In this Digital CxO Leadership Insights Series interview, Mike Vizard talks to Mark Do Couto, senior vice president for data analytics at Altair about why legacy approaches to analytics don’t meet the needs of modern digital business transformation initiatives.
Mike Vizard: Hey, folks, welcome to the latest edition of the Digital CxO Leadership Insight series. I’m your host Mike Vizard. Today we’re talking with Mark Do Couto; he is the senior vice-president for data analytics at Altair, and they just bought RapidMiner and WPS. And then we’re gonna dive into what’s going on with opensource programming languages and data analytics, and how digital transformation is gonna be driven by all of this – that was a mouthful – otherwise known as lions and tigers and bears, oh, my. Hey, Mark, welcome to the show.
Mark Do Couto: Hey, Mike, how’s it going?
Mike Vizard: So you guys just bought RapidMiner and WPS, and then, I’m not sure that a lot of people know exactly what you do. So how does one, plus one, plus one equals something five or more?
Mark Do Couto: Absolutely, no, I appreciate it. So, first, Altair, as an organization, has a data analytics component to the business. We have a complete portfolio that helps our clients with everything from data preparation to data science to visualization. And throughout that journey, we’ve looked to not only build some more features and functions and products, but look at acquisitions, as well. So RapidMiner is our most recent one; just over a week ago, we closed on RapidMiner. And RapidMiner is a, you know, low-code, no-code data science environment where data analysts, business users alike, go in and start to build AI machine learning models in that low-code, no-code environment. Great user interface, they’ve got over a million users, really known in the data space, so we thought it was a great fit for us.
And then, at the end of last year, we acquired a company called WPS out of the UK. They were known for building their own language interpreter and engine to execute SAS code, so the language of SAS code, and this really, we thought, gave our customers the ability to finally have a choice on where they’re executing some of that code, even if it’s old legacy SAS code. And really modernize where they’re coding their analytics, whether it’s R, Python, or SAS. So, we think that those two acquisitions, along with the portfolio that we already have, really strengthens the overall, you know, value that we can bring to our customers in data analytics.
Mike Vizard: How is the way we need to think about data analytics evolving in this digital transformation age? It used to be kind of a high-priest sort of thing, you needed to know a bunch of R programming languages, and you would work with SAS and there would be some folks that maybe they had some business intelligence tools on their desktop, but they were pretty much canned applications. How are we getting to the point where we’re taking the true power of, you know, what used to be known as very high-end data analytics, and democratizing it and making it more accessible to everybody?
Mark Do Couto: Yeah, I think the key, there, is how do we make it more accessible and put it in the hands of the business users, the people that really maybe understand the data. I agree with you, there was this period of time that kind of the coders were the magicians. If you knew R or if you knew Python, you were the most valuable piece in that data science group. And, you know, still very valuable, still an asset that organizations truly need, but I think it’s how do we bring that concept of data science and AI machine learning, and bring it to the business users so they can leverage their insight that they have on the data. No one knows the data better than the guys that are in it every single day, and those people that are leveraging that data really has that understanding that can then use that insight to build those better decisions.
And, you know, the algorithms and all these different tools that will help you build these machine learning models, they’re all great, but if you’re not leveraging that with real insight that you already have inherently in the data, then you can lose out on some of that additional value. So, that’s when we look at it as, you know, how do we put it in the hands of the many, in a way that we aren’t, you know, devaluating the tool, and still allowing those experts and advanced statisticians and analytical, you know, Ph.D.s, give them the ability to kind of still dig in deep with the tool but still have a broader use for it.
Mike Vizard: How do we inject that into a digital process? ‘Cause historically, people would use these tools to analyze what already occurred, and it was kind of like, you know, trying to drive the car looking through the rearview mirror. So how do we kind of move this forward to the point where we are making better decisions faster, driven by data that’s injected into the actual business process at hand?
Mark Do Couto: Yeah, I think, you know, in that sense, it’s really understanding how you’re optimizing, you know, the models that you’re building with this data. So to your point, a lot of the time we spent, historically, was looking at where things are today, right? “Here is the data that I have. What does that look like today?” and, you know, what direction do we go in based on the way that data looks. Now it’s taking historical data, taking some of these different types of machine learning models, and layering on aspects of optimization where you can bring in different factors, whether it’s factors around budgets, factors around certain constraints, factors around what objectives you’re trying to achieve. And really leverage that in these models, so you can make, you know, future type decisions that you can actually measure and monitor.
So that’s kind of the way that we work with our clients to try and help them understand kind of making those decisions and leveraging the data to make those decisions, and we think that there’s these different elements that are brought into it. Instead of just reporting on the news, trying to, you know, be more thought-provoking and be more responsive to what the data is trying to tell you that might happen in the future.
Mike Vizard: Are the people consuming this now the traditional analysts that we’ve seen forever and a day, they were business analysts and all those folks? Or is it much more the line of business executives now are directly consuming this data and trying to use it to manage a process in near-real-time, as it were?
Mark Do Couto: So I think it’s, more and more that we’ve seen over the years, it’s been moving more towards that business line. And the interesting thing there is, you know, a lot of the business line, they may not understand Python code. Or, you know, if someone talks about a regression model, let’s say, they may not see the value in that. But they know, you know, “Will this drive profit? Will this bring profitability up? Will this make that production more efficient? And what level efficiency? What percent efficiency gain can I get from that?” And I think that’s the nice thing about taking all this, you know, statistical information and putting it in such a way that the business user can get true value from it.
That’s what we’re seeing, that next step, the next evolution of looking at something that may have traditionally been kind of reporting and then predictive analytics, and taking that next step into prescriptive analytics. And that’s, I think, what the tool is starting to show our customers.
Mike Vizard: Do people trust the data to drive that level of prescriptive analytics? Because, you know, as he once joked, he said, “It’s one thing to be wrong. It’s another thing to be wrong at scale,” so how do we kind of make sure that we have some guardrails in here, in terms of what we actually make prescriptive?
Mark Do Couto: Yeah, if my statistician colleagues or Ph.D.s were talking with us right now, they would probably be saying something like, you know, that’s what champion challenger models are for, and that’s why you do a testing control environment, and kind of walk before you run, if you will. But I think that is a true testament to really understanding, to your point, how much you can trust the data. Predictive analytics, prescriptive analytics, AI machine learning, it’s only as good as the data is, and yes, you may initially see some great outcomes that a model shoots out. But when you put it into production and really start to run with it, then you see what the true business outcomes are. So we always encourage our customers to do some kind of test control group, you know, see what the results are looking like, and do a transition and shift as you go live with some of these predictive models.
A lot of our customers are in the BFSI space, so a lot of the big banks, financial services, insurance, healthcare, and a lot of them have been using these types of predictive models over the years, so it’s not, you know, nuances to them. But there are a number of customers that, you know, they dip their toe in data analytics, and they dip their toe in reporting and visualization, and they’re ready to take that next step. And we’re helping them take those steps by doing things like champion challenger models and, like I said, test and control. So, I think that really starts to build that level of comfort, and when you can actually see what some of the results and some of the lift is on that basis, you get that comfort that the data is actually showing some good results, and you can continue to move forward with it.
Mike Vizard: There has been a divide between the data science teams and the people who build applications, it’s been around for years, and yet, to achieve what we wanna talk about, just a minute ago, is I need those models to get injected into those applications. And the applications teams are running on dev ops models and they’re basically updating code at a fairly regular clip, and the data science guys are building models at a much slower rate. So how do I kind of meld those two things together in a way that gives me the optimal result, short of maybe, I don’t know, maybe it’s not a bad idea, throwing them in a room and locking the door and see what happens.
Mark Do Couto: So there’s a couple of elements, there, so one way is to try and get, you know, the people building the models to actually spend more time testing the models than they do building them. And there’s a couple of ways around this. I think this is why the concept of AutoML has become more and more popular, because you can basically click a button and build out 10, 15, 20 models, and then actually see which one’s gonna perform a little bit better, saving time in the model building process. So that’s one element of it. And the other element is, to truly understand, like, how you actually execute on the models when you wanna put’em into production, when you’re ready to say, “This is the model.” So that, you know, when it gets shifted over to that other team, they’re not looking at it saying, “Okay, now I got to reprogram all this to make it executable.”
And some of the things that we’ve found is, giving customers the choice on how the model is exported out really helps in the ability to kind of plug it in to multiple systems. So, you know, the ability to pull things out in Python or R, automatically, without even knowing how to code in that language. Being able to convert a model to XML, PMML, Java, even, and again, so it can run in these different disparate systems that might be out there, so we don’t have to give it to a group that has to start from scratch and reprogram a model that may have quickly been built. So, I think there is still some work to be done to make sure that these teams play nicely together, but I think we’re trying to focus on spending less time, you know, manually building out these models one by one, and slowing down that process. But making it easier to build the number of models, and test and validate the models so that they can feel comfortable with the one they wanna move forward with. And then convert that model in a way that plays nicely in the ecosystem that they run in.
Mike Vizard: We’ve also seen just a huge explosion in the number of open source technologies playing in this category. It used to be very much a proprietary technology space, and now we see everything from open source frameworks to programming languages and everything in it. Is the pace of innovation gonna change as a result? Or what’s the total end result of all this open source contributions in this space?
Mark Do Couto: I think the one thing that I would hope that the audience would agree with is open source isn’t going anywhere. If anything, the proliferation of open source is just gonna continue to build momentum. It’s really, how do we leverage it in a way that our organizations feel comfortable with it. And I think that’s what the key is, there. And at Altair, we embrace open source, we embrace the different coding languages and the different elements of open source, but again, within a framework and toolset that it can be leveraged, it can be tested, validated. You know, we work with a lot of different organizations that have validation teams, governance, auditors, so we wanna make sure that all the checks and balances are done before something goes into production.
And I think that’s really what the key is, is embracing what’s being populated out there, what the next best kind of model is, let’s just say, or next best algorithm. And then put it in a way that we can validate it, test it, make the organizations feel comfortable with it, because something like that goes into production. So, I don’t see the momentum slowing down, I see it continuing, and I think it’s gonna come down to, you know, the vendors that we’re used to being a little bit more proprietary, seeing if they can, you know, start to embrace it a little bit more.
Mike Vizard: What’s your best advice to folks who are starting to clearly embrace AI and data science more? But clearly, we also see a lot of people stumbling, so, what is it that you see people doing that just makes you shake your head and go, “Ah, maybe we should be smarter than that”?
Mark Do Couto: So I would say two things. One, giving up too early, right, I think there’s a frustration level or an idea that, “You know what, there’s just no point, I haven’t seen the ROI on it, yet, so let’s just stick to reporting the news and telling me what I got and stick to that.” I think if you give it some time, you give the team the ability and the tools to do what they need to do, you’ll start to see that outcome. And then I think the other element is thinking that AI and machine learning is overly-complex and only valuable to super-large organizations. I think that, you know, AI and machine learning, you know, predictive analytics in general, can really serve multiple purposes, and help organizations small or large. You don’t need to be a massive Fortune 100 enterprise to leverage the power of AI machine learning.
So, I think, you know, the most important thing when I talk to customers, you know, on a daily basis and brand-new conversations, it just comes down to, “Do you have data today? And do you have access to that data? And is there, you know, future insight you want from that data?” If those answers are yes, then I think that it’s a right move to move forward and start trying to embrace some of these tools that will help you, you know, gain further insight from your data.
Mike Vizard: And folks, and one other thing to consider there, eventually, you’re gonna be playing around with AI sooner or later, and the sooner you know about it, the happier you’re gonna be, ’cause the learning curve is just gonna get that much higher as we go along. Mark, thanks for being on the show.
Mark Do Couto: No problem. Thanks, Mike.
Mike Vizard: All right, with that, folks, thank you for watching our latest episode. You can find this one and others just like it on the digitalcxo.com website. And once again, thanks for spending some time with us. Take care.