Subscribe to The Digital Workplace
Join the journey to a better future of work
It’s easy to get carried away with the flashy lights. Implementing AI is anything but straightforward.
How can we use AI to solve business problems? And what should we do to make it work?
Sudha Jamthe is a technology futurist and a teacher of AI and AV business at Stanford Continuing Studies
Solving problems with data science
Sudha says AI isn’t a top-down or bottom-up decision—the idea to adopt AI can come from any role in the organization. AI, she adds, offers individuals a leadership opportunity to decide how to use AI in their work and how to ensure it adds value to the company.
More companies are experimenting with AI and finding ways to implement it to solve business problems. An example of this, Sudha recalls, is a project she worked on with her students as a part of her course with Capstone AI Labs. The business problem was to increase revenue by understanding customer behavior using AI. Tackling such a challenge required them to look at it from a data science perspective. Over four weeks, they analysed real industry data and offered the business three different options that helped them solve the problem.
Tech isn’t the answer
Leaders today look at technology as an answer to all of their problems. Sudha says technology is not a problem solver but an enabler if we know what we want to accomplish. “It is important to know what you want and break it down and then have a conversation with technology people, data scientists, and everybody else in the multidisciplinary team who is going to execute it.”
Sudha says 85% of AI pilots fail because leaders get too excited about implementing AI. They either approach implementation from the perspective of solving a business problem or use data to build predictive models they aren’t sure will pass for performance and accuracy. And they often change vendors or hire more data scientists instead of taking ownership and reiterating the problem.
Making AI work
Sudha says AI isn’t just about the data scientist and the business owner. Implementing AI requires a variety of skill sets. Sudha says while AI makes predictions, humans have to act on what the model tells. So the business problem needs to be designed for action. If not, the model simply won’t work.
“When you bring multiple teams into the conversation right at the stage when you’re taking the business problem and arriving at the problem statement, different people can contribute to it.”
12:36 – “Technology is not your answer. Technology is an enabler if you know what you want. So, it is important to know what you want and break it down and then have that conversation with technology people, data scientists, and everybody else in the multidisciplinary team who is going to execute it.”
21:32 – “When you bring multiple teams into the conversation right at the stage when you’re taking the business problem and arriving at the problem statement, different people can contribute to it.”
Welcome back to The Digital Workplace podcast. Today our guest is Sudha Jamthe. She is a technology futurist, and she teaches AI and AV business at Stanford Continuing Studies. Hi, Sudha, how are you?
Good, good, good. Thank you so much for having me, Neil. I’m so excited to talk to you.
Yeah, yeah. No, I’ve been actually tracking you for a long time trying to get your insight into stuff. You have a lot of things to teach us about AI and a lot of things to get into. But first, let’s start with our capture question, ‘prove your humanity here’. So Sudha, I want you to name one person that you’re close with and how you’d like to be more like them.
Whoa wow! I didn’t see that coming. Anybody who knows me knows that I enjoy meeting a lot of people and connecting with strangers. In fact, I once wrote this in my bio that I have to meet new people and learn something new. Otherwise, I’ll just combust with anything. So, that’s a lot of people. I actually would like to be more like my dog, Lulu.
Your dog. Okay.
The connection from just human to dog. I actually like the fact that she’s kind and empathetic. Because I think whatever we do, I think that’s an important thing. Even in my AI ponderings I wonder, whatever we teach AI can it really teach AI to be kind, not emulate to be empathetic and be kind. The other thing is, when I teach machine learning, I actually give the picture of Lulu and say, Lulu is a machine learning dog. She’s constantly learning. She knows when I close my laptop, whether I’m going to get up. When I’m doing this podcast, she’s not going to come and disturb because she notes my tone of voice when I’m talking publicly. And so, there’s a lot of learning and nuances that she knows. I wish I was as smart as her.
Wow, that’s great. And it’s a very human answer to say, your dog. I don’t think I can say anything more than that. For me, my wife’s grandfather just passed in the last few weeks. And he was a very interesting guy. And really someone I’ll use here just because, in a lot of ways he did all the things you weren’t supposed to do. He was very extravagant in his life, in a very good way. In his love, he would give a lot of love. Give a lot of money to people. Just do things that maybe at the start didn’t seem like common sense but they did them out of love and was willing to act in some people’s eyes in foolish ways. And I just thought that it was a neat thing. And it served him well and left a legacy around him, and that’s really cool. So that’s my answer.
Such an amazing person. And it’s so cool that you’re able to recall that aspect of him.
Yeah, for sure. So, this conversation we’re going to have today is about AI and business, which I think for most of our listeners, people who are leading digital companies, they see this as, ‘Yes, one day, I can see 10 years in the future, we’re going to be using AI’. But they don’t really know exactly what that looks like today. What they need to be doing today to get there and why they tend to fail along that way. So just give us a little bit of a background about your experience, why you’re a great person to talk about this topic of bringing AI into a business.
I’m a technology futurist. What that means is, I have many decades of experience working in the technology industry bringing technology to companies. And I looked back, and I found that I’m always working with the latest in tech. And tech is more like ‘Doctor Who’. It has a new form, and it regenerates every season. So, it keeps it exciting. So, now what I do is I’m always researching the latest in tech, but grounded in some kind of business operations to say, ‘Is this going to create new revenues? Is it going to create new markets? Is it going to create new jobs?’ And then I write about it, talk about it, teach about it. So that’s the place I’m coming from. And specifically, I look at the gap between business and technology. And I look at people who are somewhere in between. I don’t teach the data scientists. I don’t teach the person who’s just doing sales. I teach everybody in between to go innovate. And then the beauty is, everybody has a choice on how they want to do it based on what they become passionate about and bring transferable skills from the past. So, there’s so many jobs to be created, so many different ways of innovation that needs to happen. And it’s very personalized, individualized. And that’s the thing that I enjoy teaching.
Yeah. That’s amazing. When I think of CEOs wanting to get into this, I’m sure that they just kind of don’t know where to start. I think that’s the biggest thing. Maybe they can imagine, ‘AI can do this or I’m sure it could come there’. But if you try to go and you take a course online or something like that, then you might be inundated with, ‘Oh, this is trying to teach me how to program or how to actually use the AI. But I’m trying to sit here thinking about how to use it in my business’. So, where’s a good place for digital leaders to start thinking about AI?
I would say they can come to my site, Business School of AI, or I do, do a lot of talks and sessions, which is not just about money but to have conversation among peers. So, I think that would be a starting point. The thing you’ve said about how out there, there are courses which are teaching you about AI or learning how to do it from an execution standpoint, but not how to make it work for you. I think that’s a gap we are at. That’s an opportunity, I think, for an individual to take personal leadership. So, everybody who’s listening to this is out there doing some kind of job. So, somebody’s working in marketing, somebody is doing some kind of customer success management, somebody is doing customer service, product management, lot of different jobs.
The thing with AI is, right now it is not that it is starting from the top down and the CEO is saying, ‘Okay, now I’m going to make this an AI enabled business’, or it is not from bottom up and the engineer or somebody in product is saying, ‘Hey, we have created this new tech and now we want to deploy it and change our company’. In fact, there was some recent research I saw asking people, who is bringing AI to the company? And everybody said, me. And they ended up actually drawing this pie chart of different roles and it was evenly divided.
So, there’s a leadership opportunity for anybody who wants to bring AI. And I also think that’s a personal motivation that because it is new, cool, cutting edge, and kind of intriguing, it gives an opportunity to every individual to say, ‘Hey, how will I bring AI to my job role and make it add business impact for my company, thereby I improve my career?’ So, it looks selfish in some sense but that’s better motivation than thinking, ‘Hey, I’m going to take this company and I’m going to turn to an AI business into a mega billion-dollar business’, which might happen eventually. But that trigger might have been you who wants to change your department.
Yeah. I think that’s definitely where we want to get to, is we talk a lot about just the partnership between humans and technology. And when you think about teams to recognize that you’re working with colleagues now and to get to the point where everybody, every human on your team is thinking about, okay, how can I partner with this new colleague of AI and bring them in and then we can all work together. And to recognize, hey, this person’s or this technology’s best use cases aren’t being used yet. We haven’t really tapped them out yet. And technology could even say the same thing about humans. We can get a lot more out of them in other ways, too. So that’s the end goal that we wanted to get to. Give us a few examples of what some people are experimenting with, ways that they’ve successfully implemented AI into their companies.
So, one of the topics I teach is autonomous vehicle business and you use computer vision and go build how the car sees the road. That I would still say is in the realm of the data scientists. My students take that and say, ‘How do I build an AV business, which is to find data in the car, build digital twins in the car, make it connect to all kinds of other possibilities. So, it could be automotive changing transportation, or it could just be taking data in the car and connect to retail systems or healthcare systems. So, it’s several industries. So, I’m seeing a lot of possibilities of AI in the car that my students are working on.
Recently, I’m seeing two projects in a different course called Capstone AI lab that I offer, where students have got, let me pick one of them, not even two, let’s pick one. So, students work with a business partner with real industry data. And the business partner comes in and says, I have a business problem to solve. I want to bring AI to the company. I need to get started somewhere. And they bring real data. So, I have a team of students which are from different backgrounds, right? So, I have somebody from a data engineering background, somebody who’s a product manager, business manager, maybe customer success or something. It could be any kind of business role, right? Three or four people in a team, not more than that. They work and they say, Okay, what is the business problem?
So, one example we got was, they wanted to increase revenues by 10% by understanding customer behavior. Seems very straightforward and there is enough literature and courses which say AI can increase revenues for you, AI can help you understand customer behavior. All that stuff, right? They started with that. And I teach them how do you take the business problem and make it into a data science problem or a problem statement that you can give it to the data scientists to go build it. You don’t need to build it. These are not people who are coders, right? It was an amazing four weeks. My students went through this and said, Okay, how do we solve the problem? I could actually take the same data, understand my customer better, and increase my price and get more money from the same customers, or I can bring more customers similar to this profile if I understood them better, or I can actually just make the AI tell me if this customer is going to churn or stay back with me. That’s three different data science problems.
So that is how I would say, you could look at it as, hey, my job is to make sure customers don’t churn, or my job is to say what is the best channel where my customers are going to click and convert for marketing, or my job is responsible for increasing revenues from this particular business. And all three are possible directions they could go with this one business problem where they started. So that is the piece that my students have been working on and I’m very proud of it. So, they do four weeks of problem statement and after that they look at the data for two weeks and then they go fit the model and build this. This is like data science for the business user with NoCodeAI.
Sounds to me like the role of the business leader is to reinterpret the business problem in a language that data scientists can actually do something with, right?
I think that’s key. I think that we’re still in that role right now. Hopefully maybe there’s many people in the company that can start thinking in terms of that and saying, Hey, this problem that we have, the problem that we’re trying to get to, is this a technology problem? Is this a human problem? How can we go about solving that? So, what are some tricks or some helpful tips you can give to people to reorient their thinking to think, ‘Wow, this problem that I’m facing, I should seek out a technology solution for that’. Are there any key things you should look at to help make sure that that’s the right path to go down?
So, I’ve heard business users call me and say, ‘I’ve read this that AI can make predictions and do something for me’, right? Based on what we’ve seen for their industry and field. They would say, ‘Hey, I have a team of data scientists or I have a team of engineers who can learn data science. How do I go about that?’ They wouldn’t even think of this as a journey. They just think that there’s a technology answer for it. They would say, is there a data science model they should be using? Is there an AI algorithm? They expect technology to solve the problem. And so, one thing I would say is, whatever technology you use, and I’m a lover of technology, technology is not your answer. Technology is an enabler if you know what you want. So, it is important to know what you want and break it down and then have that conversation, not just with the technology person. You have to have the conversation with the data scientists, you have to have a conversation with everybody else in the multidisciplinary team who is going to execute it. So, there is the strategy piece, there’s a technology piece and the execution piece. You need all three to be aligned.
Sudha you mentioned to me earlier that, a stat you gave was 85% of AI pilots fail when they’re being out there. Why is it so high?
So, companies are getting very exuberant about bringing AI to their business. And so, they start in one of two ways. They basically say, hey, we have a lot of data and they say data trains the AI and let’s get started. And they try to go to the data scientists and say, I want you to solve this problem. The data scientist says, Okay, I will go fit a model and I will try to solve the problem. They get to a data science model, but it doesn’t quite solve the problem. It solves the problem but it’s not a performing model. So, the way data science works is, they build a predictive model. And so, it could be 99% certain, or it could be 20% certain. So, they end up getting stuck with a model. Maybe it’s 50-60% certain, but they are not sure when they take it to the customer whether it’s going to pass or fail. And so, they’re stuck in that place.
Or the other thing is, they actually don’t start with the data. They start with the business problem. The previous case that we were talking about. They don’t break it down to a proper data science problem. And they say, Hey, I want to reduce customer churn by 10%. Go to the data scientists. And the data scientist says, ‘What do you want me to do? Do you want me to give you a regression model? Do you want me to give a clustering model?’ They talk a different language. And there is this disconnect. I think the majority of the problems are stuck in that, in the latter case where there’s this disconnect of language. The thing is, they start off with a pilot, and then they feel like they’re not getting results and they’re kind of stuck. If it’s a vendor, they will say, ‘Oh, this vendor doesn’t know what they’re doing. Let me try another one’. Or if they try with the data science team, they will say, ‘Hey, we don’t have enough experience. We need to hire more data scientists.’ And I’m not seeing them stop and think, ‘I am the guardian of my business. I’m responsible for this. Why don’t I learn, and I do the right thing to translate the business problem to a data science problem and go through this as an iterative process.’
Yeah. That seems to be so central to being able to use any technology. I think we’ve gotten used to just throwing technology at different things without deeply understanding what the problem is we’re trying to solve, and we just use the tools that are there. And even like you said before, it’s even asking a better question that is going to allow the AI to succeed or that model to succeed in that. Because sometimes the problem that you have, like you said, maybe it’s a customer churn problem. But the more specific you can be with that problem and more specific you can really get at, the more useful the AI will be for that. Correct?
Yeah. And the thing with AI is, AI is predictive. So, AI might not be the answer that you need. Sometimes it just might be a very prescriptive problem. You just have to do some rule-based things and you can solve the problem. You don’t need AI for it. So, it doesn’t have to always start with, ‘Hey, I want to bring AI to the company’. It should be, ‘I want to solve a business problem. Looks like AI is a problem because so many people are talking about it like that’. And then you figure out if AI is the problem. And AI has a lot of potential. So, AI could be your solution.
So, it sounds to me, I’m trying to summarize some things here. First, you got to have a well-defined problem, something that you know exactly what you want out of this. Second, the answer to the problem needs to be predictive if you want to use AI. So, it has to be something where it’s going to suggest or predict the outcome that’s coming out of this. And third, I’m assuming you’re going to need a lot of data to be able to help the AI figure out one way or the other. Are there any other things you would add to that list of conditions that make AI a great solution?
I thought you summarized it very well. One thing I would add is you need a mix of people from the company. Variety of different skill sets. That is needed. It is not just the data scientist and the business owner who’s walking in and saying, ‘Hey, let’s use AI and solve this problem’. It is going to involve a lot of people. So, when we build a model, a model is just going to make a prediction. Right? Based on the prediction, they have to go take an action which means they have to execute. So, it is better to define the problem in such a way that it is actionable. So, what I tell my students in this ‘Data Science for Business Users’ course is, think of the AI and say, ‘Yes, it’s going to solve my problem’. Now when it solves the problem and it gives me a prediction and say, yes, this customer is going to churn or not churn; yes, this particular factor influences the customer; whatever it is telling you, right? If you are going to listen to it, what are you going to do with it?
It has to be actionable. If it is not actionable, that model is wasted. If it is actionable but you don’t have the power to do that in your company, that also is not helpful. So, you have to think about if this is actionable. So, if it is going to be creating more content because of which your customer is going to pay more money, do you have the resources? Do you have people who can help you to do that? If it is going to be checking a specific marketing channel, then do you have the support of the marketing department to act based on the recommendation your AI is going to give. Or if it says, okay, you can go to this new sales channel, then do you have the support? So, it could be different parts within the company.
I mean this is an age-old problem in companies where we are siloed, and data is sitting in different silos. And it’s the same thing. But at a fundamental level, you are going to work with multiple people who are going to get the job done. So, all that you said of taking the business problem, turning it into a data science problem, but make sure it’s actionable. Assume it’s actionable. And then you say, Okay, now I understand the root cause, and AI is going to give me a prediction and recommendation. Now how do I act on the recommendation? Who is going to help me let me get that team in. Not after the AI is built. So, that they can contribute to it. And they would bring their business acumen and say, you know what, the way you’re looking at this data is wrong. So, there’s a concept called feature engineering. When you look at the data, there might be more columns you have to add to the data. And based on that, the AI can give you a more accurate answer than if it did not have that piece of information.
So, I have one other student case where they are looking at a conference company and they want to predict what the ticket prices would be. And all the conferences have gone online because of COVID. And so, they’re planning for next year where it’s going to be in person. That’s what they’re hoping for. So now when they look at the past data and they’re making predictions, AI can say, Hey, this person signed up last time they have a propensity to sign. But there’s no guarantee that person will sign up and show up in person. Because there is one factor that they will not consider. If it is in the same location where the person is, chances are the person will show up because they find the value of your conference. Now, if it is a different location, because now you have global audience all signed up for this year online, now, if they have to travel, then they have to add the travel cost to it. Right? So, this is a very logical business thinking.
So, what the students have done is, they said, ‘Hey, look at this. But now the person might not come if they’re not in the local area’. So, the thinking was, let us add another column and say, yes or no, do they have a propensity to come? But then the AI doesn’t know why that business person added that column. The business reasoning behind that is lost. Instead, what they ended up doing is they added a column saying, what is the distance between where the person is located and the conference location. If it is zero, chances are that person is going to come if all other factors are fine. Right? Or if it is very far away and they are in Chennai and we are giving a conference here in San Jose, they’re not going to fly 10,000 miles. That is going to be expensive and going to add to the cost of the conference for them. So, that factor of ‘how far is the distance to the location’ is an important factor that comes from the person just thinking, ‘It makes business sense for us to target people who are local’. And so, if they didn’t have that factor, that would have been lost.
So, when you bring multiple teams into the conversation right at the stage when you’re taking the business problem and arriving at the problem statement, different people can contribute to it. And marketing would say, hey, I’m going to target local people. I’m not going to target people from outside. But then your list might have a whole bunch of people who came from outside. Now that data is useless, or you have to figure out a different strategy.
Sudha, I wasn’t prepared to be this excited about AI. It’s always a topic that was like, Oh, yeah, we should be doing that. We should be getting excited about it. But the clarity you brought to this conversation was really helpful. And just imagining where we can go with tools like these. And I think it goes beyond AI. There’s just kind of principles for using new technologies, whether we’re talking about other types of machine learning or IoT or blockchain. I think this framework that you’ve given us is super helpful. I would definitely recommend everyone kind of check you out. So, you mentioned a few things so far, but where should people go if they want to learn more about what you’re doing?
I think the easiest thing is you can go to my link. And if you want to check out my classes, I have two locations. One is called Business School of AI and the other is called Driverless World School. And otherwise, just find me on LinkedIn. And I love talking to people to help them figure out what do they want to do. Sometimes people are thinking about pivoting their careers into this space. Sometimes they’re just excited about the possibilities of what they can do in their job. And I see that we are all foot soldiers in companies, and we bring that together, then it’s good for us. We are growing in our career and we are doing a fun job. So that’s my recommendation. If you’re thinking of it either way, talk to me. I’m happy to help.
Excellent. Well, thanks so much for being on the show. We look forward to connecting with you again soon.
Thank you so much, Neil. This was such a pleasure. Thanks.