Season 1, Episode 5: "Is AI a dirty word?" with Vishal Gupta, CTO @ Unisys
Vishal Gupta: Once you create a policy, let's say, on a particular floor for some type of network, we could almost have AI recommend the right policies for you, for your entire digital enterprise.
Speaker 2: How have you enabled your infrastructure fundamental change over the last five years? Partnering with the business is critical. The tools exist on the cloud. Change at the rate necessary. Secure by design. That's where this [inaudible 00:00: 26 ].
Andrew: Hey, it's Andrew. And welcome to Network Disrupted, where IT leaders talk about navigating the disruption in our industry. In this episode, we talk about what merit analytics, or rather, the dirty words, artificial intelligence, has in the context of digital transformation. My guest today is Vishal Gupta, CTO at Unisys, the hundred year old IT company we all grew up around. Unisys has recently gone through a fairly significant and positive transformation, which is something Vishal has been focused on since he joined in 2018. Vishal also has a really big interest in how talent impacts any business plan. So we spend some time talking about that in this episode. Let me know what your thoughts are. You can tweet me at Network Disrupted, leave a review on Spotify or Apple podcasts, or email me at andrew @ networkdisrupted. com. So Vishal, thank you for joining us. I'm really looking forward to this conversation.
Vishal Gupta: It's a pleasure to be here. Thanks for inviting me.
Andrew: I'm very careful, for instance, at BlueCat in my role, of not leading with analytics or AI in terms of why our products are good, because it's not the fact that we have analytics that make them differentiated, it's what the outcomes are, and what value we can create. And I worry that in the industry, too many companies are leading with, ours is better because we have analytics, or we have AI, or we have ML, or we have whatever, as opposed to it being a consistent strategy across many different things that's driving outcomes.
Vishal Gupta: So I think, firstly, I want to completely agree with you that some of these technologies get hyped so much. We've seen that with blockchain. To some extent, it's with AI as well, which actually can act as a determent, ultimately, to progress because these are means to an end. These are not an end on its own. These are all about driving the right outcome. I spent a lot of years, prior to Unisys, at Symantec. And, at the time, the security industry, every vendor would talk about," Hey, my security is better because I do AI. When everybody is trying to use [ inaudible 00:02:39], I have AI. And then they start talking about the size of the data lake. So it just was a nonsensical journey. Now, having said that, I do believe there is a very real and compelling use case for AI. If we look at the IDC report, it shows the overall spend on AI in cognitive to grow from almost 24 billion in'18 to 77 billion in'22, which is a huge [ inaudible 00:00:03:11] over 37%. So we know there is something real there. The question is, what are those outcomes that could be real? What things can AI do today that actually can create some interesting things. AI is very good at things like classification and comparison. So we use the classification capability in AI to, really, not just discover different types of digital equipment, but categorize what type of equipment it is. So now we know, okay, you've got so many switches, so many routers, so many different types of equipment. And then through comparison, once you create a policy, let's say, on a particular floor for some type of network, we could almost have AI recommend the right policies for you for your entire digital enterprise. That way, you do something once and it reinforces the learning it learns from it and says, just like Amazon can say, based on the books you read, which others you might like, based upon the policies you applied, which other equipment with similar policies apply. The second, I think, very, again, low- hanging fruit in the world of AI is using AI to really improve your customer experience and sales and support process. Because no matter what business you're in, even if you're in a non- profit, you have a certain set of customers who you interact with. For example, these days, as we've seen with natural language processing, whether you leverage it from an Alexa or from a Google, there is a lot of advances. So, in the way that you interact with your customers, you should be able to leverage AI to essentially build out intelligent chatbots, which can both reduce the costs and provide a better experience, to know when to escalate to a human, when to be able to answer the questions that a customer is having or an employee is having, and to be able to essentially get to this Holy grail of both cheaper and better. Usually it's one or the other. So this whole thing around natural language processing, NLP, and mixing that with automation and AI can, I think, give almost 30% to 40% improvement in the way you interact with your employees or customers in terms of answering their questions, and without them having to wait as they used to have to in the past, in a call center environment.
Andrew: Right. But then also, and I think this is the key. Also, it's something else you're collecting data about, or something else you can measure. And I think that that becomes the reservoir. You now want to measure something else, some level of customer engagement, for instance, or customer success, or understand where, for instance, your customers are asking questions. And, any way you can measure those conversations, those interactions can lead to different, for instance, product strategies or some understanding of where your customers are struggling, or where there might be opportunities based on the sorts of things they're asking. And, yes, for sure, through natural language, but we do it on everything. I mean, all of our customer interactions with customer support, for instance, these are things that can drive a lot of intelligence on what's going on in the market. And yeah, it's this hunger for data. And then, I think, companies both have to leverage that in services, like you mentioned, certainly on the NLP side, leverage it, not build their own. But also, I think, need a strategy in general for how they're going to collect data and utilize that data from everything, from health of systems and services and security, but also higher level customer interactions and those things as well. I think just that mindset of, everything we build and deploy needs to be measurable, observable, needs to give us the telemetry so that we can figure out anything based on it, right?
Vishal Gupta: Andrew, very well said. And I hundred percent agree because any journey with AI is going to be a journey, right?
Andrew: Right. What really interests me is how we assure that the system is working based on the customer's intent. And that requires, I think this is a key part of it. To build these systems, you need the data. It's the, whatever, the network data factor, the data network effect. You need data in order to analyze the data, to understand, to build strategies, to get more data. And the more data you get, the better your value can be, the more value you can generate. And so, a big thing companies struggle with, I think, is how do they get that data? How do they get the data from the customers? How do they get enough data in order to make the analytics meaningful?
Vishal Gupta: You know, as you continuously gather a lot of data, the ability to do clustering and get insights from the data about questions that you don't even know, is very compelling. The one thing I would add is that, in addition to the focus on data, we have also seen, for example, that many times people focus so much on getting the data, cleaning the data, that they don't put sometimes enough focus on what would be the most compelling use cases, let's say, once they have the data. What are the big, the most important problems? I think it's a question of a nice balance. They need to also know which things are going to truly move the needle, which are the most important questions, and then how do you get the data to be able to, either, answer or refine those most important questions? And then the only other thing I would add is that the world of AI is a journey, because even once you have a large set of data, essentially what you are doing with AI is to predict with a certain confidence interval, let's say, as part of a prediction, an answer. And obviously the more data you get, the more accurate your predictions can be. But AI can also suffer from things like what's called false positives, which means you are saying something is true. Let's say you're at a border crossing and you're saying," Hey, this person might be a problem." And you're flagging these people. Now, maybe you're always able to get the bad people, but you also end up getting 10 good people for every bad person you get, so you end up creating a lot of lines. So in the world of AI, I think the amount of data is very key to make sure that both the accuracy and the false positives accuracy is high, and false positives are low. But then you also need to know what are the right use cases that you're really after.
Andrew: Yes. I agree with the points you made. And I've certainly seen examples of, part of it is just systems in general, right? So, I'm going to use AI or whatever to optimize this part of a system to make it leaner, faster, cheaper, whatever the case. But what's the knock on impact/ effect on its related systems? And you gave a good example with lines. So if I make this way faster, given it may have false positives, or whatever effect it might have, does the next thing and the system need to change as well? And I think people do local optimization, or think more often about local optimization, than system optimization. Are we really optimizing the right point on the system to meet our requirements? And it always comes back to requirements, what are we trying to do as a business? The exploratory side of data, I have no idea what I might find, let me just play, is a swimming pool I like to jump into a lot. But yeah, that's just one piece of the puzzle. But I think the exploratory pieces is quite interesting.
Vishal Gupta: Yeah. I think one of the areas that I've seen a lot of promise in AI is an area called anomaly detection. So AI is very good at finding patterns. So it can say, Hey, no human can really digest data coming from, say, if you've got 50 or 80,000 devices, very hard to understand what's going on. But AI is good at that. I think it can figure out the data that's coming from lots of different devices and say, this particular device never was connecting with something else and transmitting this level of data. Something strange is going on. And that could be very helpful in security, for example. There are problems that AI is very good at solving, based upon all the things you've seen. And there are ones that are more, I would say, exploratory in nature. When you start talking about the bias or the right data not being clean, or not having the right signals that you've collected as part of it. So that's where I think the balance comes in, to neither hype AI, nor to be so afraid of it that you don't leverage it. That's where the balanced approach is needed.
Andrew: Yeah, no, for sure. And anomaly detection is certainly a very interesting area to us as well. I mean, it could be something as simple as, in the world of DNS, a user- driven device peels off four or 5, 000 queries in a day to spread across a variety of different domains, split across worky type stuff and what you might be doing during lunch hour type stuff. And all of a sudden, if that number changes dramatically, or if it skews towards a certain type of domain, a single domain or two or three domains, it becomes obvious pretty quickly to a human looking at it. You don't necessarily even need AI. But, to get finer differences, you certainly do. And that's where we think about it a lot. What is the baseline? What's normal? And how can I spot an anomaly? Let's hit the talent question again. It's a conversation I have quite often. And certainly, as we've gone through our journey at BlueCat, or as I look at what my customers are investing in, it brings up this conversation over and over and over again. How do we ensure that we have the talent base to drive, whether we're talking about security or data- driven strategies or digital transformation in general? Give me your thoughts on that.
Vishal Gupta: Yeah. I think this is a great topic that you're asking because, ultimately, your talent, your ability to attract the talent, retain the talent, engage the talent, is going to be the biggest factor that determines whether your business plan is going be successful or not. And so we take this area perhaps as the most serious area in terms of the driver for success. When I joined, as I mentioned, about a year and a half back, we looked at, we've got about a couple thousand engineers through six technology centers, and did an inventory around what skills do we have versus what do we need in this new world, to be successful? And as we talked with a lot of analysts, did a lot of internal brainstorming, talked to our board. We figured there were five key technology competencies that we wanted to develop. And we created these five cross- functional teams to create the content for what we would call a learning track. Each of the learning tracks ended up having about 20 pieces of content, which took about eight hours to digest. And we thought, okay, every two months, people could do one learning track. And so over the year, they could do the five learning tracks. And we weren't quite sure what level of adoption we'll get, but we wanted to see how our team would respond to it. We wanted to measure everything, as we discussed, with data. So we wanted to, as soon as people finished any of the, either the learning track, or the inaudible of the learning track, we wanted them to quickly give us feedback on a survey, 1 to 10, how effective was the content? We were blown away by three things. One, people actually adopted this far more vigorously than we thought. And that was because they realized, knowledge is key. We can't give them job security, but they can get career security if they got the right skills. So these things gave them the right skills. And they really resonated with it. Number two, they gave us a lot of feedback. We very quickly figured out which content was good, which was not, and we drove this continuous improvement. And so, ultimately we ended up having almost a 9.3, on a scale of 10, for the overall content that we developed. So we were very pleased with that. And third, instead of it taking 10 months, which is the time we had given for people to do it, the average person got this done in seven months. So they actually embraced this far faster than we thought. And so that was that year one of our journey. Now we're developing what we call the intermediate and advanced learning tracks in these areas to be taken by a subset of the people, instead of all the people who are interested in it. And obviously we brought some talent from the outside as well.
Andrew: I think that's great. You need that combination. Certainly learning, and then certainly if there are areas where there's not enough experience where you can't have mentoring as part of that learning, for instance, and there's a little bit of hiring. But yeah, I think when talent becomes people, when you're now working with individuals, and those individuals have career ambitions and learning ambitions, and you're making it more real, more about the people that you have on the teams and aligning what you need as a company with where those people want to grow. And really, I think Nirvana is when you're able to match your need for certain skills and capabilities with the desires and ambitions of an employee base. Now you have the ingredients you need to transform what that employee base may be able to do, and at the same time, create a great deal of engagement because you have people, and those people are immersed in this learning journey as well.
Vishal Gupta: Yeah, I think it's well said. And actually, that's our focus this year, because I think one thing we did well, and one thing I would say we didn't do well. The thing we did well was we got everybody super excited and kind of trained up on these things. The thing that I think we're going to try to do better this year is to really do that match between the areas people are more interested in, and where there's a business need, and get them trained up in those areas. So their new skills match with the work they're doing. And so, we wanted to make sure everybody had a base set of skills, which now they do. And now I think for the next step of the journey, we'll hopefully do exactly what you're saying, which is try to see if we can drive their interest with alignment, with the next step of knowledge, with the alignment of the work that we know we'll have for them this year.
Andrew: Yeah, for sure. I mean, look, just at the very basic level, in software organizations, there's no business need to grow 100 software architects. There is a business need to grow a couple and then also grow some principal engineers and some other very senior roles. And so, I think it's people's desires, their competencies. And frankly, it's part of the digital transformation journey as well is, what tools and what data can I use to help solve for this? But at some level, I think it's just, what's changed in employee engagement and things like career plans. And that alignment, I think, is super difficult to do, but it goes into who you're trying to recruit as well. If you're adding net new people on teams, then where do those net new people want to grow in their career, is a crucial part of it. But I think that part of what makes a company successful is something that oftentimes isn't top of mind, and should be top of mind. Prior guest, David Mark, said it really well. At the end of the day, it doesn't matter what these technologies are. There are people that are doing these things or people that are implementing. There are people that are understanding business requirements. There are people that are making decisions. And, as employers, I think that's a critical part of what we do. Well, fantastic. It's been an absolute pleasure talking to you. I really enjoyed the conversation. I think these topics, certainly employees and people and talent, but artificial intelligence in general, and security, is something that we will be talking about for many, many years, to say the absolute obvious.
Vishal Gupta: I really enjoyed it as well, Andrew. I think it's very fascinating what you've done at BlueCat. And congratulations there on all the progress as well.
Andrew: Awesome. Well, thank you again.
Speaker 2: Thank you for listening. I'd love to know what you thought of this episode. And I'm all ears if you have a guest recommendation. You can tweet at Network Disrupted, leave a review on Spotify or Apple podcast, or email me at andrew @ networkdisrupted. com.
In this episode, I talk with Vishal Gupta, Chief Technology Officer Unisys. We discuss the value AI actually has (and doesn’t have) in an enterprise setting, common data-collection misconceptions, low-hanging fruit for organization-wide use of AI (specifically improving customer experience and classifying devices for policy application), and how Unisys is equipping its employees career security through skills training.
Vishal is the CTO of Unisys, the IT company we’ve known for over a century. Since he started in 2018, Vishal has been focused on helping Unisys’ customers apply analytics in practical ways, and on enabling his organization to attract, engage, and retain talent. We talk about both things in the episode.
P.S. Know who I should speak to next? Drop me a line at andrew [at] networkdisrupted [dot] com.
P.P.S. Short on time? I send episode summaries to my email subscribers. Add yourself to the list if you’d like.
Read more about Vishal Gupta on our blog