with OpenAI’s ChatGPT. AT&T’s legal, privacy, and compliance teams, as well as company executives, were also involved. The initiative was reviewed internally to ensure it offered a business case that would create value for the company and customers.
AI in action
A function of Ask AT&T is AskData, a feature that enables employees to query large databases using plain language. For example, managers in charge of specific geographic areas could ask, “Which stores in this region had the biggest increase in foot traffic last week?”
The tool then automatically translates the request into SQL code, the industry-standard programming language. AT&T said this process used to be done manually, as developers previously had to write their own code and create an algorithm.
Markus said more than 100,000 employees have access to Ask AT&T. The tool generates the equivalent of about 1 billion words a day, which includes automated summaries of customer calls and responses to queries.
“It makes their jobs easier so they can serve our customers better,” he said.
Markus said the tool also helps automate the software development process, including creating code, testing, integrating with other systems, and executing the programs. “Humans are always in the loop, but it’s really helping there.”
Ask AT&T is being used in other areas of the company, including the supply-chain and contracts teams. For example, the tool can provide insights into when the company is owed credits or rebates and when contracts with suppliers are up for renewal.
Did it work, and how did leaders know?
Markus said Ask AT&T has reduced the amount of time it takes to develop software by about 10% to 30%, depending on the individual task. The tool has also saved customer-service agents several minutes per call, he added.
“I have people routinely talk about how it’s helping them get through their day in a more efficient way,” Markus said. By improving productivity and efficiency, he said the gen AI tool allows AT&T to invest money it used to spend on hiring external support on new technology products and services.
Markus told B-17 that AskData executes high-level accuracy when answering complex questions. It ranks highly on the GenAI BIRD leaderboard, which measures the accuracy of generative AI platforms worldwide in translating plain text into SQL.
What’s next?
AT&T continues to add use cases for Ask AT&T, including training the tool to answer human resources questions and analyze complex financial materials. The company is also working on new capabilities, such as using real-time voice to query the tool and creating digital avatars to improve the platform’s user experience.
“The beauty is all parts of the company have a use case where we can apply this functionality and help them out,” Markus said.
