At AT&T, AI is becoming part of “core fabric,” says chief data officer

One year ago, AT&T, the world’s largest telecom company by revenue, announced a collaboration with AI cloud company H2O to jointly launch an artificial intelligence (AI) feature store for enterprises. This paid software platform enables data scientists, developers and engineers to discover, share and reuse machine learning (ML) features to speed up their AI project deployments.

Since then, the feature store has become a key part of AT&T’s vision of scaling its own AI efforts across the organization, and to “truly integrate data and AI into the core fabric of how we run the business,” Andy Markus, AT&T’s chief data officer, told VentureBeat. 

Markus, who joined AT&T in February 2020 after nearly two decades in roles at media companies such as Turner and Warner Media, said the company carries more than 543.7 petabytes of data across its global network. To manage AT&T’s data and AI at this level of scale, it has defined a common approach on how data is stored, managed, accessed and shared across the company. 

AT&T’s “North Star” for data and AI

The company relies on its Chief Data Office (CDO), he said, to set AT&T’s “North Star for data, analytics, and AI excellence.” Its mission is to harness, share and catalyze insights from the company’s massive data stores and to transform and modernize AT&T’s data platforms, data supply chain and data science ecosystem. 

Along with the feature store, the CDO uses a centralized data intelligence platform that provides a “single version of truth” for each defined data product, which empowers business managers as well as data scientists with self-serve access to datasets. 

“There’s been a really big focus to standardize on best-in-class tools that are cloud based,” he said. “We’re using state-of-the-art technologies like H2O, Good AI, Databricks and Snowflake to deliver value to the customer and to our data science community.” 

Responsible AI is also a big focus, he added. AT&T created a technology called SIFT, the System for the Integration of Fairness and Transparency in AI, and rolled it out across the company, so that all models get evaluated for potential bias. “The process not only detects the bias, but it will take the user through mitigation steps,” he said. 

Modernizing AT&T’s data and AI stack

Several years ago, AT&T’s data science efforts were “a bit of the Wild West,” said Markus.

“We had a lot of different types of technologies, a fairly disparate ecosystem of data scientists,” he explained. “Now we have a very connected data science community, everybody is working consistently with the same cutting-edge tools, and that we’re really maximizing the reuse of our data at our feature store.” 

Not unlike many legacy companies, AT&T also has had to deal with a lot of legacy tech debt, he added. 

“We had pockets of greatness, some really smart people are doing some really good things, but with a non-standard technology, there were results that weren’t shared, features that weren’t reusable,” he explained. 

Over the past two years, the company has been modernizing to a cloud-native elastic technology. This last year included an evaluation of the state of AI across AT&T, he added, which found that AI efforts have delivered billions of dollars of value on an annual basis to the company – everything from revenue enhancement to cost savings and efficiency processing. 

“We’ve brought the competency of the company up several notches by not only working with those groups that are already doing well, but bringing other parts of the company up so that they can really come to the table and leverage the great ML and AI functionality that we’ve created,” Markus said. 

AI-driven decisioning is table stakes in telecom

These days, AI-driven automation and decisioning has become table stakes for running an efficient business in the complex and cost intensive operating world of telecom – powering everything from optimized network planning, customer care and field services to protecting customers and networks.  

“The pace is continually accelerating as technology becomes more proficient at solving complex problems at the scale of AT&T and the demands of the business and our customers increase,” he said. 

While the table stakes use cases may be solved, he added, the company is now focused on next-generation challenges that continue to build on the value already created with AI.  

“Tackling the more complicated issues, both from an AI and business operations perspective, come with a steeper curve, such as developing AI-driven products and services and creating self-healing 5G networks,” he said. 

Democratizing the ability to create AI

In addition to the CDO’s feature store and centralized data platform, Markus explained that AT&T is working to democratize the ability to create AI. 

“We have a standard code-driven process for creating AI, built for the data science community,” he said. “Now we’re working to make that low-code, no-code, so that we can really democratize the ability to create AI not just to the data science community but other subject matter experts across the company.”  

If AT&T’s number one goal is to embed AI in the “core fabric” of how the company runs its business, Markus said the second is to expand on the functionality of AI-as-a-service. 

“We want to take that code-driven process and continue to advance it to what we call the citizen data scientist,” he said. “Those are the subject matter experts in the business that can create AI for their use cases, using responsible and ethical AI, and really drive more value for the company.” 

AT&T uses AI to solve business problems

That goes to what Markus said is one of his team’s core tenets – understanding the business problem, and then getting the right data in place. 

“We’re using technology to solve business problems,” he said. “We’re not doing technology for the sake of technology – so it all starts with understanding what the issue is, how that creates a challenge for the business, so it almost starts in a consultative fashion.” 

In a recent blog post, Markus highlighted several of AT&T’s most powerful AI use cases. They include using predictive AI models to avoid network outages by powering an end-to-end incident management platform that scans more than 52 million different network records, devices and customer circuits, and over 1.2 trillion daily network alerts. 

Another AI-driven solution using sampling, predictive modeling and multivariate analysis blocks nuisance robocalls by filtering through billions of daily records looking for patterns and suspicious qualities. 

And an AI-based fraud management tool evaluates millions of daily transactions, inspecting events within milliseconds against hundreds of rules – and includes an interface that allows front-line fraud team members to write, test and deploy rules themselves. 

Standardizing skill sets on data science teams

When it comes to building successful teams, Markus said the first thing he did when he arrived at AT&T was to create a standardized definition of what a data scientist is. 

“Things blur over time and we just weren’t really consistent,” he said. 

In addition, data scientists often work and collaborate with the business. “Now that we’re using common technology and data like the AI feature store, we can democratize that,” he said. That means people that are subject matter experts in fraud or network or customer care have a connection to the data science community in that part of the business. 

“I would almost call it a federated way to organize, in a very connected fashion,” he said. “So we’re not duplicating work, we’re not duplicating data, we’re working together to solve more problems.” 

The last big push toward data modernization

Markus said he thinks about his role in terms of wearing different hats. He wears three hats at AT&T, he explained – and he’s trying to get rid of all of them. 

One hat is about making sure that data and AI are used to deliver meaningful and significant value to AT&T. The second is about ensuring that data and AI are first-class assets of the company. And the third is modernizing the company’s data and AI ecosystem. 

“2023 is the last big push towards that,” he said. “We’ll have the bulk of our technology in a new modern environment with an updated set of tools.” 

And then, he said, “Hopefully we can retire that hat.”