[00:00:00] Dr Genevieve Hayes: Hello, and welcome to Value Driven Data Science, the podcast that helps data scientists transform their technical expertise into tangible business value, career autonomy, and financial reward. I’m Dr. Genevieve Hayes, and today I’m joined by three globally recognized innovators and leaders in AI, analytics, and data science.
[00:00:24] Bill Schmarzo, Mark Stouse, and John Thompson. Bill? Also known as the Dean of Big Data, is the AI and Data Customer Innovation Strategist for Dell Technologies AI Spear Team, and is the author of six books on blending data science, design thinking, and data economics from a value creation and delivery perspective.
[00:00:49] He is an avid blogger and is ranked as the number four influencer worldwide in data science and big data Analytica. And he’s also an adjunct professor at Iowa State University, where he teaches AI driven innovation. Mark is the CEO of proofanalytics. ai, a causal AI company that helps organizations understand and optimize their operational investments in light of their targeted objectives, time lag and external factors.
[00:01:23] Known for his ability to bridge multiple business disciplines, he has successfully operationalized data science at scale across large enterprises. Driven by his belief that data science’s primary purpose is enabling better business decisions. And John is EY’s global head of AI and is the author of four books on AI data and analytics teams.
[00:01:49] He was named one of DataIQ’s 100 most influential people in data in 2023. and is also an adjunct professor at the University of Michigan, where he teaches a course based on his book, Building Analytics Teams. Today’s episode will be a tough love wake up call for data scientists on why you are failing to deliver real business value and more importantly, what you can do about it.
[00:02:17] So get ready to boost your impact. Earn what you’re worth and rewrite your career algorithm. Bill, Mark, John, welcome to the show.
[00:02:25] Mark Stouse: Thank
[00:02:26] Bill Schmarzo: Thanks for having us.
[00:02:27] John Thompson: to be here.
[00:02:28] Dr Genevieve Hayes: Only 11 percent of data scientists say their models always deploy. Only 10 percent of companies obtain significant financial benefits from AI technologies and close to 85 percent of big data projects fail. These statistics, taken from research conducted by Rexa Analytics, the Boston Consulting Group and Gartner respectively, paint a grim view of what it’s like working as a data scientist.
[00:02:57] The reality is, you’re probably going to fail. And when that reality occurs, it’s not uncommon for data scientists to blame either the executive for not understanding the brilliance of their work, or the corporate culture for not being ready for data science. And maybe this is true for some organizations.
[00:03:20] Particularly those relatively new to the AI adoption path. But it’s now been almost 25 years since William Cleveland first coined the term data science. And as the explosive uptake of generative AI tools, such as chat GPT demonstrate with the right use case. People are very willing to take on AI technologies.
[00:03:42] So perhaps it’s finally time to look in the mirror and face the truth. Perhaps the problem is you, the data scientist. But if this is the case, then don’t despair. In many organizations, the leadership just don’t have the time to provide data scientists with the feedback necessary to improve. But today, I’m sitting here with three of the world’s best to provide that advice just for you.
[00:04:09] So, let’s cut to the chase what are the biggest mistakes you see data scientists making when it comes to demonstrating their value?
[00:04:18] Mark Stouse: I think that you have to start with the fact that they’re not demonstrating their value, right? I mean, if you’re a CEO, a CFO, head of sales really doesn’t matter if you’re trying to make better business decisions over and over and over again. As Bill talks about a lot, the whole idea here is economic,
[00:04:39] and it is. About engaging, triggering the laws of compounding you’ve got to be able to do stuff that makes that happen. Data management, for example, even though we all agree that it’s really necessary, particularly if you’re launching, you know, big data solutions. You can’t do this sequentially and be successful.
[00:05:04] You’re going to have to find some areas probably using, you know, old fashioned math around causal analytics, multivariable linear regression, things like that, to at least get the ball rolling. In terms of delivering better value, the kind of value that business leaders actually see as valuable
[00:05:29] I mean, one of the things that I feel like I say a lot is, you have to have an understanding of your mission, the mission of data science. As somebody who, as a business leader champions it. Is to help people make those better and better and better decisions. And if you’re not doing that, you’re not creating value.
[00:05:52] Full stop.
[00:05:53] Bill Schmarzo: Totally agree with Mark. I think you’re going to find that all three of us are in violent agreement on a lot of this stuff. What I find interesting is it isn’t just a data scientist fault. Genevieve, you made a comment that leadership lacks the time to provide guidance to data scientists. So if leadership Is it treating data and analytics as an economics conversation if they think it’s a technology conversation is something that should be handled by the CIO, you’ve already lost, you’ve already failed, you already know you failed,
[00:06:24] Mark mentioned the fact that this requires the blending of both sides of the aisle. It requires a data scientist to have the right mindset to ask questions like what it is that we’re trying to achieve. How do we create value? What are our desired outcomes? What are the KPIs metrics around which are going to make your success?
[00:06:39] Who are our key stakeholders? There’s a series of questions that the data scientist must be empowered to ask and the business Leadership needs to provide the time and people and resources to understand what we’re trying to accomplish. It means we can go back old school with Stephen Covey, begin with an end in mind.
[00:07:01] What is it we’re trying to do? Are we trying to improve customer retention? We try to do, you know, reduce unplanned operational downtime or improve patient outcomes. What is it we’re trying to accomplish? The conversation must, must start there. And it has to start with business leadership, setting the direction, setting the charter, putting the posts out where we want to go, and then the data science team collaborating with the stakeholders to unleash that organizational tribal knowledge to actually solve
[00:07:32] Dr Genevieve Hayes: think a lot of the problem comes with the fact that many business leaders see data science as being like an IT project. So, if you’ve got your Windows upgrade, the leadership It gives the financing to IT, IT goes along and does it. And then one morning you’re told, when you come into work, your computer will magically upgrade to the latest version of Windows.
[00:07:55] So no one really gets bothered by it. And I think many business leaders treat data science as just another IT project like that. They think they can just Give the funding, the data scientists will go away and then they’ll come in one morning and the data science will magically be on their computer.
[00:08:15] Bill Schmarzo: Yeah, magic happens, right? No, no, magic doesn’t happen, it doesn’t happen. There has to be that leadership commitment to be at the forefront, not just on the boat, but at the front of the boat saying this is the direction we’re going to go.
[00:08:29] John Thompson: That’s the whole reason this book was written. The whole point is that, analytics projects are not tech projects. Analytics projects are cultural transformation projects, is what they are. And if you’re expecting the CEO, CFO, CIO, COO, whoever it is, to go out there and set the vision.
[00:08:50] That’s never going to happen because they don’t understand technology, and they don’t understand data. They’d rather be working on building the next factory or buying another company or something like that. What really has to happen is the analytics team has to provide leadership to the leadership for them to understand what they’re going to do.
[00:09:12] So when I have a project that we’re trying to do, my team is trying to do, and if we’re working for, let’s say, marketing, I go to the CMO and I say, look, you have to dedicate and commit. that your subject matter experts are going to be in all the meetings. Not just the kickoff meetings, not just the quarterly business review, the weekly meetings.
[00:09:36] Because when we go off as an analytics professionals and do things on our own, we have no idea what the business runs like. , we did analytics at one company that I work for. We brought it back and we showed it to the they said, the numbers are wildly wrong. And we said, well, why? And they said, well, you probably don’t understand that what we do is illegal in 10 US states.
[00:10:00] So you probably have the data from all those 10 states in the analysis. And we did. So, we took it all out and they look down there and go, you got it right. It’s kind of surprising. You didn’t know what you were doing and you got it right. So, it has to be a marriage of the subject matter experts in the business.
[00:10:17] And the data scientists, you can’t go to the leadership and say, tell us what you want. They don’t know what they want. They’d want another horse in Henry Ford’s time, or they glue a, a Walkman onto a radio or something in Steve Jobs time. They don’t know what they want. So you have to come together.
[00:10:36] And define it together and you have to work through the entire project together.
[00:10:42] Mark Stouse: Yeah, I would add to that, okay, that a lot of times the SMEs also have major holes in their knowledge that the analytics are going to challenge and give them new information. And so I totally agree. I mean, this is an iterative learning exchange. That has profound cultural implications.
[00:11:11] One of the things that AI is doing right now is it is introducing a level of transparency and accountability into operations, corporate operations, my operations, your operations, that honestly, none of us are really prepared for. None of us are really prepared for the level of learning that we’re going to have to do.
[00:11:36] And very few of us are aware of how polymathic. Most of our challenges, our problems, our objectives really are one of the things that I love to talk about in this regard is analytics made me a much better person. That I once was because it showed me the extent of my ignorance.
[00:12:01] And when I kind of came to grips with that and I started to use really the modicum of knowledge that I have as a way of curating my ignorance. And I got humble about it made a big difference
[00:12:16] John Thompson: Well, that’s the same when I was working shoulder to shoulder with Bill, I just realized how stupid I was. So, then I just, really had to, come back and, say, oh, God nowhere near the summit, I have a long way to go.
[00:12:31] Bill Schmarzo: Hey, hey, Genevie. Let me throw something out there at you and it builds on what John has said and really takes off on what Mark is talking about is that there is a cultural preparation. It needs to take place across organizations in order to learn to master the economies of learning,
[00:12:48] the economies of learning, because you could argue in knowledge based industries that what you are learning is more important than what you know. And so if what you know has declining value, and what you’re learning has increasing value, then what Mark talked about, and John as well, both city presenting data and people saying, I didn’t know that was going on, right?
[00:13:09] They had a certain impression. And if they have the wrong cultural mindset. They’re going to fight that knowledge. They’re going to fight that learning, oh, I’m going to get fired. I’m going to get punished. No, we need to create cultures that says that we are trying to master the economies and learning and you can’t learn if you’re not willing to fail.
[00:13:29] And that is what is powerful about what AI can do for us. And I like to talk about how I’m a big fan of design thinking. I integrate design thinking into all my workshops and all my training because it’s designed to. Cultivate that human learning aspect. AI models are great at cultivating algorithmic learning.
[00:13:50] And when you bring those two things together around a learning culture that says you’re going to try things, you’re going to fail, you’re going to learn, those are the organizations that are going to win.
[00:13:59] John Thompson: Yeah, you know, to tie together what Mark and Bill are saying there is that, you need people to understand that they’re working from an outmoded view of the business. Now, it’s hard for them to hear that. It’s hard for them to realize it. And what I ask data scientists to do that work for me is when we get a project and we have an operational area, sales, marketing, logistics, finance, manufacturing, whatever it is.
[00:14:26] They agreed that they’re going to go on the journey with us. We do something really simple. We do an exploratory data analysis. We look at means and modes and distributions and things like that. And we come back and we say, this is what the business looks like today. And most of the time they go, I had no idea.
[00:14:44] You know, I didn’t know that our customers were all, for the most part, between 70 and 50. I had no idea that our price point was really 299. I thought it was 3, 299. So you then end up coming together. You end up with a shared understanding of the business. Now one of two things is generally going to happen.
[00:15:05] The business is going to freak out and leave the project and say, I don’t want anything to do with this, or they’re going to lean into it and say, I was working from something that was, as Bill said, declining value. Okay. Now, if they’re open, like a AI model that’s being trained, if they’re open to learning, they can learn what the business looks like today, and we can help them predict what the business should look like tomorrow.
[00:15:31] So we have a real issue here that the three of us have talked about it from three different perspectives. We’ve all seen it. We’ve all experienced it. It’s a real issue, we know how people can come together. The question is, will they?
[00:15:46] Dr Genevieve Hayes: think part of the issue is that, particularly in the area of data science, there’s a marked lack of leadership because I think a lot of people don’t understand how to lead these projects. So you’ve got Many data scientists who are trained heavily in the whole technical aspect of data science, and one thing I’ve come across is, you know, data scientists who’ll say to me, my job is to do the technical work, tell me what to do.
[00:16:23] I’ll go away and do it. Give it to you. And then you manager can go and do whatever you like with it.
[00:16:29] Mark Stouse: Model fitment.
[00:16:31] Dr Genevieve Hayes: Yeah. And then one thing I’ve experienced is many managers in data science are, you know, It’s often the area that they find difficult to find managers for, so we’ll often get people who have no data science experience whatsoever
[00:16:46] and so I think part of the solution is teaching the data scientists that they have to start managing up because they’re the ones who understand what they’re doing the best, but no one’s telling them that because the people above them often don’t know that they should be telling the data
[00:17:08] John Thompson: Well, if that’s the situation, they should just fire everybody and save the money. Because it’s never going to go anywhere. But Bill, you were going to say something. Go ahead.
[00:17:16] Bill Schmarzo: Yeah, I was going to say, what’s interesting about Genevieve, what you’re saying is that I see this a lot in not just data scientists, but in a lot of people who are scared to show their ignorance in new situations. I think Mark talked about this, is it because they’re, you think about if you’re a data scientist, you probably have a math background. And in math, there’s always a right answer. In data science, there isn’t. There’s all kinds of potential answers, depending on the situation and the circumstances. I see this all the time, by the way, with our sales folks. Who are afraid we’re selling technology. We’re afraid to talk to the line of business because I don’t understand their business Well, you don’t need to understand their business, but you do need to become like socrates and start asking questions What are you trying to accomplish?
[00:18:04] What are your goals? What are your desired outcomes? How do you measure success? Who are your stakeholders ? You have to be genuinely interested In their success and ask those kind of questions if you’re doing it to just kind of check a box off Then just get chad gpt to rattle it off But if you’re genuinely trying to understand what they’re trying to accomplish And then thinking about all these marvelous different tools you have because they’re only tools And how you can weave them together to help solve that now you’ve got That collaboration that john’s book talks about about bringing these teams together Yeah
[00:18:39] Mark Stouse: is, famously paraphrased probably did actually say something like this, . But he’s famously paraphrased as saying that he would rather have a really smart question than the best answer in the world. And. I actually experienced that two days ago,
[00:18:57] in a conversation with a prospect where I literally, I mean, totally knew nothing about their business. Zero, but I asked evidently really good questions. And so his impression of me at the end of the meeting was, golly, you know, so much about our business. And I wanted to say, yeah, cause you just educated me.
[00:19:21] Right. You know, I do now. And so I think there’s actually a pattern here that’s really worth elevating. So what we are seeing right now with regard to data science teams is scary similar to what happened with it after Y2K, the business turned around and looked at him and said, seriously, we spend all that money,
[00:19:45] I mean, what the heck? And so what happened? The CIO got, demoted organizationally pretty far down in the company wasn’t a true C suite member anymore. Typically the whole thing reported up into finance. The issue was not. Finance, believing that they knew it better than the it people,
[00:20:09] it was, we are going to transform this profession from being a technology first profession to a business outcomes. First profession, a money first profession, an economics organization, that has more oftentimes than not been the outcome in the last 25 years. But I think that that’s exactly what’s going on right now with a lot of data science teams.
[00:20:39] You know, I used to sit in technology briefing rooms, listening to CIOs and other people talk about their problems. And. This one CIO said, you know, what I did is I asked every single person in my organization around the world to go take a finance for non financial managers course at their local university.
[00:21:06] They want credit for it. We’ll pay the bill. If they just want to audit it, they can do that. And they started really cross pollinating. These teams to give them more perspective about the business. I totally ripped that off because it just struck me as a CMO as being like, so many of these problems, you could just do a search and replace and get to marketing.
[00:21:32] And so I started doing the same thing and I’ve made that suggestion to different CDOs, some of whom have actually done it. So it’s just kind of one of those things where you have to say, I need to know more. So this whole culture of being a specialist is changing from.
[00:21:53] This, which, this is enough, this is okay , I’m making a vertical sign with my hand, to a T shaped thing, where the T is all about context. It’s all about everything. That’s not part of your. Profession
[00:22:09] John Thompson: Yeah, well, I’m going to say that here’s another book that you should have your hands on. This is Aristotle. We can forget about Socrates. Aristotle’s the name. But you know. But , Bill’s always talking about Socrates. I’m an Aristotle guy myself. So, you
[00:22:23] Bill Schmarzo: Okay, well I Socrates had a better jump shot. I’m sorry. He could really nail that
[00:22:28] John Thompson: true. It’s true. Absolutely. Well, getting back , to the theme of the discussion, in 1 of the teams that I had at CSL bearing, which is an Australian company there in Melbourne, I took my data science team and I brought in speech coaches.
[00:22:45] Presentation coaches people who understand business, people who understood how to talk about different things. And I ran them through a battery of classes. And I told them, you’re going to be in front of the CEO, you’re going to be in front of the EVP of finance, you’re going to be in front of all these different people, and you need to have the confidence to speak their language.
[00:23:07] Whenever we had meetings, we talk data science talk, we talk data and integration and vectors and, algorithms and all that kind of stuff. But when we were in the finance meeting, we talked finance. That’s all we talked. And whenever we talked to anybody, we denominated all our conversations in money.
[00:23:25] Whether it was drachma, yen, euros, pounds, whatever it was, we never talked about speeds and feeds and accuracy and results. We always talked about money. And if it didn’t make money, we didn’t do it. So, the other thing that we did that really made a difference was that when the data scientists and data scientists hate this, When they went into a meeting, and I was there, and even if I wasn’t there, they were giving the end users and executives recommendations.
[00:23:57] They weren’t going in and showing a model and a result and walking out the door and go, well, you’re smart enough to interpret it. No, they’re not smart enough to interpret it. They actually told the marketing people. These are the 3 things you should do. And if your data scientists are not being predictive and recommending actions, they’re not doing their job.
[00:24:18] Dr Genevieve Hayes: What’s the, so what test At the end of everything, you have to be able to say, so what does this mean to whoever your audience is?
[00:24:25] Mark Stouse: That’s right. I mean, you have to be able to say well, if the business team can’t look at your output, your data science output, and know what to do with it, and know how to make a better decision, it’s like everything else that you did didn’t happen. I mean it, early in proof, we were working on. UX, because it became really clear that what was good for a data scientist wasn’t working. For like everybody else. And so we did a lot of research into it. Would you believe that business teams are okay with charts? Most of them, if they see a graph, they just totally freeze and it’s not because they’re stupid.
[00:25:08] It’s because so many people had a bad experience in school with math. This is a psychological, this is an intellectual and they freeze. So in causal analytics, one of the challenges is that, I mean, this is pretty much functioning most of the time anyway, on time series data, so there is a graph,
[00:25:31] this is kind of like a non negotiable, but we had a customer that was feeding data so fast into proof that the automatic recalc of the model was happening like lickety split. And that graph all of a sudden looked exactly like a GPS. It worked like a GPS. In fact, it really is a GPS. And so as soon as we stylized.
[00:26:01] That graph to look more like a GPS track, all of a sudden everybody went, Oh,
[00:26:10] Dr Genevieve Hayes: So I got rid of all the PTSD from high school maths and made it something familiar.
[00:26:16] Mark Stouse: right. And so it’s very interesting. Totally,
[00:26:21] Bill Schmarzo: very much mirrors what mark talked about So when I was the new vice president of advertiser analytics at yahoo we were trying to solve a problem to help our advertisers optimize their spend across the yahoo ad network and because I didn’t know anything about that industry We went out and my team went out and interviewed all these advertisers and their agencies.
[00:26:41] And I was given two UEX people and zero data. Well, I did have one data scientist. But I had mostly UX people on this project. My boss there said, you’re going to want UX people. I was like, no, no, I need analytics. He said, trust me in UX people and the process we went through and I could spend an hour talking about the grand failure of the start and the reclamation of how it was saved at a bar after too many drinks at the Waldorf there in New York.
[00:27:07] But what we’ve realized is that. For us to be effective for our target audience was which was media planners and buyers and campaign managers. That was our stakeholders. It wasn’t the analysts, it was our stakeholders. Like Mark said, the last thing they wanted to see was a chart. And like John said, what they wanted the application to do was to tell them what to do.
[00:27:27] So we designed this user interface that on one side, think of it as a newspaper, said, this is what’s going on with your campaign. This audience is responding. These sites are this, these keywords are doing this. And the right hand side gave recommendations. We think you should move spend from this to this.
[00:27:42] We think you should do this. And it had three buttons on this thing. You could accept it and it would kick into our advertising network and kick in. And we’d measure how effective that was. They could reject it. They didn’t think I was confident and we’d measure effectiveness or they could change it. And we found through our research by putting that change button in there that they had control, that adoption went through the roof.
[00:28:08] When it was either yes or no, adoption was really hard, they hardly ever used it. Give them a chance to actually change it. That adoption went through the roof of the technology. So what John was saying about, you have to be able to really deliver recommendations, but you can’t have the system feel like it’s your overlord.
[00:28:27] You’ve got to be like it’s your Yoda on your shoulder whispering to your saying, Hey, I think you should do this. And you’re going, eh, I like that. No, I don’t like this. I want to do that instead. And when you give them control, then the adoption process happens much smoother. But for us to deliver those kinds of results, we had to know in detail, what decisions are they trying to make?
[00:28:45] How are they going to measure success? We had to really understand their business. And then the data and the analytics stuff was really easy because we knew what we had to do, but we also knew what we didn’t have to do. We didn’t have to boil the ocean. We were trying to answer basically 21 questions.
[00:29:01] The media planners and buyers and the campaign managers had 21 decisions to make and we built analytics and recommendations for each Of those 21
[00:29:10] John Thompson: We did the same thing, you know, it blends the two stories from Mark and Bill, we were working at CSL and we were trying to give the people tools to find the best next location for plasma donation centers. And, like you said, there were 50, 60 different salient factors they had, and when we presented to them in charts and graphs, Information overload.
[00:29:34] They melted down. You can just see their brains coming out of their ears. But once we put it on a map and hit it all and put little dials that they could fiddle with, they ran with it.
[00:29:49] Bill Schmarzo: brilliant
[00:29:50] Mark Stouse: totally, totally agree with that. 100% you have to know what to give people and you have to know how to give them, control over some of it, nobody wants to be an automaton. And yet also they will totally lock up if you just give them the keys to the kingdom. Yeah.
[00:30:09] Dr Genevieve Hayes: on what you’ve been saying in the discussion so far, what I’m hearing is that the critical difference between what data scientists think their role is and what business leaders actually need is the data scientists is. Well, the ones who aren’t performing well think their role is to just sit there in a back room and do technical work like they would have done in their university assignments.
[00:30:33] What the business leaders need is someone who can work with them, ask the right questions in order to understand the needs of the business. make recommendations that answer those questions. But in answering those questions, we’re taking a data informed approach rather than a data driven approach. So you need to deliver the answers to those questions in such a way that you’re informing the business leaders and you’re delivering it in a way that Delivers the right user experience for them, rather than the user experience that the data scientists might want, which would be your high school maths graphs.
[00:31:17] Is that a good summary?
[00:31:20] John Thompson: Yeah, I think that’s a really good summary. You know, one of the things that Bill and I, and I believe Mark understands is we’re all working to change, you know, Bill and I are teaching at universities in the United States. I’m on the advisory board of about five. Major universities. And whenever I go in and talk to these universities and they say, Oh, well, we teach them, these algorithms and these mathematical techniques and these data science and this statistics.
[00:31:48] And I’m like, you are setting these people up for failure. You need to have them have presentation skills, communication skills, collaboration. You need to take about a third of these credits out and change them out for soft skills because you said it Genevieve, the way we train people, young people in undergraduate and graduate is that they have a belief that they’re going to go sit in a room and fiddle with numbers.
[00:32:13] That’s not going to be successful.
[00:32:16] Mark Stouse: I would give one more point of dimensionality to this, which is a little more human, in some respects, and that is that I think that a lot of data scientists love the fact that they are seen as Merlin’s as shamans. And the problem that I personally witnessed this about two years ago is when you let business leaders persist in seeing you in those terms.
[00:32:46] And when all of a sudden there was a major meltdown of some kind, in this case, it was interest rates, and they turn around and they say, as this one CEO said in this meeting Hey, I know you’ve been doing all kinds of really cool stuff back there with AI and everything else. And now I need help.
[00:33:08] Okay. And the clear expectation was. I need it now, I need some brilliant insight now. And the answer that he got was, we’re not ready yet. We’re still doing the data management piece. And this CEO dropped the loudest F bomb. That I think I have ever heard from anybody in almost any situation,
[00:33:36] and that guy, that data science leader was gone the very next day. Now, was that fair? No. Was it stupid? For the data science leader to say what he said. Yeah, it was really dumb.
[00:33:52] Bill Schmarzo: Don’t you call that the tyranny of perfection mark? Is that your term that you always use? is that There’s this idea that I gotta get the data all right first before I can start doing analysis And I think it’s you I hear you say the tyranny of perfection is what hurts You Progress over perfection, learning over absolutes, and that’s part of the challenge is it’s never going to be perfect.
[00:34:13] Your data is never going to be perfect, you got to use good enough data
[00:34:17] Mark Stouse: It’s like the ultimate negative version of the waterfall.
[00:34:22] John Thompson: Yeah,
[00:34:23] Mark Stouse: yet we’re all supposedly living in agile paradise. And yet very few people actually operate
[00:34:30] John Thompson: that’s 1 thing. I want to make sure that we get in the recording is that I’ve been on record for years and I’ve gone in front of audiences and said this over and over again. Agile and analytics don’t mix that is. There’s no way that those 2 go together. Agile is a babysitting methodology. Data scientists don’t do well with it.
[00:34:50] So, you know, I’ll get hate mail for that, but I will die on that hill. But, the 1 thing that, Mark, I agree with 100 percent of what you said, but the answer itself or the clue itself is in the title. We’ve been talking about. It’s data science. It’s not magic. I get people coming and asking me to do magical things all the time.
[00:35:11] And I’m like. Well, have you chipped all the people? Do you have all their brain waves? If you have that data set, I can probably analyze it. But, given that you don’t understand what’s going on inside their cranium, that’s magic. I can’t do that. We had the same situation when COVID hit, people weren’t leaving their house.
[00:35:29] So they’re not donating plasma. It’s kind of obvious, so, people came to us and said, Hey, the world’s gone to hell in a handbasket in the last two weeks. The models aren’t working and I’m like, yeah, the world’s changed, give us four weeks to get a little bit of data.
[00:35:43] We’ll start to give you a glimmer of what this world’s going to look like two months later. We had the models working back in single digit error terms, but when the world goes haywire, you’re not going to have any data, and then when the executives are yelling at you, you just have to say, look, this is modeling.
[00:36:01] This is analytics. We have no precedent here.
[00:36:05] Bill Schmarzo: to build on what John was just saying that the challenge that I’ve always seen with data science organizations is if they’re led by somebody with a software development background, getting back to the agile analytics thing, the problem with software development. is that software development defines the requirements for success.
[00:36:23] Data science discovers them. It’s hard to make that a linear process. And so, if you came to me and said, Hey, Schmarz, you got a big, giant data science team. I had a great data science team at Hitachi. Holy cow, they were great. You said, hey, we need to solve this problem. When can you have it done?
[00:36:38] I would say, I need to look at the problem. I need to start exploring it. I can’t give you a hard date. And that drove software development folks nuts. I need a date for when I, I don’t know, cause I’ve got to explore. I’m going to try lots of things. I’m going to fail a lot.
[00:36:51] I’m going to try things that I know are going to fail because I can learn when I fail. And so, when you have an organization that has a software development mindset, , like John was talking about, they don’t understand the discovery and learning process that the data science process has to go through to discover the criteria for success.
[00:37:09] Mark Stouse: right. It’s the difference between science and engineering.
[00:37:13] John Thompson: Yes, exactly. And 1 of the things, 1 of the things that I’ve created, it’s, you know, everybody does it, but I have a term for it. It’s a personal project portfolio for data scientists. And every time I’ve done this and every team. Every data scientist has come to me individually and said, this is too much work.
[00:37:32] It’s too hard. I can’t
[00:37:34] Bill Schmarzo: Ha, ha, ha,
[00:37:35] John Thompson: three months later, they go, this is the only way I want to work. And what you do is you give them enough work so when they run into roadblocks, they can stop working on that project. They can go out and take a swim or work on something else or go walk their dog or whatever.
[00:37:53] It’s not the end of the world because the only project they’re working on can’t go forward. if they’ve got a bunch of projects to time slice on. And this happens all the time. You’re in, team meetings and you’re talking and all of a sudden the data scientist isn’t talking about that forecasting problem.
[00:38:09] It’s like they ran into a roadblock. They hit a wall. Then a week later, they come in and they’re like, Oh, my God, when I was in the shower, I figured it out. You have to make time for cogitation, introspection, and eureka moments. That has to happen in data science.
[00:38:28] Bill Schmarzo: That is great, John. I love that. That is wonderful.
[00:38:30] Mark Stouse: And of course the problem is. Yeah. Is that you can’t predict any of that, that’s the part of this. There’s so much we can predict. Can’t predict that.
[00:38:42] Bill Schmarzo: you know what you could do though? You could do Mark, you could prescribe that your data science team takes multiple showers every day to have more of those shower moments. See, that’s the problem. I see a correlation. If showers drive eureka moments, dang it.
[00:38:54] Let’s give him more showers.
[00:38:56] John Thompson: Yep. Just like firemen cause fires
[00:38:59] Mark Stouse: Yeah, that’s an interesting correlation there, man.
[00:39:05] Dr Genevieve Hayes: So, if businesses need something different from what the data scientists are offering, why don’t they just articulate that in the data scientist’s role description?
[00:39:16] John Thompson: because they don’t know they need it.
[00:39:17] Mark Stouse: Yeah. And I think also you gotta really remember who you’re dealing with here. I mean, the background of the average C suite member is not highly intellectual. That’s not an insult, that’s just they’re not deep thinkers. They don’t think a lot. They don’t
[00:39:37] John Thompson: that with tech phobia.
[00:39:38] Mark Stouse: tech phobia and a short termism perspective.
[00:39:43] That arguably is kind of the worst of all the pieces.
[00:39:48] John Thompson: storm. It’s a
[00:39:49] Mark Stouse: It is, it is a
[00:39:50] John Thompson: know, I, I had, I’ve had CEOs come to me and say, we’re in a real crisis here and you guys aren’t helping. I was like, well, how do you know we’re not helping? You never talked to us. And, in this situation, we had to actually analyze the entire problem and we’re a week away from making recommendations.
[00:40:08] And I said that I said, we have an answer in 7 days. He goes, I need an answer today. I said, well, then you should go talk to someone else because in 7 days, I’ll have it. But now I don’t. So, I met with him a week later. I showed them all the data, all the analytics, all the recommendations. And they said to me, we don’t really think you understand the business well enough.
[00:40:27] We in the C suite have looked at it and we don’t think that this will solve it. And I’m like, okay, fine, cool. No problem. So I left, and 2 weeks later, they called me in and said, well, we don’t have a better idea. So, what was that you said? And I said, well, we’ve coded it all into the operational systems.
[00:40:43] All you have to do is say yes. And we’ll turn it on and it was 1 of the 1st times and only times in my life when the chart was going like this, we made all the changes and it went like that. It was a perfect fit. It worked like a charm and then, a month later, I guess it was about 6 months later, the CEO came around and said, wow, you guys really knew your stuff.
[00:41:07] You really were able to help us. Turn this around and make it a benefit and we turned it around faster than any of the competitors did. And then he said, well, what would you like to do next? And I said, well, I resigned last week. So, , I’m going to go do it somewhere else.
[00:41:22] And he’s like, what? You just made a huge difference in the business. And I said, yeah, you didn’t pay me anymore. You didn’t recognize me. And I’ve been here for nearly 4 years, and I’ve had to fight you tooth and nail for everything. I’m tired of it.
[00:41:34] Mark Stouse: Yeah. That’s what’s called knowing your value. One of the things that I think is so ironic about this entire conversation is that if any function has the skillsets necessary to forecast and demonstrate their value as multipliers. Of business decisions, decision quality, decision outcomes it’s data science.
[00:42:05] And yet they just kind of. It’s like not there. And when you say that to them, they kind of look at you kind of like, did you really just say that, and so it is, one of the things that I’ve learned from analytics is that in the average corporation, you have linear functions that are by definition, linear value creators.
[00:42:32] Sales would be a great example. And then you have others that are non linear multipliers. Marketing is one, data science is another, the list is long, it’s always the non linear multipliers that get into trouble because they don’t know how to show their value. In the same way that a linear creator can show it
[00:42:55] John Thompson: And I think that’s absolutely true, Mark. And what I’ve been saying, and Bill’s heard this until he’s sick of it. Is that, , data science always has to be denominated in currency. Always, if you can’t tell them in 6 months, you’re going to double the sales or in 3 months, you’re going to cut cost or in, , 5 months, you’re going to have double the customers.
[00:43:17] If you’re not denominating that in currency and whatever currency they care about, you’re wasting your time.
[00:43:23] Dr Genevieve Hayes: The problem is, every single data science book tells you that the metrics to evaluate models by are, precision, recall, accuracy, et
[00:43:31] John Thompson: Yeah, but that’s technology. That’s not business.
[00:43:34] Dr Genevieve Hayes: exactly. I’ve only ever seen one textbook where they say, those are technical metrics, but the metrics that really count are the business metrics, which are basically dollars and cents.
[00:43:44] John Thompson: well, here’s the second one that says it.
[00:43:46] Dr Genevieve Hayes: I will read that. For the audience it’s Business Analytics Teams by John Thompson.
[00:43:51] John Thompson: building analytics
[00:43:52] Dr Genevieve Hayes: Oh, sorry, Building
[00:43:54] Mark Stouse: But, but I got to tell you seriously, the book that John wrote that everybody needs to read in business. Okay. Not just data scientists, but pretty much everybody. Is about causal AI. And it’s because almost all of the questions. In business are about, why did that happen? How did it happen? How long did it take for that to happen?
[00:44:20] It’s causal. And so, I mean, when you really look at it that way and you start to say, well, what effects am I causing? What effects is my function causing, all of a sudden the scales kind of have a way of falling away from your eyes and you see things. Differently.
[00:44:43] John Thompson: of you to say that about that book. I appreciate that.
[00:44:46] Mark Stouse: That kick ass book, kick
[00:44:48] John Thompson: Well, thank you. But, most people don’t understand that we’ve had analytical or foundational AI for 70 years. We’ve had generative AI for two, and we’ve had causal for a while, but only people understand it are the people on this call and Judea Pearl and maybe 10 others in the world, but we’re moving in a direction where those 3 families of AI are going to be working together in what I’m calling composite AI, which is the path to artificial, or as Bill says, average general intelligence or AGI.
[00:45:24] But there are lots of eight eyes people talk about it as if it’s one thing and it’s
[00:45:29] Mark Stouse: Yeah, correct. That’s right.
[00:45:31] Dr Genevieve Hayes: I think part of the problem with causal AI is it’s just not taught in data science courses.
[00:45:37] John Thompson: it was not taught anywhere. The only place it’s taught is UCLA.
[00:45:40] Mark Stouse: But the other problem, which I think is where you’re going with it Genevieve is even 10 years ago, they weren’t even teaching multivariable linear regression as a cornerstone element of a data science program. So , they basically over rotated and again, I’m not knocking it.
[00:46:01] I’m not knocking machine learning or anything like that. Okay. But they over rotated it and they turned it into some sort of Omni tool, that could do it all. And it can’t do it all.
[00:46:15] Dr Genevieve Hayes: think part of the problem is the technical side of data science is the amalgamation of statistics and computer science . But many data science university courses arose out of the computer science departments. So they focused on the machine learning courses whereas many of those things like.
[00:46:34] multivariable linear analysis and hypothesis testing, which leads to things like causal AI. They’re taught in the statistics courses that just don’t pop up in the data science programs.
[00:46:46] Mark Stouse: Well, that’s certainly my experience. I teach at USC in the grad school and that’s the problem in a nutshell right there. In fact, we’re getting ready to have kind of a little convocation in LA about this very thing in a couple of months because it’s not sustainable.
[00:47:05] Bill Schmarzo: Well, if you don’t mind, I’m going to go back a second. We talked about, measuring success as currency. I’m going to challenge that a little bit. We certainly need to think about how we create value, and value isn’t just currency. John held up a book earlier, and I’m going to hold up one now, Wealth of Nations,
[00:47:23] John Thompson: Oh yeah.
[00:47:25] Bill Schmarzo: Page 28, Adam Smith talks about value he talks about value creation, and it isn’t just about ROI or net present value. Value is a broad category. You got customer value, employee value, a partner stakeholder. You have society value, community value of environmental value.
[00:47:43] We have ethical value. And as we look at the models that we are building, that were guided or data science teams to build, we need to broaden the definition of value. It isn’t sufficient if we can drive ROI, if it’s destroying our environment and putting people out of work. We need to think more holistically.
[00:48:04] Adam Smith talks about this. Yeah, 1776. Good year, by the way, it’s ultimate old school, but it’s important when we are As a data science team working with the business that we’re broadening their discussions, I’ve had conversations with hospitals and banks recently. We run these workshops and one of the things I always do, I end up pausing about halfway through the workshop and say, what are your desired outcomes from a community perspective?
[00:48:27] You sit inside a community or hospital. You have a community around you, a bank, you have a community around you. What are your desired outcomes for that community? How are you going to measure success? What are those KPIs and metrics? And they look at me like I got lobsters crawling out of my ears.
[00:48:40] The thing is is that it’s critical if we’re going to Be in champion data science, especially with these tools like these new ai tools causal predictive generative autonomous, these tools allow us to deliver a much broader range of what value is And so I really rail against when somebody says, you know, and not trying to really somebody here but You know, we gotta deliver a better ROI.
[00:49:05] How do you codify environmental and community impact into an ROI? Because ROI and a lot of financial metrics tend to be lagging indicators. And if you’re going to build AI models, you want to build them on leading indicators.
[00:49:22] Mark Stouse: It’s a lagging efficiency metric,
[00:49:24] Bill Schmarzo: Yeah, exactly. And AI doesn’t do a very good job of optimizing what’s already happened.
[00:49:29] That’s not what it does.
[00:49:30] John Thompson: sure.
[00:49:31] Bill Schmarzo: I think part of the challenge, you’re going to hear this from John and from Mark as well, is that we broaden this conversation. We open our eyes because AI doesn’t need to just deliver on what’s happened in the past, looks at the historical data and just replicates that going forward.
[00:49:45] That leads to confirmation bias of other things. We have a chance in AI through the AI utility function to define what it is we want our AI models to do. from environmental, society, community, ethical perspective. That is the huge opportunity, and Adam Smith says that so.
[00:50:03] John Thompson: There you go. Adam Smith. I love it. Socrates, Aristotle, Adam
[00:50:08] Bill Schmarzo: By the way, Adam Smith motivated this book that I wrote called The Economics of Data Analytics and Digital Transformation I wrote this book because I got sick and tired of walking into a business conversation and saying, Data, that’s technology. No, data, that’s economics.
[00:50:25] Mark Stouse: and I’ll tell you what, you know what, Genevieve, I’m so cognizant of the fact in this conversation that the summer can’t come fast enough when I too will have a book,
[00:50:39] John Thompson: yay.
[00:50:41] Mark Stouse: yeah, I will say this, One of the things that if you use proof, you’ll see this, is that there’s a place where you can monetize in and out of a model, but money itself is not causal. It’s what you spend it on. That’s either causal or in some cases, not
[00:51:01] That’s a really, really important nuance. It’s not in conflict with what John was saying about monetizing it. And it’s also not in conflict with what. My friend Schmarrs was saying about, ROI is so misused as a term in business. It’s just kind of nuts.
[00:51:25] It’s more like a shorthand way of conveying, did we get value
[00:51:31] John Thompson: yeah. And the reason I say that we denominated everything in currency is that’s generally one of the only ways. to get executives interested. If you go in and say, Oh, we’re going to improve this. We’re going to improve that. They’re like, I don’t care. If I say this project is going to take 6 months and it’s going to give you 42 million and it’s going to cost you nothing, then they’re like, tell me more, and going back to what Bill had said earlier, we need to open our aperture on what we do with these projects when we were at Dell or Bill and I swapped our times at Dell, we actually did a project with a hospital system in the United States and over 2 years.
[00:52:11] We knocked down the incidence of post surgical sepsis by 72%. We saved a number of lives. We saved a lot of money, too, but we saves people’s lives. So analytics can do a lot. Most of the people are focused on. Oh, how fast can we optimize the search engine algorithm? Or, how can we get the advertisers more yield or more money?
[00:52:32] There’s a lot of things we can do to make this world better. We just have to do it.
[00:52:36] Mark Stouse: The fastest way to be more efficient is to be more effective, right? I mean, and so when I hear. CEOs and CFOs, because those are the people who use this language a lot. Talk about efficiency. I say, whoa, whoa, hold on. You’re not really talking about efficiency. You’re talking about cost cutting.
[00:52:58] Those two things are very different. And it’s not that you shouldn’t cut costs if you need to, but it’s not efficiency. And ultimately you’re not going to cut your way into better effectiveness. It’s just not the way things go.
[00:53:14] John Thompson: Amen.
[00:53:15] Mark Stouse: And so, this is kind of like the old statement about physicists,
[00:53:18] if they’re physicists long enough, they turn into philosophers. I think all three of us, have that going on. Because we have seen reality through a analytical lens for so long that you do actually get a philosophy of things.
[00:53:38] Dr Genevieve Hayes: So what I’m hearing from all of you is that for data scientists to create value for the businesses that they’re working for, they need to start shifting their approach to basically look at how can we make the businesses needs. And how can we do that in a way that can be expressed in the business’s language, which is dollars and cents, but also, as Bill pointed out value in terms of the community environment.
[00:54:08] So less financially tangible points of view.
[00:54:11] Bill Schmarzo: And if I could just slightly add to that, I would say first thing that they need to do is to understand how does our organization create value for our constituents and stakeholders.
[00:54:22] Start there. Great conversation. What are our desired outcomes? What are the key decisions? How do we measure success? If we have that conversation, by the way, it isn’t unusual to have that conversation with the business stakeholders and they go I’m not exactly sure.
[00:54:37] John Thompson: I don’t know how that works.
[00:54:38] Bill Schmarzo: Yeah. So you need to find what are you trying to improve customer retention? You’re trying to increase market share. What are you trying to accomplish and why and how are you going to measure success? So the fact that the data science team is asking that question, because like John said, data science can solve a whole myriad of problems.
[00:54:54] It isn’t that it can’t solve. It can solve all kinds. That’s kind of the challenge. So understanding what problems we want to solve starts by understanding how does your organization create value. If you’re a hospital, like John said, reducing hospital acquired infections, reducing long term stay, whatever it might be.
[00:55:09] There are some clear goals. Processes initiatives around which organizations are trying to create value
[00:55:18] Dr Genevieve Hayes: So on that note, what is the single most important change our listeners could make tomorrow to accelerate their data science impact and results?
[00:55:28] John Thompson: I’ll go first. And it’s to take your data science teams and not merge them into operational teams, but to introduce the executives that are in charge of these areas and have them have an agreement that they’re going to work together. Start there.
[00:55:46] Bill Schmarzo: Start with how do you how does the organization create value? I mean understand that fundamentally ask those questions and keep asking until you find somebody in the organization who can say we’re trying to do this
[00:55:57] Mark Stouse: to which I would just only add, don’t forget the people are people and they all have egos and they all want to appear smarter and smarter and smarter. And so if you help them do that, you will be forever in there must have list, it’s a great truth that I have found if you want to kind of leverage bills construct, it’s the economies of ego.
[00:56:24] Bill Schmarzo: I like
[00:56:24] John Thompson: right, Mark, wrap this up. When’s your book coming out? What’s the title?
[00:56:28] Mark Stouse: It’s in July and I’ll be shot at dawn. But if I tell you the title, but so I interviewed several hundred fortune, 2000 CEOs and CFOs about how they see go to market. The changes that need to be made in go to market. The accountability for it all that kind of stuff. And so the purpose of this book really in 150, 160 pages is to say, Hey, they’re not all correct, but this is why they’re talking to you the way that they’re talking to you, and this is why they’re firing.
[00:57:05] People in go to market and particularly in B2B at an unprecedented rate. And you could, without too much deviation, do a search and replace on marketing and sales and replace it with data science and you’d get largely the same stuff. LinkedIn,
[00:57:25] Dr Genevieve Hayes: for listeners who want to get in contact with each of you, what can they do?
[00:57:29] John Thompson: LinkedIn. John Thompson. That’s where I’m at.
[00:57:32] Mark Stouse: Mark Stouse,
[00:57:34] Bill Schmarzo: And not only connect there, but we have conversations all the time. The three of us are part of an amazing community of people who have really bright by diverse perspectives. And we get into some really great conversations. So not only connect with us, but participate, jump in. Don’t be afraid.
[00:57:51] Dr Genevieve Hayes: And there you have it, another value packed episode to help you turn your data skills into serious clout, cash, and career freedom. If you found today’s episode useful and think others could benefit, please leave us a rating and review on your podcast platform of choice. That way we’ll be able to reach more data scientists just like you.
[00:58:11] Thanks for joining me today, Bill, Mark, and John.
[00:58:16] Mark Stouse: Great being with
[00:58:16] John Thompson: was fun.
[00:58:18] Dr Genevieve Hayes: And for those in the audience, thanks for listening. I’m Dr. Genevieve Hayes, and this has been value driven data science.
