October 13, 2022
Andy Markus explains how making AI accessible to business users creates value
As one of the largest telecom companies in the world, AT&T sees petabytes of data flow through its networks every day. Chief Data Officer Andy Markus leverages AI not only to manage and secure that data, but also develop tools to empower employees across the company to do their jobs better.
AI Business recently sat down with Markus to talk about his endeavors at AT&T. What follows is an edited transcript of that conversation. To listen to the podcast, click on the player below.
AI Business: AT&T manages petabytes of data on its global network. What kind of approaches do you take to manage it and how does AI play a critical role?
Andy Markus: We do see a ton of data. The most recent stats are that we see about 530 petabytes a day across the network. Not all of the data is curated; all of the data is stored in raw form. What we try to do to make the data most useful is we do try to curate it – and create views and aggregations that the business most needs, and then we attempt to democratize those across the business.
We create data libraries by types of data. Those data libraries are curated and we eventually create business-ready datasets. That's where most of the company leverages its data. Only the data science applications in some of the network operations areas are leveraging the 500 petabytes of data. We try to curate it to make the data into the most usable fashion for most of our business users.
All that data we inventory in a central data repository we call amp. An amp is a searchable and findable environment that allows us to take this data, allow people to find it, and to democratize it – and to really maximize the reuse of that data.
One of the challenges that all companies have is that people will take copies of data, they'll make more copies of data, and now you have data duplication all over the place. So by documenting it, making it searchable and findable, and then easily accessible, we're trying to solve that issue at AT&T.
AI Business: Talking about democratization of data, that means the tech teams and data teams have to work cross-functionally with business teams. So how do you accomplish that? Usually, those two groups speak different languages.
Markus: I have three hats to wear at AT&T. The first one is, we have to modernize and transform our data and our ecosystems. That's a full time job right now; that job will be over at some point. The next hat I wear is to really execute – in partnership with the business (side) − data and AI use cases. And then the third hat is that of the chief data officer, to make sure that our data and AI are first-class citizens, optimized assets of the corporation.
With these last two hats, we're working with the business (side) all the time. We're at the table with them, helping them to do really complicated things with data in AI, making sure that we're delivering the value that is needed for business. And then from a data governance perspective, with the chief data officer hat, we're there ensuring that everybody is at the table, maximizing their own skillsets around data competency, so that we really can become a data-driven company.
… We actually have a business engagement function within the chief data office. The goal of that function is to be like internal data consultants where we sit at the table with consumers of our data and people across the firm, so that we're there to help them solve their business problems − whether from a Center of Excellence standpoint, consulting at a high level, or to really roll up our sleeves and do work with them to execute their use case. That function is super-important for making us successful. … We’re not doing technology for technology's sake. We're doing technology, data and AI to drive value to the business.
AI Business: Can you give us some examples of successful collaborations between the data side and the business side?
Markus: We'll work with the business on network planning, or how to leverage data to optimize (say) how we pay taxes, or how we look at the fraud. We work together to create solutions. Everything from marketing applications to applications to identify incidents in the network, using data and AI to optimize how we roll trucks, how we dispatch our field services reps − we've got so many applications that are core to how AT&T operates.
AI Business: What are some of the challenges that you've encountered and solved while you're implementing this?
Markus: Making sure we all talk the same language is really important. Ultimately, communication, transparency, and a sense of partnership are key. … We have to have transparency so that the business understands what we're doing and we understand what the business is doing, so that we don't collide or swim (different) lanes; we're working together. And then, you know, we can resolve any overlap, if there are, if there happens to be overlap.
Another part of being a good partner in making sure you have this good connection is really listening. … If you think across AT&T, some of our business units are really advanced in their data competency. And so when we engage, it's really with a light touch. It’s more about ensuring we have the right architecture involved, that we've had the data documented correctly. That business unit doesn’t need a lot of help in executing their use case. In other cases, the business (units) aren't quite as mature and so there's more of a hands on role. It really depends on the use case and the part of the business we're working with.
Unlocking business value
AI Business: In one of your blogs, you've written about citizen data scientists at AT&T. Who are they and why are they important?
Markus: They are everybody across the business. Let me set the stage for that: We have to have the technology that allows citizen data scientists to be engaged. A code-driven process is already pretty standard across AT&T for developing AI. But what we're working on is creating that to be low-code, no-code, so that we can engage smart people across the company who can leverage this technology, (giving them the) knowledge to develop AI solutions that are important for their part of the business. You don't have to have a professional data scientist for every application; you're using this technology with your subject matter expertise to create that functionality (you need).
… We've done a lot with AI across the company. We just recently chalked up close to $2.5 billion to $3 billion of value just in the last 12 months at AT&T with AI. And there's still so much more we can do. By leveraging this technology that anybody can use to create robust, but responsible AI, we can even create a lot more value − maybe at a magnitude of 2x to 3x, or 5x or more.
And the reason why that's so important is professional data scientists, good ones, are hard to find. They're like unicorns. … So our path forward to create more productive AI across AT&T is to enable that (low-code, no-code) technology. Smart people across the business can do it, and do it well.
AI Business: That's brilliant, giving non-technical people enough technical chops so they can do their jobs best in their domain. But first of all, you've got to get buy in from them. How do you do that?
Markus: It's a good question. But I will say that it’s not that hard. AT&T, at its core, is a pretty technical company. We have a lot of smart people that are really receptive to this concept. They're ready to engage.
The way we approach it is (to say), ‘we can't do things today the way we've done things in the past 20 years. We have to break the mold, we have to change the status quo, we have to attack things in a new way to really be that connectivity company.’ And so I think people are ready to go on that journey. They just need the technology and the support to get there.
AI Business: What advice can you give other companies that want to embark on a similar data journey as AT&T?
Markus: We have to have the tools in place to make the technology they need to use available. So I mentioned AI as a service as that foundational platform that we're creating. That's a big part of it, but you also have to have that empowering technology. At AT&T, so many of our business users, data users, are actually BI or business intelligence users. What we do is plug them into the right technology, like a Power BI or Tableau. They don't necessarily need an AI as a service, but we need to make sure that they have a toolset that's very easy to use but powerful in its own right. And then just continue to find other ways to create self-service opportunities for employees.
We have another concept that we've created called data-platforms-as-a-service, which is a data-pipelining self-service tool. It allows the business to bring in and democratize data without having a data engineer involved. So it's about enabling and providing the tools that the business needs, to do what users want to do and making sure … we're all doing things in a coordinated and connected fashion.
AI Business: Can you differentiate between business intelligence, and AI powered intelligence or insights?
Markus: I would say maybe it's different, not necessarily just better. For most BI output applications, it's more around understanding what the data is telling you often in a visual way, sometimes in a numerical way, but just getting the insights, or the story from the data. With AI, so much is about the predictive nature – understanding the data that we have today, how it's going to predict actions and outcomes in the future. And so, it's taking a use case with a different light.
AI Business: When encouraging employees to do-it-yourself with AI, do you build or do you buy?
Markus: Well, both. Another thing I tell the team … is don't try to build a commodity. If it's a commodity, we're going to buy it. When I talk about our tool sets, really what we have as a best-in-class, commercial offering, (are ones) wrapped in a common platform with some AT&T special sauce.
I'll give you a really good example. With AI as a service, the foundation (consists of) commercial tools like Databricks, or H2O. But in the middle of that is what we call the AT&T real-time feature store for AI. We built the feature store, actually, in partnership with and licensed from H2O … because the functionality that we needed at our scale or complexity didn't exist in the market.
So it's a combination of build and buy. We don't build where we don't need to. But when we do build it, I think it's really powerful. The AI feature store has, at this point, almost 20,000 of AT&T’s best AI features across the company. It allows users across AT&T to dip into that feature store, and immediately start building their models without having to go through the standard steps that most data scientists have to go through because those features are already there. So the model development process is greatly shortened and the features are there to superpower those models.
AI Business: How do you ensure privacy and security of your data?
Markus: Security and privacy are top of mind with everything that we do. Those are my core concerns as the chief data officer and we work really closely with our chief security officer and our chief privacy officer (to address them).
At AT&T, all of our data is tightly governed by both policy and user access by process. So that you can't access data unless you go through a very controlled process to ensure that the access is needed and granted. And then all of our use cases are reviewed with privacy (in mind). We tackle our modeling and our data applications, almost in a privacy-by-design way, where we deal with privacy issues up front. So we're not scrambling on the back end when we we’re trying to apply the application.
Another thing that's really important to us, and we spent a lot of time on it, is responsible AI. … We have our AT&T AI operating guidelines. For all of our AI, we envision going through our responsible AI process so that it's documented; it goes through a process so that bias can be evaluated and corrected. … With the sensitive data that we have, with the trust we have from our customers, there's nothing more important to them than to protect their data and their privacy.
Is AI sentient?
AI Business: What's next for the chief data officer? What are you working on?
Markus: We're almost two thirds of the way through a transformation to the cloud. So we've got to finish that … and get the bulk of our workloads to the cloud. That's one thing. The other is, I talked a lot about the AI-as-a-service functionality. There's still work to be done there. For an example, the responsible AI component needs to be automated.
And then we're working on new things to add on to AI-as-a-service, like hyperlocal analytics, and some more NLP functionality, as well as more personalization functionality. And then we have a lot of great use cases for graph technology. One of the really great things that we've done recently (is on robocalling.) Robocalling is a huge issue for telecom. We have suppressed a ton of robocall traffic − nuisance traffic − using standard data NAT (Network Address Translation) techniques. But we've also employed graph technology to see not just the number coming through, but the relationship of that number to other numbers in their connected network. … And there are a lot more applications of graph technology that we have across AT&T.
AI Business: We're going to do a fun lightning round. I'm going to give you a word or a phrase and then you tell me what comes to mind: AI is sentient.
Markus: Not yet.
AI Business: Will it ever be sentient?
Markus: Not in my lifetime.
AI Business: Text-to-image generator
AI Business: Crypto/blockchain
Markus: Two very different topics. Blockchain is very important for certain applications. Crypto, wave of the future.
AI Business: NFTs
AI Business: Explainable AI
Markus: Foundational. As we think about our commitment to responsible AI, AI has to be explainable. We have to be able to explain it to our customers.