How AT&T trained its AI assistant—chief data officer on how the tech is being used

AT&T’s internal AI assistant Ask AT&T has a wide array of functions from guiding employees through complex problems and assisting marketers to, perhaps, some less obvious functionalities, such as helping with personal home renovation tips. That last item was added for Andy Markus, AT&T’s chief data officer, who helped develop the Ask AT&T platform—and needed some home-improvement advice. “I ask it things all the time,” Markus told Ad Age in a recent interview. “I ask it things about work. I ask it things about my personal life, just things that I’m dealing with. I could be having some construction done in our house right now, some remodeling work. And, so, I ask a lot of contractor-type questions. But from a work perspective, some of the complex business cases that we’re doing, we can ask Ask AT&T to help.” AT&T is one of the 20 largest advertisers in the U.S., spending more than $2.46 billion in 2023, according to the Ad Age Datacenter. The Dallas-based telecoms marketer has adopted generative AI like so many brands, working with large language models from OpenAI, Google and Meta along with other tools to develop a bespoke platform. The platform coalesced into an agent called Ask AT&T, which is trained on the brand’s vast reserves of data to instill deep knowledge about the company. Beyond embracing AI in the office, AT&T and other carriers sell Apple’s iPhones, the latest mobile offering to come out with new AI features for consumers. On Tuesday, Apple launched AI, or what it calls Apple Intelligence, on iPhones, an addition that affects how AT&T markets phones and manages data across its network. Read about Apple’s new ad campaign for Apple AI With all the changes going on, Ad Age spoke with Markus to understand the influence of AI across the business. Check out the latest Ad Age events and award programs How is AT&T preparing for Apple launching AI on iPhones? Our focus is on connectivity, ensuring that all those devices that need access to information have the bandwidth and the capacity that they need to operate well, and we’re building our network to serve that connectivity. How will Apple’s AI release affect consumers? It’s going to become a true productivity enhancer for all types of consumers. It’s going to be so embedded that I don’t think we can imagine a world without it. And so Apple is at the very beginning of that journey. The services that we’re creating, in an adjacent swim lane to Apple, are at the beginning of that journey. But if you think six to 12 months out, our lives are very different than they were before, in a good way. How has AT&T embraced generative AI? Our generative AI journey has affected everything: how we operate the network, how we are using it to protect our customers and our company from fraud, how we’re communicating to our customers, how we’re doing back-office work. What third-party AI platforms does AT&T use? We couldn’t hitch our wagon to any one technology. We had to be able to change as the technology adapted. We are very interoperable. We use technology from Open AI. We use Google. We use a lot of open-source technology. A lot of the stuff that we’re doing right now is with the Llama family of models from Meta. What has AT&T developed with generative AI? We built a system called Ask AT&T (an internal AI assistant) and it has a front end for the average employee at AT&T. We really integrated it at the back end of the screens and the systems that run our company. We had to really mix and match AI technologies, or else we would be paying way too much, more than we ever budgeted. We’re paying, truly, like a percent of what we would be paying if we used an off-the-shelf model. Check out the latest Ad Age events and award programs How did you train Ask AT&T? We’ve made it really smart in AT&T terms. It understands our company. And when we talk about accuracy, which is a big thing that we use to see how we’re doing, the accuracy is off the charts, because we’ve made it smart on how AT&T works. If you took a model off the shelf, it doesn’t matter if it’s the best model from Open AI, the best model from Google. It doesn’t matter. It’s only going to know so much about AT&T, and how we work and how we operate. But by making it smart, now we’re able to get accuracy rates that are human-level accuracy rates. OK, AI is running throughout your organization—how about within marketing? We’re definitely using AI throughout the marketing process. There are traditional AIs embedded throughout how we market. There is also work around customer journeys and how we think about journeys in the future will be even more AI-enhanced. If we use generative AI to help us with the creative for a campaign, we still will have a human verify that we have the right information, that it’s appropriate for the consumer, for the given campaign, for the given use case. (An AT&T representative added that AI also factors into the company’s programmatic and digital advertising strategy. “AI-driven audience discovery and performance analysis tools have made it easier for marketers to generate insights and evaluate campaign effectiveness using natural language queries,” the representative said by email. “AT&T is also testing and exploring how text and image generation could accelerate campaigns and enable tailored messaging for different audience segments, languages and formats.”) What are you excited about developing through AI? We’re spending a lot of time on agents, breaking down a bigger problem into smaller problems, and creating a certain set of skills for these agents that are reusable across the company. We think there’s a lot of benefit there in solving really complex use cases that people thought were unsolvable with. What AI use case has been effective within AT&T? We spend so much effort and capacity on developing software, and the entire software development life cycle at AT&T is being reimagined with generative AI at the center. It’s agent-driven, from the user story [at the start of the coding process] on down, everything else is driven off of generative AI and automation, with humans in the loop, of course. It’s helping increase the efficiency of software processes between 20% and 40%. How did AT&T go about launching AI across the company? This is not just a science project. This is about driving true bottom-line business value. I’m on the technology side of the organization, but we worked with our partners on the business side of the organization, and we created what we call a generative AI transformation office, and we funded that office, and that office works with the CFOs across the company. As we bring use cases to the office, they will run a detailed business case, and then we prioritize the business cases that we have across the firm to ensure that we’re working on the ones that drive the most value for the company. And then that’s what we as technologists work on. The magic of the playbook is that these are not just proof of concept. These are very intentional investments in how we can invest in processes that can change the way AT&T operates that really drops to the bottom line, and we ran that entire play this year. We’re going to run it again next year as well. In a year, we have two times the return on the dollars that we’ve invested, and we expect that to grow.

How AT&T trained its AI assistant—chief data officer on how the tech is being used

Featured Stories