“The first step to really making the best use of AI is to understand what it really is and how it applies to the context of your business. Because every business is different, and AI may or may not be applicable to your business.” 

Intro:
Welcome to the Agile Digital Transformation Podcast, where we explore different aspects of digital transformation and digital experience with your host, Tim Butara, content and community manager at Agiledrop. 

Tim Butara: Hello, everyone. Thanks for tuning in. I'm joined today by Kavita Ganesan, trusted AI advisor, founder of Opinosis Analytics, and author of The Business Case for AI, who has over 15 years of experience in artificial intelligence. And in today's episode, we'll be talking about strategies and best practices for introducing AI to your business. 

Kavita Ganesan: Thank you, Tim. Thank you for having me here. 

Tim Butara: So if you're okay with the intro, I suggest we jump straight into our topic for today. 

Kavita Ganesan: Yes. 

Tim Butara: To kind of get a better idea of where we're at in terms of AI evolution and adoption. Let me ask you, so what is the current state of AI evolution and adoption in the world? 

Kavita Ganesan: Sure. Yeah. So let me start off by setting the context for the audience. So AI is basically trying to mimic human intelligence within a computer. And this is in the form of intelligent software. And people often think of AI as one particular thing, but really it's a coming together of sub disciplines. 

Like NLP is to mimic language understanding in a computer. Computer vision is to mimic human vision capabilities within a computer. Then machine learning is to mimic how humans make decisions within a computer. But with machine learning, you use large amounts of data to learn how to make those decisions. So it's a coming together of different disciplines to solve specific use cases. So, for example, for self-driving cars, you use a lot of computer vision to help the car see and navigate themselves on the road. 

And in the early days, AI was a very research intensive field. The techniques that were proposed in the 1940s, 1950s could not be easily used to solve practical business problems because the computation power just was not there. But over time, this actually changed. Like around 2011, cloud computing actually became a really popular thing. So you could actually rent cloud resources and run algorithms pretty cheaply. And if you would have done that many years ago, you'd need a supercomputer, which is really expensive. 

And around 2011, also, big data really started to take off. So companies realized, hey, I can use this data and all these computational resources to use AI and solve practical business problems. So that's when AI really started to take off within business applications. But the adoption of AI itself has been pretty slow. And that's because initially the tools to develop and deploy these AI solutions were not very refined. But now things have changed. A lot more tools, a lot more ways to easily go from an idea to deployment of AI solutions. And you'll see a lot of this AI adoption in large tech companies, in financial companies, and also within startups. 

Tim Butara: So with this in mind, with all the changes and kind of the new reality for AI now, and with the tooling, I assume, becoming better, improving, what are the best practices, what are the tactics for succeeding with AI, and maybe how do these differ between industries, between companies? 

Kavita Ganesan: Yeah. So I think the first step to really making the best use of AI is to understand what it really is and how it applies to the context of your business. Because every business is different, and AI may or may not be applicable to your business. So understanding what the sub disciplines are, how it relates to the problems you have in your company, so that will help you see how you can use this tool to solve problems in your company. 

So understanding is one of the first things you want to do. And I've seen a lot of cases where people just try to apply AI for the sake of using AI and it goes nowhere. And then they get disappointed and say that AI is not a really good technology to use, but really, AI is suitable for very specific problems. And these are usually high volume problems, complex decision making problems. And you usually have to be generating a lot of data from the manual process to start with. 

Another thing for succeeding with AI is to understand what is the return on investment you're getting by using AI versus a simpler method, or even manually. So maybe a task is taking you ten days to accomplish, but by using AI, you can do it within minutes or seconds. So understanding the ROI will help you really get value from the investment. Because AI is not cheap. It's not just about simple software engineering. It's, once you develop the model, you'll have to maintain the model. You'll have to improve the model over time, and the model can dip in its performance. So it's a long term commitment. So you want to ensure that the ROI is there. 

And the third thing is, if you're looking for the long term adoption of AI, then you need to think about addressing the foundational building blocks. So, for example, AI systems are heavily data dependent. You need a lot of data to learn how to make decisions. So your company may not be already collecting data aggressively, or for a particular initiative, you don't have the data. So jumping right into AI in those circumstances is going to set you back.

So what you want to do is identify those types of foundational pieces, like data, cultural elements. Like, you may have some fear in your company about using AI. You may not have the infrastructure to develop and deploy models. So by addressing those challenges, and you can also concurrently pilot AI at the same time, but addressing those challenges will help you with a long term adoption of AI. 

Tim Butara: Yeah, I think that we’re going to talk about challenges and hesitations a little bit more in a later part of the episode, but right now it sounds to me like you don't really need to have super in-depth technical skills to make good use of AIs. I mean, from what we discussed now, it's more a combination of some technical skills, not extremely tech savvy, and some business acumen and kind of this adoption, this mindset. Yes. 

Kavita Ganesan: So it's more understanding it at a very high level. Like anything you want to apply in your business, you'll have to understand it at a high level. So you need to know what it is. And once you know what it is, what are the risks, what are the costs, and how to prepare for it, then you'll be able to see how best to apply it. And when it comes to the implementation of it, you can buy off-the-shelf solutions, and you may need a software engineer to integrate it within your workflow. 

If you want to build from scratch, that's also possible. You don't have to do it yourself. You can outsource it to companies that do AI development. So there are many ways to get it implemented without being technically savvy. But on the high level, you do have to understand how AI is relevant to your business and what it is. 

Tim Butara: Yeah. This ties back to your previous point about having to know the ROI, if you want to determine, if you want to be able to determine if the investment is actually going to pay you off, both in terms of finances and in terms of time saved. 

Kavita Ganesan: Yes. And a lot of times you'll find that using just clean software engineering will do it, you don't need AI, or maybe doing it manually is a cheaper solution than using AI. So you'll have to weigh all of those before deciding, hey, AI is the right solution for me and I'm going to explore its use. 

And also, way to get value from AI solutions that are already deployed is to– actually, before deployment, is to test it. So even if you're buying off-the-shelf solutions, you want to test it on your data for your particular use case. And it might be that those off-the-shelf solutions may not work. So you'll have to custom build. So testing is the key here. 

Tim Butara: It really sounds like there's a lot of potential issues and a lot that can go wrong. And you have to be really, I guess, thorough and meticulous in how you adopt AI, how you proceed with your adoption, if you're going to scale based on that, how you're going to scale, as you just said. Yes, as I just confirmed. 

Kavita Ganesan: Yeah. You have to be very strategic about AI because it's not a simple software engineering. 

Tim Butara: So in this context, I mentioned just a few minutes ago that we would talk more in depth about challenges. So what have you seen or experienced to be the main challenges or concerns or hesitations of implementing AI and how should companies, businesses approach resolving and addressing these? 

Kavita Ganesan: Yeah, so one of the challenges that I've seen is, there is a lot of fear about AI as a technology. So some people think that they are going to contribute to the bad of humanity because of things like what they've seen with Facebook, where algorithms can cause more harm than do good. And this is not just with your customers, but also internally with employees. There may be fears, there may be confusion. They don't know what you're going to do with this technology. 

So one way to address it is through education and also taking a stance within the company. So what are you going to do with all this AI if you were to use it, what you will and will not do with AI applications, like, how would you address the issues of bias, the algorithmic challenges you've seen with companies like Facebook. So, through education, I think you can address a lot of fears and people may be more empowered to consider this as a technology they would use to solve their business problems. Otherwise, there may be hesitation to even think about AI as a solution. 

Tim Butara: Yeah, I mean, there's a lot of negative stigma associated with it, on one end because of Sci-Fi and movies and stuff like that. And on the other hand, the more concrete and more kind of tangible aspects of it, such as, you mentioned Facebook and algorithms and the potential misuse of the data that's collected and analyzed, which also ties to the bias. I guess similarly to a lot of new breakthrough technology. Right. It has the potential to either do a lot of good or if misused, if exploited, it can lead to a lot of bad things as well. 

Kavita Ganesan: A lot of bad things. And a lot of it is a function of human behavior rather than the AI itself. So we have to take a stand as a company on what we will and will not do. 

Tim Butara: I think most of it is actually a function of the human input. Right. Even if one aspect is the people using and managing the AI, but it begins with humans creating AI. I guess we're not at the point yet where AI is creating AI. I guess that would be a sound of alarm. 

Kavita Ganesan: Yes, we are not at that point yet, but in many years we may get there.

Tim Butara: In this context, it's also, I guess, one of the hot recent topics in the field of AI has also been ethical AI and this concept of trust-based AI. I guess that these are super important for kind of addressing and tackling these concerns and hesitations. 

Kavita Ganesan: Yes, definitely. And another challenge that I would say with the adoption of AI is some of the building blocks are not there. Like, for example, if you want to get started with an initiative, the data for that initiative may not already be there. So then what do you do? So what you can do in those instances is to have an AI roadmap. So instead of going straight into implementation, have a few AI ideas at hand, and then think about what data is needed to support those initiatives and how you can acquire that data. So maybe starting manually may be a good solution, because manually generated data is always good data because it's generated by humans. And also that will give you a better handle of the problems that you have at hand. 

Tim Butara: Really good piece of advice. I I think that listeners will really appreciate this one. 

Kavita Ganesan: Yes. And it will help you understand what exactly is the problem you're trying to solve, what's your input and what's your output? And then you can get AI system to replicate that. So that's one way of starting with AI without the data. Another way is to, like, you may have to crowdsource this because there's no way to collect it internally. It's very dependent on the problem itself. But you want to brainstorm what are the AI applications you can develop, and then start the data collection early on. 

Tim Butara: It only makes sense, right? Even in a constantly changing, uncertain world filled with disruptions, you still have to plan things out and think things through before jumping headfirst into everything. Especially if it's new technology that hasn't really seen massive adoption. Although more and more people are starting to realize that they're basically interacting with some forms of AI on a daily basis, if they use social media. We talked about algorithms, but it's still not as widely implemented, at least not in its fullest potential, I guess. 

Kavita Ganesan: Yes, and people are struggling at different points either during data collections, during deployment, on which problems they're applying AI to. So they're struggling at different areas, really. 

Tim Butara: And next question. We already dipped our toes a little bit into this one. So how do you think the field of AI will continue to evolve in the future? What are we likely to see? Maybe in the coming months, in the coming years, in ten years? 

Kavita Ganesan: Yeah. So I think there are few broad areas, and one area I think is in making AI systems more explainable. So today AI systems are very black box in nature. So if I predict that you have a high risk of a certain type of cancer, doctors currently don't know why that AI made such predictions. So if the AI can explain why it arrived at those conclusions, then that's going to help the physician confirm their diagnosis or confirm that you do indeed have a high risk of cancer. 

So that's one very large area that's up and coming. There's a lot of research going on. And then the second area is in common sense reasoning. So what we talked about earlier. So AI systems today are very task oriented. So you train it to accomplish a specific task, it can very well do that. But if you ask it, the same AI system, if you ask a question that you will ask a two year old, it may not be able to answer that. So common sense reasoning is very much lacking, but there is progress in that area and you're going to see more and more of that over the next decade. But it's going to be slow. It's not going to be as intelligent as humans are at this time. But we will get there, I guess, at some point. 

Tim Butara: I think that's almost inevitable. Yeah. 

Kavita Ganesan: Yes. And the third area, I think, is relying less on data. So right now, AI systems require large amounts of data to learn a specific task. So there is a lot of research going on, on how can I use less data to learn a similar task that I previously would have learned using large amounts of data? And this is called, like, one-shot learning, if you've heard of a few-shot learning.Yeah, this is a new area of research that's up and coming. And I think this also made great strides because not all companies can afford to collect large amounts of data, so they want to be able to start with limited amounts of data. 

Tim Butara: That's very interesting because it almost seems to contradict one of the main points that we've discussed so far, that is that AI is heavily dependent, or rather, a successful implementation of AI is heavily dependent on data. So right now I really can't imagine what that would look like. But I guess that's why we're talking about this in the part of the episode when we're talking about future potential. 

Kavita Ganesan: Yes, these are the future potential. And I think it will become a tool that we can use down the road.

Tim Butara: As for the other predictions, I think especially the first one is different from this one in the sense that it really aligns with what we talked about previously. We talked about trust and ethics. And explainability to me, in this context, it's almost equated with transparency, right, having insight into how the particular AI was able to infer that insight. 

Kavita Ganesan: And that's one way we can build trust between humans and algorithms through these types of evidence from the algorithms themselves. 

Tim Butara: And, excellent point because my next point about that was that, you mentioned that that would allow them, the actual humans, to make more accurate diagnosis. So this kind of reinforces one of the observations that we made several times during this podcast in different episodes. That the trick to leveraging, to making good use of AI is to make it work well in tandem with humans. Not as a replacement for humans, not as in any kind of other sense, but just enhancing the work that humans do and facilitate. 

Kavita Ganesan: AI systems, they take over roles that we may not already want to do. We may be making a lot of mistakes, but AI systems don't get tired, they don't get distracted. So they can repeatedly do things at the same level of quality, I guess, so to say. So that's one area that we really need to be thinking about. We may not already be wanting to do those jobs, so why not offload it to an AI system? And we can be the data generators for the systems. We can be the supervisors for the systems. We can help improve those systems. So we still need humans, very much need humans in the loop. 

Tim Butara: That's really good to hear. So no fear of AI overlords eliminating us, replacing us. We're good on that note? 

Kavita Ganesan: Yes, for now, we are good on that note. For sure. Yeah. 

Tim Butara: Okay. Awesome. Now we've almost reached the end of this very fascinating and interesting discussion, Kavita and I just want to ask you before we finish, what would be your top words of advice for business leaders who maybe are having trouble with their AI implementation or aren't seeing success with it… just, yeah? 

Kavita Ganesan: Yeah, it depends on where you're having the problem. So let's say you've already deployed your AI system, but you don't know what it's doing, if it's improving your business metrics, you don't know what value it's creating for you. So in this case, I would say go back to your metrics. So the first thing you want to check is your model metrics. So is your model performing adequately in production, and how was it performing during development? 

So you want to make sure that your model is built to standard. So once you know that you have a good model, now you want to see what is your goal of deploying the AI system? Was it to improve productivity? So then you should be tracking maybe the time reduction in completing a specific task. And what is AI doing in delivering on those calls? Is it improving? Is it staying the same, or is it deteriorating? 

So if it's improving, then it's on the right path. Maybe you just need some improvements to improve the model so it further gives you more productivity improvements. Or if it’s staying stagnant, why? We want to investigate, why is it stagnant? Is it because the workflow is too complicated that people have to go through additional steps to consume the AI output? So that's going to make your workflow– add friction to your workflow, essentially. 

Tim Butara: So that would be the inverse of what you would want to achieve by implementing AI, right? Instead of your job getting easier, it would get harder. 

Kavita Ganesan: Harder. And I've seen this happening where you develop this really cool AI solution, but you add friction in people's workflow, then that doesn't really help. Yeah. And if your metric is getting worse, there is no improvement but deterioration, then you want to go back to the drawing board and figure out, hey, what's going on here? Did we develop a bad model or did we not deploy it correctly? Is in the wrong place that it's being used? 

So you really want to investigate into those issues using your business success metrics as your North Star essentially. And you also want to know what your user thinks, like, the consumers of the AI output. So what do they think? Do they perceive this as a long term solution? If they're not happy with it, they're not going to use it later down the road. They may use it during testing, but after it's deployed, they may just go back to the old way of doing things. 

So you want to ensure that the consumers are happy and they may also spot other issues, like they may say, hey, I like this AI solution, but the workflow is not good for us, the way it's being deployed is not good for us. Or they may see model issues that we may have missed during model evaluation. So all these three components need to come together. The model success, business success, and user success. And I discussed a lot of this in chapter 14 of my book. 

Tim Butara: Actually, since we're already finishing, can you tell us more about the book? More about where listeners could potentially reach you, order the book, learn more about you?

Kavita Ganesan: Sure. The book is about how companies, essentially leaders, domain experts, anybody who wants to use AI, how they can get started with it in their business. So it starts with a very high level of what is AI, so what I talked about earlier. And then it dives into what are some of the myths of AI, your use cases where you can use AI within your business and how do you find opportunities. 

So you're not just going for the wrong software automation opportunities, but you're going for real AI opportunities, and then things like whether you should build AI systems or buy. So it walks you through the whole thing for how you can apply AI within your business. And you can find more details about this on my website. It's Kavita, K-A-V-I-T-A dash Ganesan dot com. So just go to my website and you'll see some information about the book. 

Tim Butara: Awesome. I'll make sure to link both the website and a direct link to where they can order the book so that they have everything at the palm of their hands. Because it certainly seems like the book is the best way to upgrade the knowledge and insights you gained from this episode and kind of start your AI journey for real. 

Kavita Ganesan: Yeah, that's correct. And it will be especially helpful if you're new to AI and you don't want to be intimidated by all the hype around. So, yeah, that's a good way to get started. 

Tim Butara: Well, Kavita, as I already said, this was a fantastic conversation. I think we covered a lot of the really important aspects, the positive aspects, the negative aspects, the best practices. I think that this will definitely be very enjoyable episode for all of our listeners as well. So thanks for being our guest today. 

Kavita Ganesan: Yeah, thank you for having me. I really enjoyed this. 

Tim Butara: Awesome. Likewise. Have a great day. 

Kavita Ganesan: Thank you. 

Tim Butara: And to our listeners, that's all for this episode. Have a great day everyone, and stay safe. 

Outro:
Thanks for tuning in. If you'd like to check out our other episodes, you can find all of them at agiledrop.com/podcast as well as on all the most popular podcasting platforms. Make sure to subscribe so you don't miss any new episodes, and don't forget to share the podcast with your friends and colleagues.