In this bonus episode of the Agile Coaches’ Corner podcast, Christy Erbeck, the Chief People Officer at AgileThought, is serving as your guest host for today’s conversation with Dr. Jerry Smith.
In their conversation today, they discuss how to make AI work for your enterprise. Dr. Jerry Smith explains why understanding causality is critical for AI-driven business transformation, and how data science and analytics can help enterprise clients transform and become the digital winners that they desire to be.
- How AgileThought aids enterprises in understanding AI-driven business transformation:
- Come up with a working set of definitions for AI, machine learning, and data science
- How AgileThought helps their enterprise clients solve their problems:
- The question: “What is data?” should be asked (Dr. Jerry Smith’s answer: “Data is the debris of human activity; it’s because of us, not in spite of us”)
- Note: Data is not just spontaneously created in your data systems; it’s created from an application which captures an interaction between a human being (you, your customers, or your admins/salespeople) and that system
- Note: The data we see is because of human actions
- When we look at our capabilities, we should be asking the fundamental question: “What data in our enterprise is causal to our business outcomes?”
- For example, ask: “What data that you have spent time collecting is directly causing your revenue to perform the way it does?”
- The very first thing to ask is: “What is causal?”
- Once you know the causal data, you can go back to the application and the human and say, “How do I change the human behavior so that the application picks up the new behavior and changes the data?” This result is causal-based data engineering for AI, and is the only way to change your organization
- The question: “What is data?” should be asked (Dr. Jerry Smith’s answer: “Data is the debris of human activity; it’s because of us, not in spite of us”)
- AgileThought helps companies institutionalize data science, machine learning, and AI at the enterprise level by breaking down the process (as shown below), so that each and every process resides in infrastructure and a set of capabilities
- There are three kinds of data: Your enterprise, your IT, and your opensource – the goal is to get this data into a single machine learning record
- This single machine learning record is critical in showing all of the variables in columns and observations in rows – from there, you can do basic analytics, and then, data science
- Data scientists make sense of the data and create models out of the data, so the data no longer has to be used
- In the machine learning phase, data scientists try to predict what these models are trying to do and how they’re going to change under certain variables
- Note: AI is about prescriptions; making decisions
- Note: The biggest value is not in generating or reading reports; it is in making an appropriate decision based on these reports
About Dr. Jerry Smith: Dr. Jerry Smith is AgileThought’s Managing Director of Analytics and Data Science. As a practicing AI & Data Scientist, thought leader, innovator, speaker, author, and philanthropist, Dr. Jerry Smith is dedicated to advancing and transforming businesses through evolutionary computing, enterprise AI and data sciences, machine learning, and causality.
Transcript [This transcript is auto-generated and may not be completely accurate in its depiction of the English language or rules of grammar]
Intro: [00:02] Welcome to Agile Coaches’ Corner by AgileThought. The podcast for practitioners and leaders, seeking advice to refine the way they work and pave the path to better outcomes.
Christy Erbeck: [00:11] Welcome to the Agile Coaches’ Corner. My name is Christy Erbeck, chief people officer at AgileThought. And I am the guest host today talking with Dr. Jerry Smith, who is AgileThought’s managing director of analytics and data science. This conversation will be all about how to make AI work for the enterprise. Dr. Jerry, welcome to the call.
Dr. Jerry Smith: [00:35] Well thank you for having me, and it’s really great to be part of the team here at AgileThought.
Christy Erbeck: [00:41] Really happy to have you here. And just in some of the conversations that I’ve been witness to in the last several weeks since you’ve joined have been super interesting, there’s so much, in this realm of data science and AI, particularly that I think we can learn from you and help our clients succeed and be, you know, to really transform into that digital enterprise that they’re seeking to become.
Dr. Jerry Smith: [01:06] Yeah, I think you’re, I think you’re right about that. Um, you know, AI has gone through a couple of different phases, right? Uh, you know, going back to the eighties, it went through that, you know, “What is AI?” And it was a big experimentation. I remember back in the original days, just spending time on something called a back propagation network and teaching it how to play tic tac toe and simple games like that today, all that stuff is done, right? I mean, we’ve got these great, awesome capabilities, uh, building, uh, you know, deep neural networks that are genetically optimized and feature extraction. And it’s wonderful stuff. The problem is not in the neural networks, but the real challenge we’re having today is in the development of enterprise systems based on it. And I think that’s the conversation that most executives are looking for today. They know this stuff works within some degree of tolerance, but they just don’t know how to build it out. And if we can provide any insight to folks on, on how to do that, I think that’s the best value add.
Christy Erbeck: [02:12] That’s a great question. How do we help them figure out a method to the madness of a process, an approach that will make sense for them? How do we do that?
Dr. Jerry Smith: [02:24] Yeah, it starts off by coming up with sort of a normal set of definitions, right? I mean, in this role today, AI has been upscounded by everybody. You know, AI took for one group of people. It means something different for another. And what we’ve done is we’ve broken AI out separately from machine learning and data sciences. And why have we done that? Because it makes us easier. It makes it easier for us to build stuff. So the way we started AgileThought is we go from the range on the far left, you were looking at a graphic, you’d see your data, right? Your three kinds of data, your enterprise, your it, and your open source. That’s all the stuff that’s outside the company, that data then comes in and is data engineer. What does it do? We got to bring it all together. We got to bring the realtime stuff, the bat stuff together. We got to composite it. And the goal of that is to get that into a single machine learning record. And that’s the most important thing. If you look at most of the time a data scientists, which we haven’t got to yet spends this time, it’s in data engineering, right? So those high end people are wasting 80% of their time orchestrating data. So at the enterprise level, we need to bring that in. We have, we have great partners like with Microsoft and Amazon and Databricks and others that are capable of doing that. So we create that machine learning record, that single instance that says, here’s all my variables in columns. And here’s all my observations in rows. Now the fun starts. From there, we can do basic analytics, which are descriptive. How much data do I have? How often does it change? What does it look like? Quality, stuff like that. But then we move into data science, and this is where we differ from most. And it’s, it’s an operational difference. We can argue over the details of it, but the operational differences data scientists are really designed to make sense out of the data. Right? If you think about IQ in the United States, we got 350 million people. If you ask somebody what’s the average IQ, you don’t want to have to read every row. Every column of that information come up with. A data scientist comes back and says, listen, so 115 points normally distributed plus, or minus a 15 to 18 point standard deviation. That’s what a data scientist does. He makes sense makes models out a data. So we no longer have to use the data in the machine learning phase. We try to predict what those models are going to do both in time and space. How’s, how’s your, how’s your, how’s the IQ going to change over the next year? How’s the IQ differ from left to right coast, right? That sort of stuff. What we’re doing is we’re using the models and predicting how they’re going to change underlying variables. And then finally AI is about what prescriptions. It’s about making decisions. Our biggest value we have in an organization, the human being, is not generating reports, reading reports, even predicting something it’s making an appropriate decision based on it. So artificial intelligence, artificial human beings is designed to do that. So the very first thing we do in helping companies, institutionalize data science, machine learning, AI at the enterprise level is break down the process in that way so that each and every process then resides in infrastructure and then equally important a set of capabilities. AgileThought is very capabilities oriented. And we’ll probably talk about that a little bit later on.
Christy Erbeck: [05:47] I was just going to say, what, when you talk about capabilities, what does that mean to you? And again, how, how do our enterprise clients take what we have to offer and help them solve problems?
Dr. Jerry Smith: [06:02] Yeah. It’s, a great, great question. Um, and anytime I hear the word help solve problems, my brain sort of triggers in the, you know, very, there’s a couple of ways to answer that one. That’s a good one. Let me think on that one. Uh, the first piece is let’s, let’s go back to the capabilities. Um, one of the Rosetta stone phrases that I’d like to introduce to folks is, is the concept around what is data? Remember? We started off on the left hand side, right? We started off with data and typically nobody even asked the question, what is data? And, you know, you’d get all sorts of answers, ones and zeros, you know, bits and bytes and that sort of stuff. And I say, data is the debris of human activity. It’s because of us, and not in spite of us. You got to think about that for a second. That one phrase will set you aside and release you from all sorts of problems that organizations have. Data is the debris of human activity. It’s because of us not in spite of us. And what I mean by that is data isn’t just spontaneously created in your data systems. Data comes from what? An application. That application does what? It captures an interaction between a human being, your customers, or your admins, your salespeople, and that system. So the data we see is because of our human capabilities, our actions. And why is that important? If a CEO comes along and says, you know, team I’m really unhappy with our revenue, we just don’t have an admin go in and multiply by two. One it’s highly illegal, but it just doesn’t work that way. So what do we have to do? The very first thing we do and this again, separates us differently is when we’re, when we look at our capabilities, is to ask the fundamental question, what data in your enterprise is causal to your business outcomes? You want revenue to go up? Okay, great. You just spent $25 million on developing a big, you know, Hadoop base, a big as your base data warehouse out there, what data that you’ve been spending all that time collecting and, and maturating is directly causing your revenue to, to perform the way it does. Most, customers don’t know that your health or your employees, your, I mean, your health over your, constituents, your patients, you know, hospital, what data in your enterprise directly relates to it. So the very first thing we do is you ask the question, what’s causal? Now, why is that important? Think about it. We just said, data is the debris of our human activity. We’ve now identified with one of the many capabilities that is part of our capability process, that the causal data, the rest of stuff can correlate, which is great. It’s great for you know, creating red flags and stuff like that. But once you know, the causal data, what can you do? You can go back to the application and then go back to the human and say, how do I change the human behavior? So that, that application picks up that new behavior and changes the data, which results in what the new business school that’s causal based data engineering for AI. It is the only way that you could actually change your change, your organization.
Christy Erbeck: [09:22] It sounds pretty much of a game changer for organizations, if they can get to that causality. And they can really define that, then it sounds as though understanding that will help them make better business decisions remain relevant in the marketplace, and even perhaps make different decisions than they would have they not had that information.
Dr. Jerry Smith: [09:47] Absolutely. Right. I mean, if you, if let’s just take an example, let’s take a large scale financial service firm that does credit card processing for retailers. On an annual basis, this one I’m thinking about loses somewhere between a billion and a half and $2 billion a year because people don’t repay their cards. Right, so when we started working with them, the very first question I ask is what, right? You know me by now, right? We start with the data side of the house. We do that. We do the expression. Data is the debris of human activity because of us, not in spite of us, the very question I ask is why don’t they repay their credit card bills? And do you have data that captured that relationship and they’ve turned to each other? And they said, we don’t know why. We have programs in place, you know, call out programs, uh, asset retrieval, write off programs, but we don’t know why they don’t repay their bills. And that was a, that actually got everybody to stop and think. So we did the process. We actually took all their data. We brought it in, data engineered it. Did the machine owning record. Again, variables as columns, rows as observation, the row being you paid, you didn’t pay all the variables in there, uh, about 1,400 different variables that were captured, large scaled variable. This is a huge solution space. Um, and then what we found out was interesting, actually, the problem that we ended up getting onto wasn’t why they didn’t repay. We actually found a cohort of people that didn’t pay that started to pay. And the team thought, Oh, that must be due to our programs. We said, really? So let’s take a look. We did the cause out, by the way, this is the only time you’ll hear me say that we use mutual information theory, which is an advanced statistical process to actually measure causality. So when somebody says, well, how do you measure it? Right? It’s an entropy measurement within data. It’s pretty sophisticated. It’s kind of fun. You get a chance to say mutual information theory and several times at a party and that and $1.25 you still can’t buy a cup of coffee at Starbucks. That’s a different issue. So, so we went in and we found out, it turns out none of their programs were causally related to this, which meant what these people were repaying based upon some other external intrinsic value. And what they ended up doing at the end was doing one simple thing. When you were classified as this demographic, they’d send you an email saying, Hey, looks like you’re going through some difficult times. We’re here to help. And that was it.
Christy Erbeck: [12:17] I love that. I love the fact that they were able to dig into what was happening with their customers in that way.
Dr. Jerry Smith: [12:23] Yeah. Now, you ask the question, well, what did this cost them? Right. Well, okay. They didn’t reduce their multi million dollar program by that amount right but what it costs them was in the area of better clients that said, thank you because these people were financially responsible just on downtimes. And that’s all they did. So the, you know, this notion of starting off with a question, why is that happening? Figuring out what’s causal is so important. And then from there we can do the fun stuff, right? If that isn’t fun enough, once you have the data that’s causal, Oh, by the way, there are three categories of customers that come out of that. You’re a CIO CXO working in this space, you’re a director who’s responsible for the warehouse. Somebody sits down and says, I want this business value to be optimized. Let’s go look at our data. And three things can come back. First is, Hey boss. None of the data we collect is causal for driving our business problems. What does that say? That says, you’re not in control of your organization. You’ve got external factors that are responsible and are out of your control. What I mean by control is you haven’t collected the data. Now your people may be doing things, but you can’t prove they’re doing things through data. So we got to fix that. That’s a transformative opportunity for you. That’s the one of the best outcomes to say, okay, we’ve got to transform what new applications, what new journeys with our customers, what new customers do we need at the other extreme is, gosh, you know, we’ve got a substantial amount of data that is or related or outcomes, and we can control it. Now it’s all about maturation, right? It’s all about continuing that and exploiting it. And then you’ve got that kind of wishy washy space in between, right? That 50% area where we kind of have some stuff and we don’t have some stuck. And in this particular space, that’s an opportunity to evolve. Bring in third party data sources. Most of the times that we see people, there is a causal element, but it’s not a strong one. And when you bring in IT and open source data, you end up drawing in more causal capabilities. So once you have a causality, that’s when you get into prediction prescription, right? We build with that causal element, we build digital surrogates of the thing. And for example, I love this example of fish salmon, right? Europe has something like 26, 27 salmon farms over there. Question is, is how do we make all salmon grow at the same rate, make them fat over the same amount of time with the same outcome. And right now it isn’t unified. And so the idea is, again, go collect all the data. What’s causal making a fat salmon, right? Feeding programs, nutrition values. So water, temperature, air temperature, phase of the moon, whether Jerry wore black boots or not during the feeding example, we then figure out what’s causal. We’ll walk through that process. It turns out Jerry wearing black boots has nothing to do with the salmon getting fat, but the phase of the moon does. Very interesting. So then from there, once you have the causal data, you can build a prescriptive surrogate model. We build a digital surrogate of a salmon and it says, I’m a salmon. You give me that data. I’ll tell you if I get fat or not over how much time, how powerful is that? How do you use that prediction? Again, we went from data science now into the machine learning predictive model based upon those characteristics, while we now develop an AI prescription. And what it does is it looks through all the possible combinations of those inputs. It’s an optimization problem, genetically. And it says, given all these combinations, what are you going to do? Salmon says, “eh, I get fat, but it takes me a long time.” No one says “I don’t get fat and I die.” Another one says, “I get that really fast, but it takes a lot of resources.” So we optimize those inputs to achieve a peak. And then what do we do with that prescription? We go to the field. We tell our people start feeding this way. At this time, you know, wearing these clothes at this phase of the moon, we go to our sales people and we say, start saying these things. We go to our, our, our hospitals, and we say, use this prescription when it comes to discharging people so that they don’t return. And then we wait, what do we wait for? Wait for people to change new interactions, those applications that collect that information, new data wash, rinse, repeat you walk through that process of causality, digital surrogates and prescriptions. Your business has to change. There’s no way it cannot change. That’s what we mean at AgileThought. And we talk about our transformative process, our Transform, our Build and our Run is tightly integrated together to move your company. That is the power of AI.
Christy Erbeck: [17:02] That is fantastic. Dr. Jerry, I think that we could have many more conversations like this to learn how data science and analytics can help our enterprise clients transform and become the digital winners that they desire to be. So thank you for joining us today. And I look forward to our next conversation.
Dr. Jerry Smith: [17:26] Well thank you for having me. And I appreciate the opportunity to talk with you and all the folks that are listening. And the one piece of advice that I would leave you is something that was told to me a long time ago, that AI is really too important to be left to the IT folks. We’re really smart IT people, but artificial intelligence is really about changing your business. And if you don’t understand what it is as a business person, then you’re not bad. You’re not wrong. It’s the people helping you out. And at at AgileThought, that’s what we’re here to do. We’re here to help you through this process and make it useful. Just like your employees are helping you out.
Christy Erbeck: [18:00] That’s great. We’ll leave it there. Thank you again for joining us on the Agile Coaches’ Corner. And I look forward to our next conversation, Dr. Jerry.
Outro: [18:10] This has been the Agile Coaches’ Corner podcast brought to you by AgileThought. The views, opinions and information expressed in this podcast are solely those of the hosts and the guests, and do not necessarily represent those of agile thought. Get the show notes and other helpful tips for this episode and other episodes at agilethought.com/podcast.