Mathematical Optimization: Disrupting Decision Intelligence – Digital CxO Podcast EP118

Hello and welcome to the digital CXO podcast. I am Amanda Ani. And with me today I have Duke Pucci.

He is the CEO of Gu Robbi Optimization. How are you doing? I'm doing well.

Thanks for having me, Amanda. Happy to have you on the show. So can you talk a little bit about Gu Robbie optimization?

What services do you provide? Sure, sure. So we, uh, we're an analytics software company.

We live in the space of artificial intelligence, but in a space that not many people know about. You know, most people know about machine learning, you know, being able to predict outcomes from big streams of data. And ours a little bit different.

Ours would tell you the optimal decision to make. So you have, uh, many, you know, business operations have very, very complex decisions that need to be taken. And what mathematical optimization will do is take a look at that entire decision space, which is sometimes, you know, trillions of possible solutions to a given decision.

And it will tell you what is the optimal decision to take. If you're trying to optimize something, let's say you're trying to optimize your revenue or your profit as a company, mathematical optimization will define that dec decision space and then give you what, what is the best outcome. So if, if you're trying to maximize profit, it'll give you the best outcome to get there.

So it's, it's a little bit different than predictive analytics. It's known as prescriptive analytics 'cause it prescribes the ideal outcome to take. Hmm.

That is a new term that I have not yet heard before. So share a little bit more. Do you have some use case examples that you could give our audience?

Sure, sure. So, uh, let's see. Uh, let's say you are, uh, an ice cream company and you have a production facility and you've got produce the right amount of ice cream to supply the, the warehouses or supply the retail establishments, um, with the right amount of inventory at the right, in the right amount of time.

But behind all of that, you have a certain number of constraints. And those constraints could be the amount of packaging you have for, for a one pint container. It could be the amount of cacao that you have, uh, on hand.

It could be the amount of fresh strawberries or cream or, or all of that. net or something like that. And in that you would describe all of those variables.

You know, how many production lines you have, how much vanilla you have on stock, how much cacao you have on stock. Then you'd, you'd build in all of the constraints. And the constraints are the limitations, right?

You can't produce a million pints of chocolate ice cream if retailers only need half a million pints. So you build out this entire sequence in a mathematical model and that model will come back and tell you precisely what to produce. So in this example, uh, you might have 500 pounds of cacao on stock and only a hundred pounds of vanilla.

Um, the model will see those constraints and will decide, Hey, you need to produce more, more chocolate ice cream right now because that cacao is gonna go out of stock or out of date. And then you're gonna have a lot of wasted, uh, uh, material on your hands. So it looks at all of these hundreds of thousands of different variables and constraints, and then builds the perfect production plan for you to take advantage of all of those variables, keeping within your constraints and maximizing your profit or maximizing your revenue.

And it's used, it's really used across probably 60 or so different industries. So 80% of the, the Fortune 500, uh, out there is using g**o or using mathematical optimization, and you can use it for any sort of problem. We have, uh, lumber companies that use it to figure out what parts of the forest to fell first to maintain certain sustainability standards.

Um, and then we've got companies on the other end of the spectrum that are, you know, making, making cookies as profitably as they can or delivering nutritional products to babies, uh, that are in the horn of Africa, that at the risk of starving, delivering those at the right time, um, so that we're, they're saving lives. So it, it, it, it fills really any need out there that you have. Uh, which is what's really fascinating about it.

It's very, very flexible. Um, but it typically hides behind the scenes. Most people don't see it.

It's usually built into some other sort of application that's helping you figure out what to produce or how to deliver things. Uh, like Uber, Uber is a, is a, uh, customer of ours, and you could imagine they're trying to figure out how to route drivers in the most cost-effective manner, and also how to route drivers so that people aren't waiting too long for, for their Uber. Um, so all sorts of problems can be solved with, uh, with math optimization.

Great. Use case examples. So would this method be in lieu of the other methods that we mentioned, or would it be along with these other methods?

Yeah, it's very complimentary to, uh, to other methods. So a lot of our customers today use, um, uh, data science techniques like machine learning. They'll use those to predict, uh, demand.

So like going back to the ice cream example, um, you'll use predictive analytics to figure out how much, uh, demand there will be for a given product. And then, uh, second in that flow, they'll use mathematical optimization to figure out how to live up to that demand in the most profitable way. So it, it does work in concert with those other techniques that are out there, which makes it really handy to the, the typical data scientist out there.

And is this a relatively new technology concept or has this been around a while? And like you said, it's just behind the scenes, so we just don't know a lot about it. Yeah, it's actually, it's been around since the forties, uh, 1940s.

Um, it really didn't become useful to commercial, um, users until the late nineties, and that's because there wasn't enough compute power. These problems require, sometimes they require significant amounts of, of computing resources. Um, but I'd say really in the last 20 years is when it's really picked up.

You know, you've had this whole advent of what, what's known as big data, so people now have access to big streams of really clean data. Then you had computing power that got a lot bigger. You know, anybody can go and, and, and get access to hundreds of Amazon machines in this day and age.

Um, those two things helped, uh, optimization to, to pull out of the shadows a little bit, but it's, it still stays a little bit behind the scenes because most people, uh, really can't work with optimization. You, it really requires an expert, somebody who's trained in operations research that knows how to take a problem and build a digital twin for that problem and, and then solve the, the problem as that digital twin. So we're getting there.

Um, people, people have more awareness of it than they had in the past. In fact, um, you know, a lot of the analyst firms like Gartner and Forrester now talk specifically about optimization and the role that it plays in decision making. But we still have a bit of a, we still have a little bit of ways to go before every CTO out there knows precisely the benefits that optimization can bring to them.

And I'm glad you mentioned digital twin, because now that is a term I've been hearing a lot about the digital twin technology. And so would, would any digital twin technology be using this, um, mathematical optimization? Many of them would, yeah.

So if, if you have a, a factory that you've replicated in a digital, in a digital space, you would then be using optimization to figure out ways to optimize that production line. Um, so yeah, many times if you have a digital twin, you have optimization built into it somewhere. And that is fascinating technology, especially in the, the health sector, uh, seeing how they're using the digital twin technology along with ai, which is of course what everybody's talking about the last couple of years.

Yeah. Uh, so what, from your perspective, um, if a business is trying to incorporate, uh, mathematical optimization, uh, into their technology, what is the first Step? So the first step would be to find the people in your organization that can do this sort of thing.

So back in the old days, it was only people that had an operations research background. Now it's most data scientists out there. Um, so you find the, the analytics team or the data science team within the organization.

The next step in the process is that that data science team then creates that digital twin of, of the issue that you're trying to solve. So they spend time with stakeholders, they under understand the situation, they understand all of the inputs to the problem, and that takes probably a month to do it properly. And then once that digital twin is built, then it's as simple as having a piece of software like g**o, which is just known as a solver.

Uh, that piece of software then reads that digital twin and you tell the software what you're trying to do, you can either maximize or minimize something. So you would work with that data science team and you'd say, Hey, for this particular problem type, we're trying to minimize waste or minimize pollution or maximize this or maximize that. Um, you then run the, run the solution and that there's your, your answer.

Then you make the adjustments to the production plan or to the dispatch plan for the trucks or, you know, whatever you're using the, the software for. So from, from your experience, do you think that there, uh, is a great need for people with this skillset? Are are uh, companies experiencing a shortage of people with this skillset?

There's definitely a shortage. Um, the, the us uh, bureau of Labor Statistics predicts that both data science and operations research, those two roles are gonna grow by 25% a year every year for the next decade. And that's because, you know, if, uh, most of the, most of the decision making that big businesses take today are all built out of large data streams, right?

So you need people that can predict, um, outcomes and you need people that can find the best outcome for you. So yeah, there's definitely a shortage. I mean, there's been a shortage of data scientists, uh, to the tune of, of several million worldwide for the last five or six years.

I don't know if we'll ever fill the hole of all the demand, but uh, if I was coming outta college right now, I'd want to be a data scientist. No, no question about it. You can write your ticket in many ways.

That's good to know because I know, um, on the other side of that coin, we maybe hear a lot about how AI is gonna be this, this big solution and, and they're not gonna need data scientists anymore 'cause AI is gonna do it all. How likely do you think that is? What are your Yeah, Uh, there's no question that AI is gonna change the, the landscape of things.

Um, I think you'll still always need smart analytics professionals because ai, you know, there are hallucinations, right? It, it makes the wrong call. Um, sometimes it does get better and better over time, but I think, I think the need for analytics professionals is always gonna be out there.

But yeah, AI is going to replace some of the more basic roles that data scientists are playing today. I mean, we see that here at our company. It's helping us in, in a number of different ways, whether it's, uh, the way we build the software, the way we test the software.

Um, what we find though is the benefits of AI just allow our experts to focus more on the, on the things that are really important. And they use AI to do more of the routine things. So I, I do think companies are always gonna need deep experts, but it's, yeah, it's really changing the landscape out there.

There's no, no question about it. And it's exciting from my standpoint, it's exciting to see, I mean, at G**o we think that there's a huge benefit to using large language models to build those digital twins. 'cause right now building digital twin could take several months.

It could be a, you know, 10,000 or a hundred thousand lines of code and you could use in many cases LLMs to do a lot of that heavy lift for you. So you get all of that sort of heavy work outta the way and then you can use your data scientists for the fine work, like the detailed work within those models. I think there's a lot of value that can be built there.

Awesome. Well, if there was one key takeaway you could leave our audience with today, what would that be? Oh, I guess it's, it, uh, it's around mathematical optimization.

It is such a powerful technology that a lot of people aren't really aware of. I think when you're in a business context and you're facing a problem that could have hundreds or thousands or millions of possible solutions and you've really gotta drive some sort of efficiency, you've got to look to mathematical optimization to, to do that data science, you know, machine learning can't do it. You've gotta make sure that you have a, a good understanding of what problems you can bring to bear, uh, mathematical optimization.

'cause it is a very, very powerful technique. Alright, thank you so much for coming on the show and sharing that with us today. You're welcome.

Thanks for having me. And thank you to our audience. Stay tuned.

There's more.