The value of creative AI

Podcast Transcript

Kimberly Boyd: As companies look to future-proof their businesses, there's certainly a lot of hype around artificial intelligence and machine learning. Companies are looking to AI and ML to deliver better customer experience, to improve productivity, sometimes in the hopes of cutting costs or simply to keep up with the competition. But do companies truly understand the potential of AI and ML for their non-operational business strategies? Beyond the popular use of replacing tedious and repetitive tasks with automation, lie more creative and strategic opportunities in solving business problems and improving product innovation.

Welcome to Pragmatism in Practice, a podcast from ThoughtWorks, where we share stories of practical approaches to becoming a modern digital business. I'm your host Kimberly Boyd. Today I'm joined by Jarno Kartela. Jarno joined the ThoughtWorks family earlier this year through the acquisition of Fourkind, a team of data scientists who blend machine learning, data science, design, and engineering to help clients lower the perceived risk of adopting new approaches to artificial intelligence. Welcome, Jarno.. Where do you begin with when you want to help organizations and leaders understand how to use technology, how to use AI and ML to create value? What's the best way to begin that process?

Jarno Kartela: I think the best way is to find the right narrative in doing so, because a lot of these projects are change management projects by nature. We need to find the right narrative and we need to find the right problem to solve and then we can work back from that to technology. I see a lot of companies working on AI and working on machine learning for the sake of the technology itself. A lot of those projects are having a hard time because it's much more motivational for the technology folks as well to work on a business problem. I usually start with the business stakeholders and trying to understand what's the problem we want to solve and what's the narrative of doing so, and does it even require technology? If it does, how can we match it in the right way? I'm looking for that right problem to solve.

Kimberly: Yes. I'd love to dig in a little more there. When you say the right problem, right… I'm sure organizations have a lot of challenging problems that could benefit from technology. When we say the right problem, what exactly does that mean? What should organizations be looking for?

Jarno: I think a good way of thinking about it is to think that what are the different things we do as a business and what are the different things we do as an organization within itself? What are the things we do in a sort of maybe mundane, but like repetitive fashion and what other things we do that really make the core of our business?

In terms of strategy, in terms of tactics, in terms of creativity, what are the processes and things and ways of working that we really need to delve into so that we create competitive advantage? From that, I think you start to see the value continuum. You can use machine learning, you can use applied data science, you can use data in all bits of the value chain and all bits of the value chain in terms of what the business is doing, but also what the organization itself is doing.

We don't need to stick to doing the most obvious things in machine learning such as personalization, and dynamic pricing and all these things that we see constantly in the market. We can also use machine learning and AI to augment strategic planning, we can use it to augment decision making, we can use it to basically improve all of these creative processes that we are not accustomed to be using computers and machines and so on as co-workers.

I think a lot of merit goes to thinking about the entire value chain and then thinking about the entire value chain of how the organization actually works and what are the different tasks that we need to do to keep the organization rolling. All of those, I think, are good areas for solving and basically adapting AI and machine learning into. Then you just need to be really smart about what are the things that will get us most return on investment? What are the things that are feasible? What are the things that the organization actually wants to do? When you take all of those things into account, you start to build this narrative that this may be the right direction that we should go towards in taking advantage of all of this emerging technology.

Kimberly: It sounds like it might even be helpful if organizations did a exercise before even picking up problem space to say, "Hey, let's map out what fundamentally makes up our business. What are the tasks? What are our core competencies, our core differentiators," before they even begin to tackle using AI, whether it's for more of the automated or more of the creative?

Jarno, I have to imagine, when you're talking about areas of the business that are core to value or core to competitive advantage, or those things that are viewed as really strategic or creative, people probably get a little nervous about saying, "Hey, let's automate these or let's enhance these with technology," because they probably say, "Wow, no, no, that can't possibly help. That's something that's a people driven activity." How do you get them to overcome that initial objection and start to consider using technology in more of a creative capacity?

Jarno: Yes, I think it really boils down to you have to make the people actually want the thing that you're doing. Just trying to push machine learning and AI to the agenda of creative people, it's a difficult road to take. You have to start with the co-planning, co-designing, co-creating. What's the problem? What are the scenarios and things and systems we could throw at it? What are the pros and cons of each? Together with all of those stakeholders, and with that team and with those creatives come up with, "Okay, maybe we should actually try something like machine learning as a co-worker," but not go with the solution first. I think it's a dead end.

You have to come up with a process that allows for ideas, that allows for co-creation, that allows for brainstorming about new concepts of tackling whatever the business problem may be. If it seems reasonable, go with something that takes emerging technology into it. Most likely that is the case for many problems, but you still have to do that homework, and you still have to go down that route because you want to have the people backing that idea. Because if you don't, the technology will not solve that problem.

Kimberly: If it could, I think we will solve most of the world's challenges if we could figure out how technology could overcome people not wanting to change. Is it fair to say that if people are coming to the table with a point of view of saying, "I have this problem and I know I want to use AI, I know I want to use technology to solve it," that's actually the wrong approach? You need to actually just start with more of that, "Let's explore all the possibilities, let's be willing to get in that creative space first, before having a definitive point of view," because it sounds like people are set up to fail if they come into it with that approach.

Jarno: I think so. It's also a really natural way for humans and for experts and everyone to think in that fashion because if you think about what are the tools that organizations use to tackle problems, they tend to organize themselves differently. They tend to create goals and KPIs differently, but like most of the time, you see this problem between, "We need to get this business," and technology alignment.

What if you thought about the problem first and then thought about how is it that you should organize around it, I think it would solve a lot of problems. I think it's the same in emerging technology, it's the same in AI, that you cannot start with the solution and then think, "How should I go about creating the narrative on how does this fit to a business problem?"

We really have to start with the business problem. That creates a chasm I think between business and technology, because there's this profound problem that the business usually does not trust the technology folks to solve the business problem. The technology folks feel that business doesn't understand the set of possibilities with technology. There's this lack of trust between two parties. It doesn't really help if you go into that mix and say that this is the year that we're going to invest in AI and machine learning, because yet again, I think it's only going to widen the gap between business and technology. You have to go with the business problem first.

Kimberly: I wish we could use AI technology to close that trust gap that exists. That would be great if that was possible. We've talked just a little bit, I think more theoretically around how organizations should approach having a problem in mind and having an open mind about how to apply technology. I know you've worked with a lot of organizations and actually done this and actually used AI in creative or decision-making scenarios. Can you bring that to life for us a little bit and talk us through where you've seen some areas of success with organizations who have applied that?

Jarno: I think three come to mind as really different examples. One is obviously organizations that maybe five years ago set that they want to be data-driven, but then they came to the realization that, "Actually, we don't necessarily want to be data-driven, because like data can only tell us so much about what's happening in the world, it can only tell us so much about what's happening in our customer base, and most notably it can only tell so much about the future."

Then organizations start to think, "What is the way that I should use all of this new technology to improve how I do strategic planning, how I do simulation, how I do all of this horizon 2, horizon 3 stuff that I need to do as a CXO?" Then we worked on these types of problems most lately in the area of sustainability in which we, together with the clients, started to think, "What are the best scenarios to go towards zero CO2 impacts and what are the mechanisms of doing so?"

In order to succeed in that problem, we need to do three things. Number one is, we need to model how the business works. We need to model it as a dynamical system of things, because if you think about organizations, they are so complex that if you change one bit of it, you cannot trace the effects of that in isolation of the entire company. You have to model the system in some way. There are different ways of doing that. We went with system dynamics. You can do agent-based modeling as well, there's a number of things you can do. The core idea is, you need to model how the business works so that when you change bits of it, you can see what's the impact on the total scenario. That's one, you have to model the business in a conceptual way.

The second one is, obviously, you have to have some idea what's going to happen in the future. You're going to have to have assumptions, research, data points, just talking about what the future may hold and writing all of those assumptions down as data points, as pieces of information and so on. You have to have the common campfire around which you gather ideas about the future. That's the second one I think you need to do.

The third one is that, you need to bring all of these to life through simulation. You need to bring all of these to life through scenario modeling. Then ultimately, if you are able to do that, you can answer the strategic question of, "How should we go towards zero CO2 impact or how should we go towards whatever the strategy is in the most robust, in the most impactful way out of all of these 50 scenarios that we've thought about?"

That's possible today. I think that's really cool, but that's really underused in the market, let's say, of applied data science, applied machine learning and so on because we're not used to doing that in the decision science umbrella of things, because we're scared of using machine learning and artificial intelligence when we go to strategy, when we go to the CXO level of making decisions. That's one, I think. Scenario modeling simulations that's certainly something that's going to hit the mainstream in five years, and the overall umbrella decision science is certainly going to be the next big thing, I think.

A second one, from a completely different perspective, was the case we did for Mackmyra which is the AI created whisky. In that case, we were not trying to find the best scenario.

We're not trying to predict what's the most optimal outcome. We try to create a model, a machine, a thing, a concept that augments the creative experts. When I say augments, I mean that instead of trying to pinpoint what's the right answer, let's try to create a model that can mimic creativity in a way. Let's try to understand how you make whiskey which was fun by the way to work out.

Kimberly: Yes, it sounds like it.

Jarno: Because you get to do tasting and sampling and so on, which was not too bad. Let's try to understand how you make whiskey. Great. Let's try to understand what things, what pieces of data and information does the creative expert use to create the perfect whiskey. Then let's try to figure out if we can use machine learning, if we can use generative modeling, if can use this idea of computational creativity to create the co-worker for that creative experts.

That's exactly what we ended up doing. We created this model that was able to take these data points from making whiskey. Previous recipes, customer ratings, online reviews and all of the ingredients that the company has, and it was able to produce this list of, "These are recipes that we think you should try out. These are recipes that are somewhat like what you've done before, but really different in the space of things that you've done before." They are something that the brand could have done, but they are really far from any previous recipe that they have done so far. In that way, it's trying to push the creative expert out of their comfort zone.

When we did that, we actually-- I think we created 15 different recipes, and then from those we selected 5 for sampling. Then out of those five, the one that we have on store shelves now is labeled Intelligens, which is the AI creative whiskey. There was no human involved in the process of somehow altering the recipe, but there was a human involved in the process in selecting the best one.

We created this generative mind that was able to create recipes, and not trying to find the right answer but to push, where we go with whiskey making? Where could we go in terms of what the next product could be? I think that's another area that's certainly really interesting. It's not decision science. It's like this notion of computational creativity. It's this notion of creative AI and how to use machines to generate things that we have not thought about before. It's trying to go around that creative experts block on creating new things.

Kimberly: Yes. I'm probably going to grossly simplify what you've just described here, but what it sounds to me is, it's a technology enabled co-worker that helps you brainstorm. It helps you come up with new opportunities to try out faster. I'm sure a master distiller could go in a room, and it would probably take them a couple years to come up with the variety of all those recipes. Whereas with their enhanced co-worker, they can get to that short set and then focus on doing what they do best is tasting and trying the recipes and then deciding which one to go with. Is that an accurate way to describe it?

Jarno: Yes, absolutely. You can immediately think what are all of the areas that you can use this approach in. You can it in architecture, you can use it in writing, you can use it in marketing copies, you can use it in virtually anything. You can use it to imagine what a scenic city could look like in 20 years. You can use it to imagine all of these things.

In that respect, it's the complete opposite of decision science, which is trying to augment how we do the most robust, the best decision and how to do that in a science and data backed way. That's what I find interesting because obviously the third one, the third example of how this comes to life is the self-optimizing products and the dynamic pricings and the reinforcement learning things, which we can do in really novel ways today. I think what we should talk about more is creativity, is strategic thinking, is decision-making because I think we're losing out a lot on the potential of applied data science if we only use it to automate things.

Kimberly: Yes. Hearing you talk, I'm like, "Why hasn't this taken off?" Because I feel like pretty much every industry or every function could benefit from quickly being exposed to variety of new ideas or options in whatever field they're in. I heard you mention marketing and marketing copy. As a marketer, I want you to build me a model so I can get some help generating campaigns for us. In all seriousness, I would love to understand why do you think it hasn't taken off in fields of architecture, retail, all the ones you just previously named it, because it seems like there's a huge opportunity space there for people who want to jump on the decision science train.

Jarno: I think it takes a lot of courage from the company and its culture to do so. I think that's the fundamental thing. I'm not saying that companies are not courageous enough. I'm saying that it's a leap of faith to do so. It was certainly a leap of faith for the whiskey company to be the first one in the world and say that, "Okay, let's challenge these hundreds of years old industry. Let's say that machines can do whiskey as well." That's a crazy idea. It's a leap of faith to do that.

It takes a lot of courage. It takes a phenomenal culture of the organization to say that, "We want to disrupt how this thing is working. We want to disrupt the entire industry." I think it's going to be more abundant as time goes on because we are going to see a lot of examples. It's going to be more mainstream, but I think now it's the time for the early adopters. It's the time for the first ones on the markets to reap the benefits because a lot of these companies are working on, "How can we push forward our optimized processes and optimized systems and things," and that's all great. I think that's obviously something you need to do.

To get to outsized value, to get to these things that your neighbor and your competitor is not doing, you have to take the leap of faith. I think this has been true for decades now, in terms of technology, in terms of adopting a new way of working, adopting a new way of managing the business and so on, it's always a leap of faith. What I think is the leap of faith today it's not a leap of faith anymore in terms of technology. It's a leap of faith in terms of, "Are we courageous enough to try that?" I think the ones that are we'll certainly reap the benefits.

Kimberly: Yes, right? Everyone who's willing to take a risk, I think earlier than others, it's typically where you see that outsize value that you talked about. Say I'm an individual in an organization that is willing to take more of a gamble and step out, apply technology to more of the creative or decisions scenarios, are there any new skill sets or capabilities that I need as an individual to appropriately work with technology in that capacity?

Jarno: I think the overall understanding of what data is, is rather important today. If you think about the roles and how we've gone about managing technology in businesses, I think it came up, firstly, we had IT, that was responsible for all of these basic systems and ERP and so on, then the e-commerce wave and the digitalization wave came and we created the chief digital officers, the CDOs that are now really prominent as the bridge between business and technology.

What I think the next wave is, I think it's certainly data because we have all of these roles that are-- basically, they look like something that's highly that is highly valued in terms of how we manage technology, how we manage e-commerce, how we manage digital growth, but they are not necessarily equipped to manage what's the data transformation that the company will inevitably will absolutely face in the upcoming 5 to 10 years.

It's not like you have to understand AI and you have to understand applied data science. I think understanding what data is, how can you use it? Where does it come from? How does it work and how can you apply that to create business value? I think that's going to be a skill that's going to be required from a lot of folks in the upcoming years, because organizations are not equipped to handle that in their current roles and competencies, because they have evolved from digitalization, they have evolved from information technology, but they have not evolved from, "How can we use data as a business asset?" There's a lot of grounds to cover in order to get to AI. I think in order to get to AI, you have to get to data first and understand the value prop of that.

Kimberly: Everyone needs to fundamentally improve their data acumen and then maybe start to get in the habit of saying, "Is this a scenario where we could be using data or we could be enhancing our ways of thinking with it?" That's a change. That's a shift in mindset, which like you mentioned earlier is the hardest part. It's not actually building the models or any of that. It's getting people to think differently.

Jarno: Yes, absolutely. I think another point is that it's tricky for us as humans to let go of control. It's tricky for us to say that, "Let's use technology as a co-worker." It's difficult to say that, "I'm not in charge of whatever campaign or price or piece of content is laying on my websites, I want the computer to handle it." I think that was a problem 5 years ago or 10 years ago that we were like, "Well, no, we know what's the right price to pay for a specific product."

Now we've let go of that and we're like, "Okay, bring on dynamic pricing, bring on all of this modern technology to handle all of those customer interactions in the digital platform," but we're not ready to let go of control when it comes to strategic and creative tasks. I think that's totally fine. We need a lot of examples, we need to build trust in the people who are creating these systems, we need to build trust in, "How are we going about to use this technology in our day to day?"

I think what will help quite a bit in building that trust to technology is understanding what data actually is? How is it created? What are the pros and cons? When can you use it? When can you not? When should you actually do something entirely else and so on, because when you understand that, then I think it's easy to understand that, "Okay, let's have this like generative model help me in my creative task because I understand what type of data is being fed into it. I understand the inner workings of it on how from that it comes about creating new concepts and things and products and recipes and whatnot." Building trust to technology, from my perspective, these days, is a lot about building trust in data, and building trust in data requires competencies of understanding what that is about.

Kimberly: Just like you would need to understand whatever the product or service is that your company takes to market, you equally need to know what your data is in your organization as well, is what I think I hear you saying.

Jarno: Yes, absolutely. I think it's certainly crucial. I'm not saying that everyone has to learn all of these developer tools and things, but core idea of what data is and how that's different from information. How that's different from the systems within that data lies? I think it's important to understand.

Kimberly: Maybe an unfair question for you, but we fast forward 5 years from now, where do you think we'll be with using data and technology in creativity and decision-making? Or where would you like to see us five years from now?

Jarno: What I'd like to see is that we're going to have these new players in different industries that have built their company around the idea of data. Not data consultancies, not anything like that, but these companies that see that data is actually the glue between all of these silos. Data is the glue between manufacturing logistics, customers, e-commerce and so on.

If we're able to see these companies that really understand this, and don't go shopping for pieces of technology for a specific function, but rather think that an organization is this living thing and what that living thing creates that we can use in the digital platform being that logistics or e-commerce or something else is in fact data. So that when businesses understand that, what is the thing that really goes through the entire value chain is the data and the fingerprint that we leave behind on it, I think that will lead to outsized value. That will certainly change the game for many industries.

Because when you start to think that, that we are not anymore buying systems for logistics alone, we're creating a system for logistics that sits on top of the data backbone that is our company, and that is our company's value chain, that's going to unlock a lot of innovation. That's going to be a game changer in the markets, because that will lead us to understanding causal effects in our decision-making. That will lead us to creating completely transparent value chains for the end customer.

That will lead us to endless possibilities in optimization, because we are not anymore optimizing our logistics, we are not optimizing our supply chain, we are not optimizing our growth or seeing less churn. We're optimizing the entire company as a whole because it's this living thing that creates data on top of which we can create optimized services and tools and systems. I think that's true for any industry. We just need to figure it out.

Kimberly: You paint an exciting picture of what could be possible there. I hope we could pick up this conversation five years from now, and hopefully be in a place where we are optimizing the full organization like you shared. One last question for you, AI created whisky. Do you like that better than the old fashioned human-created kind?

Jarno: I think that's the unfair question. When we did the AI whisky, we certainly wanted to understand that this is going to be the first time that a lot of people will try whiskey, because a lot of people will try it because it's made with machine running. It's made with AI. A lot of people in the technology space who may have never had whiskey, will now have whiskey for the first time. We had to make a compromise, which was that you have two styles of whiskey. You have elegant whiskeys, and then you have smoky whiskeys. Elegant whiskey is respective of around 80% of the entire world's whiskey markets. We had to make an elegant one. I am a smoky whisky drinker.

Kimberly: Ah. [laughs]

Jarno: My answer is twofold. I prefer a smoky whiskey, but we haven't tried to make a smoky whiskey with AI at least yet. Until I get to try that, I cannot answer that question.

Kimberly: Can't fully answer it. Got it. It sounds like that's another project there that needs to be tackled is the AI-generated smoky whisky. Jarno, thanks so much for joining me today. I really enjoyed our conversation. It's given me a lot of food for thought about how I can use AI and data in the marketing field. Hopefully we'll have an opportunity to do some collaboration together down the road.

Jarno: Absolutely. 

Kimberly: Thanks so much for joining us for this episode of Pragmatism in Practice. If you'd like to listen to similar podcasts, please visit us at thoughtworks.com/podcasts. Or if you enjoy the show, help spread the word by rating us on your podcast platform.