 Live from downtown San Francisco. It's theCUBE. Covering IBM Chief Data Officer Strategy Summit 2018. Brought to you by IBM. Welcome back to the IBM Chief Data Officer Strategy Summit in San Francisco. We're here at the Park 55. My name is Dave Vellante and you're watching theCUBE, the leader in live tech coverage. Hashtag IBM CDO Seth Dobern is here. He's the Chief Data Officer for IBM Analytics. Seth, good to see you again. Good to see you again. Anytime, CUBE alum. Thanks for coming back on. Asim Tuwari. Tuwari? Tuwari, sorry. Tuwari, yes. Asim Tuwari, I can't read my own writing. Head of Data Science and Advanced Analytics at Verizon. In from Jersey to East Coast Boys, three East Coast Boys. Three East Coast Boys. Yeah. Welcome gentlemen. Thank you. So, Asim, you guys had a panel earlier today. Let's start with you. What's your role? I mean, we talk you're the de facto Chief Data Officer at Verizon. Yes, I'm responsible for, you know, all the data ingestion platform, big data, and the data science for Verizon, for wireless, wireline, and the enterprise businesses. And it's a relatively new role at Verizon. You were saying previously you were a CDO in a financial services organization. Common that a financial service organization would have a Chief Data Officer. How did the role come about at Verizon? Are you Verizon's first CDO or? I was actually brought in to really pull together the analytics and data across the enterprise because there was a realization that data only creates value when you're able to get it from all the different sources. So we had separate teams in the past. My role was to bring it all together to have a common platform, common data science team to drive revenue across the businesses. So, Seth, this is a big challenge, obviously. We heard Caitlin this morning talking about the organizational challenges. You got data in silos. Indipal and your team are basically, I call it dog-fooding. Champaigning. Champaigning, yeah, okay. But you have a similar challenge. You get a big company complex. A lot of data silos come in. Yeah, I mean, IBM is really, think of it as five companies, right? I mean, and any one of them would be a Fortune 500 company in and of themselves. And so even within each of those, there were silos, and then Indipal trying to bring them across, the data from across all of them is really challenging. And honestly, the technology part, the bringing it together is the easy part. It's a cultural change that goes along with it. That's really, really hard to get people to think about. It is, you know, IBM's or Verizon's data and not their data. And that's really how you start getting value from it. So that's a cultural challenge you face. Is, okay, I've got my data. I don't want to share. And how do you address that? Absolutely, you know, governance and ownership of data, having clear roles and responsibilities, ensuring there's this culture where people realize that data is an asset of the firm. It is not your data or my data. It is firm's data and the value you create for the businesses from that data. So it is a transformation. It's changing the people, culture aspect. So there's a lot of education. You know, you have to be an evangelist where multiple hats to show people the value, why they should do. And obviously, I had an advantage because coming in, Verizon management was completely sold to the idea that the data has to be managed as an enterprise asset. So business was ready and willing to own data as an enterprise asset. And so it was relatively easier. However, it was a journey to try to get everyone on the same page in terms of ensuring that it wasn't a solid mentality. This was the enterprise asset that we need to manage together. A lot of organizations tell me that, first of all, you got to have top-down buy-in. Clearly, you had that. But a lot of times I hear that the C-suite says, okay, we're going to do this, but the middle management is sort of, they got a P&L, they've got to make their plan, and it takes them longer to catch up. Did you face that challenge? And how are you addressing it? Absolutely. What we had to do was really make sure that we were not trying to boil the ocean, that we were trying to show the value. So we found champion, for example, finance, was a good champion for us where we used the data analytics to really actually launch some very critical initiatives for the firm, asset-backed securities, for the first time Verizon launched ABS, and we actually enabled that. So that created the momentum, if you will, is to, okay, there's value in this. And that then created the opportunity for all the other business to jump on and start leveraging data, and then be all willing to help and be part of the journey. So Seth, before you joined IBM, obviously the company was embarking on this cognitive journey, you know, Watson, the evolution of Watson, the kind of betting a lot on cognitive. But internally, you must have said, well, if we're going to market this externally, we better become a cognitive enterprise. So one of the questions that came up on the panel was what is a cognitive enterprise? So you guys, have you defined it and love to ask a sim the same question? Yeah, so I mean, a cognitive enterprise is really about an enterprise that uses data and analytics and cognition to run their business, right? And it's, but you can't just jump to being a cognitive enterprise, right? It's a journey or a ladder, right? Where you got to get that foundational data in order, then you got to start even being able to do basic analytics. Then you can start doing things like machine learning and deep learning, and then you can get into cognition. It's not a just jump to the top of the ladder because there's just a lot of work that's required to do it. And you can do that within a business unit. The whole company doesn't need to get there. And in fact, you'll see within a company, different parts of the company will be at different stages, kind of to Sim's point about partnering with finance. And that's my experience, both at IBM and before I joined. You find a partner that's going to be a champion for you. You make them immensely successful and everyone else will follow because of shame, because they don't want to be outcompeted by their peers. So similar definition of a cognitive enterprise? Absolutely, in fact, what I would say is cognitive is a spectrum, right? Where most companies are at the low end of the spectrum where using data for decision making, but those are reports, BI reports and stuff like that. But as you evolve to become smarter and more AI machine learning, that's when you get into predictive, where you're using the data to predict what might happen based on prior historical information. And then that evolution goes all the way to being prescriptive, where you're not only looking back and being able to predict, but you're actually able to recommend action that you want to take. Obviously, with the human involvement, because governance is an important aspect to all of this, right? But so completely agree that the cognitive is really covering the spectrum of prescriptive, predictive, and using data for all your decision making. And this actually gets into a good point, right? I mean, I think, you know, as Sim has implemented some deep learning models at Verizon, but you really need to think about what's the right technology or the right use case for that. So there's some use cases where descriptive analytics is the right answer, right? There's no reason to apply machine learning or deep learning. You just need to put that in front of someone. Then there are use cases where you do want deep learning, either because the problem is so complex or because the accuracy needs to be there. And I go into a lot of companies to talk to senior executives and they're like, we want to do deep learning. And you ask them what the use case is and you're like, really, that's rules, right? It gets back to Occam's razor, right? The simplest solution is always the answer. It was always the best answer. And so really understanding from your perspective, you know, having done this at a couple of companies now, kind of when do you know when to use deep learning versus machine learning versus just basic statistics? Well, how about that? How do you parse that? Absolutely. So, you know, like anything else, it's very important to understand what problem you're trying to solve. When you have a hammer, everything looks like a nail and deep learning might be one of those hammers. So, you know, what we do is make sure that any problem that requires explainability, interpretability, you cannot use deep learning because you cannot explain. When you're using deep learning, it's a multi-learn neural network algorithm. You cannot explain why the outcome was what it was. So for that, you have to use more, you know, simpler algorithms like decision tree, like regression, classification. And by the way, 70 to 80% of the problem that you have in the company can be solved by those algorithms. You don't always use deep learning. But deep learning is, you know, a great use case and algorithm to use when you're solving complex problems. So, for example, you know, when you're looking at doing friction analysis as to, you know, customer journey path analysis, that tends to be very noisy. You know, you have billions of data points, you know, that you have to go through for an algorithm. That is, you know, a good fit for deep learning. So, we're using that today. But, you know, those are narrower set of use cases where it is required. So, it's important to understand what problem you're trying to solve and where you want to use deep learning. And to use deep learning, you need a lot of labeled data, right? And that's often labeled data. So, and that's often a hurdle to companies using deep learning, even when they have a legitimate deep learning use cases, just the massive amount of labeled data you need for that use case. As well as scale, right? Because, you know, the whole idea is that, you know, when you have massive amounts of data with a lot of different variables, you need deep learning to be able to make that decision. That means you've got to have scale and real-time capability within the platform that has the elasticity and compute to be able to crunch all that data. You know, initially, when we start on this journey, our infrastructure was not able to handle that. You know, we had a lot of failures. And so obviously, you know, we had to enhance our infrastructure. And you spoke to Samit Gupta and Ed earlier about, you know, GPUs and flash storage and the need for those types of things to do these complex, you know, deep learning problems. We struggled with that even inside IBM when we first started building this platform is how do we get the best performance of ingesting the data, getting it labeled and putting it into these models, these deep learning models and some of the instances we use that. Yeah, my takeaway is that that infrastructure for AI has to be flexible. You've got to be great granularity. It's got to not only be elastic, but it's got to be, sometimes we call it plastic. It's got to sometimes retain its form, right? And then when you bring in some new unknown workload, you've got to be able to adjust it and without ripping down the entire infrastructure, you have to purpose-built a whole next set of infrastructure, which is kind of how we built IT over the years. And I think Dave, too, when you and I first spoke four or five years ago, it was all about commodity hardware. Right, it was going to Hadoop ecosystem, minimizing, you know, getting onto commodity hardware. And now you're seeing a shift away from commodity hardware in some instances towards specialized hardware because you need it for these use cases. And so we're kind of making that, we shifted to one extreme and now we're kind of shifting and I think we're going to get to a good equilibrium where it's a balance of commodity and specialized hardware for big data as much as I hate that word and advanced analytics. Well, yeah, even your cloud guys, all the big cloud guys that used to, you know, five, six years ago, it's all commodity stuff and now it's a lot of custom because they're solving problems that they can't solve with the commodity. I want to ask you guys about this notion of digital business. To us, the difference between a business and a digital business is how you use data. And so as you become a digital business, which is essentially what you're doing with cognitive and AI, historically you might have been organized around, I don't know, your network. And certainly you've got human skills that are involved and your customers. I mean, IBM in your case, it's your products, your services, your portfolio, your clients. Increasingly, you're organizing around your data, aren't you? Which brings back to cultural change, but what about the data model? I presume you're trying to get to a data model where the customer service and the sales and the marketing aren't separate entities. I don't have to deal with them when I talk to Verizon. I deal with just Verizon, right? And that's not easy when the data is on sale. So how are you dealing with that challenge? So customer is at the center of the business model. Our motto and our goal is to provide the best products to the customers, but even more important, provide the best experience. It is all about the customer, agnostic of the channel, which channel the customer is interacting with. The customer, for the customer is one Verizon. So the way we are organizing our data platform is, first of all, breaking all the silos. We need to have data from all interactions of the customer that is all digital, that's coming through, and in creating one unified model, essentially, that essentially teaches all the journeys and all the information about the customer, their events, their behavior, their propensities and stuff like that. And then that information using algorithms like predictive prescriptive and all of that make it available in all channels of engagement. So essentially, you have common intelligence that is made available across all channels. So whether the customer goes to a point of sale in a retail store or calls, call center, talks rep, or is on the digital channel, is the same intelligence driving the experience. And whether a customer is trying to buy a phone or has an issue with a service-related aspect of it, and that's the key, which is centralized intelligence from Common Data Lake, and then deliver a seamless experience across all channels for that customer. Independent of where I bought that phone, for example. Exactly, maintaining the context is critical. If you went to the store and you're looking for a phone and you didn't find what you're looking for, you want to do some research, if you go to the digital channel, you should be able to have a seamless experience where you should know that you went and you're looking for the phone. Or you call the care and you're asked the agent about something. So having that context be transferred across channels be available so that customer feels that we know who the customer is and provide them with a good experience is the key. We have limited time, but I want to talk about skills. It's hard to come by. We talk about that. It's number five on Interpol, so the list of things you got to do as a CDO. Sometimes you can do M&A by the weather company. You get a lot of skills, but that's not always so practical. How have you been dealing with the skills gap? Look, skill is hard to find. Data scientists are hard to find. The way we envisioning our talent management is two things we need to take care of. One, we need solid big data engineers because having a solid platform that has real-time streaming capability is very critical. Second, data scientists. It's hard to get. However, our plan is to really take the domain experts who really understand the business, who understand the business process and the data and give them the tools, automation tools for data science that essentially will put it in a box for them in terms of which algorithm to use and enable them to create more value. So while we will continue to hire specialized data scientists who are going to work on much more of the complex problems, the scale will come from empowering and enabling the domain experts with data science capabilities that automates choosing model development and algorithm. And presumably grooming people in-house. Grooming people in-house, and I actually break it down a little more granular. I even say there's data engineers, there's machine learning engineers, there's optimization engineers, and there's data journalists. They're the ones that tell the story. And I think we were talking earlier, I've seen about, it's not just PhDs, right? You're not just looking for PhDs to fill these roles anymore. You're looking for people with master's degrees, and even in some cases, bachelor's degrees, and with IBM's new collar job initiative, we're even bringing on some, what we call P-TECH students, which are five-year high school students, and we're building a data science program for them. We're building apprenticeships, which is, you've had a couple of years of college, building a data science program, and people look at me like, I'm crazy when I say that, but the bulk of the work of a data science program, of executing data science, is not implementing machine learning models, it's engineering features, it's cleaning data. With basic Python skills, this is something that you can very easily teach these people to do, and then under the supervision of a principal data scientist, or someone with a PhD or a master's degree, they can start learning how to implement models, but they can start contributing right away with just some basic Python skills. And then five, seven years in, they're domain experts. All right, guys, got to jump, but thanks very much, Simp, for coming on and sharing your story. Is that always a pleasure? Yep, good to see you again, Dave. Thank you, Dave. You're welcome. Keep it right there, buddy. We'll be back with our next guest. This is theCUBE live from IBM's CDO strategy summit in San Francisco. We'll be right back.