Mike Connell - Digital DNA

We spoke with Alex about unlocking the potential of AI in R& D. We'll link that episode in the show notes for those who may have missed it. And today, Mike and I will be talking about digital DNA. So Mike, before we get into that, do you want to add anything to the intro or are we good to go? I guess just

Mike Connell: A little bit about me in case it's helpful context. So I have a long history in AI. I did cognitive science. At MIT, which was grounded in computer science and AI as the focus. And then I moved to Harvard where I investigated how we could apply AI to understand human learning. That was really my interest was human learning. And I was kind of using AI as a set of tools to reverse engineer that.

And then. So my interest was how do we apply those insights to improve education and that sort of a thing. So I've been working with NSTOD since 2007 and I've supporting digital transformation and scientific R& D, as you said. And as part of my role here, I developed the digital transformation program, just in case that's useful for context.

Tim Butara: Yeah, I mean, it's always, always useful to have a little bit more context. I never want to, you know, I, I think. Always the guest has the best idea of, of what information is the most relevant. So, so thank you for this extended intro, Mike. And so right off the bat, we said that we're going to be talking about digital DNA.

So what is digital DNA? What's meant by that?

Mike Connell: So digital DNA is obviously a metaphorical term and it really refers to what you can think of as. And organizations kind of intrinsic digital culture and their digital capabilities. So in part, of course, it means having digital infrastructure and digital capabilities like people's skills and, you know, having digital data, but that's just table stakes because a lot of companies have that, that don't really have a kind of a sophisticated or a mature digital DNA.

It's really about being digital at the core. And I think part of what is most important about that is that it means that you're able to use. Kind of the affordances of digital data, digital tools, digital skills in order to execute kind of a tight data driven decision making loop. And part of that is being able to respond rapidly to things that are happening in the market or in your space, but it also means being able to learn and adapt rapidly.

And so, you know, we work in scientific R and D like. Materials science and drug discovery in areas like that, and as an example there, the most of the workflows that are there are they were designed at a time when you didn't really have that kind of digital foundation. And so they were designed around humans as the center, right?

They were the humans were assumed to be the kind of decision making and computing elements in those workflows. And so. Those are not, those are like manual processes, very human centric. And you know, and a lot of the design is based on the assumption that that's how you're going to do it. And so one way to think about digital DNA is that you, you can kind of now take a fresh look at, okay, if we were to design that process today and we.

Didn't assume that humans were the main computing illness. We've got AI and we've got simulation. We've got all these advanced computational capabilities. How would we design that today using all of those affordances? And so part of it is, of course, efficiencies and you get better results and things. But what I'm pointing at is part of digital DNA.

That's emergent. Almost. It's kind of an added feature is. You've got the data there and you can use that data to make all kinds of decisions and have all kinds of insights that may or may not have to do with the task that generated the data, which currently that's, that's not the case in the manual world.

Mostly the data are just used for a task and then they're kind of discarded or maybe they try to use it later, but it's very difficult.

Tim Butara: Yeah, that was a great way to start things off. Yeah. And so I'm wondering, a lot of that sounded very similar to just digital strategy. So I'm interested, how does digital DNA both relate to as well as differ from digital strategy?

Mike Connell: Yeah, that's a good question. Good thing to tease apart. So, digital DNA is really more like the underlying foundation that defines the capabilities, what you can do, right? And what you can do with digital is more than what you can do without it. In most domains. And so it's kind of that potentiality or that capability digital strategy is really a set of choices that you make about what you're going to do with digital, right?

How you're going to win in your market, how you're going to, where are you going to play and that sort of a thing. I mean, in reality, Digital strategy refers to, I think, two different things. One is what's our strategy for becoming more digital, right? How are we going to get from our manual processes to more digitally enhanced ones?

So people kind of refer to that as their digital strategy, but then there's also the digital strategy, that's your kind of, you know, go to market strategy or your business strategy. And that, that's kind of a rethinking of, well, like I said, how are we going to rebuild our operating models? How are we going to rebuild our business models or our enterprise models?

Kind of. Getting past the assumptions of the past and looking with fresh eyes at what digital affords us that wasn't possible before. And you know, what's holding us back because we built it around a different paradigm when we didn't have all this stuff available. So anyway, those are two meanings of strategy, but they both come down to choices about how you're going to use digital, whether it's for changing yourself or for winning in the market.

Whereas digital DNA is the set of capabilities that you put in place in order to, you know, to, that you build a strategy on top of.

Tim Butara: So they should kind of work in tandems in some way.

Mike Connell: Oh, yeah. Yeah. So I would say that your capability set, whether it's kind of pre digital or digital. It enables you to, to do certain things.

And so your strategic options are different. If you're on a digital, if you have digital DNA or digital foundation, then the strategies that you can pursue are very different than the ones that you could pursue if you don't have that. So it opens up a world of strategic possibilities, I guess you could say, in terms of like the objectives you can go after the the, the economics of going after them, like how expensive it is to do things.

So for one, like, like I said, part of digital DNA is. This ability to use the digital matrix and the data that are coming in, in order to have insights, make decisions, learn and adapt rapidly. That's one of the radical things that changes when you get, when you have digital DNA versus kind of the old, old style, you know, manual kind of foundation.

It's very, it becomes, it's failing fast and cheap, right? You can try things and there's not, there's not a big consequence for it. But when you're in a manual world or kind of the old world, Yeah. You don't try things that you don't have high confidence are going to work because it's just way too expensive and risky to fail.

So it could be catastrophic to fail. You know what I mean? And so that changes the game completely. Like what you can do in a world where you can fail fast and fail inexpensively, that opens up all kinds of strategic possibilities that just aren't available otherwise.

Tim Butara: And these were already some key characteristics of a healthy digital DNA, right?

So, so what are some other signs of healthy digital DNA? Yeah, that's true. We're kind of

Mike Connell: touching on that. So I think, I think it's easiest probably to talk about this using a contrast pair, right? So taxi companies, I'm going to use a generic one that everybody's familiar with. So taxi companies don't have digital DNA, right?

They have cars, they have drivers, people call up on the phone. Now they can layer data. Technologies onto that, like they got credit card processing systems at some point in the cab that was helpful incrementally, but it didn't fundamentally change the business model of the operating model. They got GPS, which is helpful right to drivers, especially inexperienced drivers who don't know the area to get around.

They maybe got apps where you can kind of call for a cab, but fundamentally, they're still operating the same model, the same operating model, the same business model. And so these things are kind of incremental. Enhancements locally to a task that you're doing Uber, on the other hand, provides the same kind of car service from nominally from the end users perspective, but they don't own any cars.

They don't maintain cars. They don't maintain insurance. They don't hire any drivers. And they have and you know, they're I mean, they're they're I forget what they're worth now, but tens of billions of dollars, 50 billion or something, right? A taxi company has, has trouble becoming worth 10 million probably locally.

And so it can't scale beyond that. So that's kind of the, the sign of a healthy digital DNA. You can kind of see in that contrast, it means that they have things like data infrastructure. And I mean, a big part of this is. That though, kind of all the activities, there's data representing the activities that you're involved in.

Let me give you an example from a domain that we work in of materials development, like people are trying to develop new materials for the market. So typically, you know, they want to come up with a new plastic or something. One that is high temp, it's, you know, it's got a high melting point. And it baffles noise cause you want to use it for an engine cover.

So. Typically there's kind of this science process where they have to go through and they do trial and error to find a plastic that will satisfy those requirements. And so the data that they're generating, they're make, they're mixing up a batch of some kind of plastic that they think will work and then they measure, you know, tensile strength, how does it break, what's its melting point, all these things that matter in the application.

Those data are kind of exhaust from the materials development process. They have to collect them in order to get this task done. And then what happens is they, they don't manage the data well. They don't think of the data as Okay. Kind of a first class object. And so people replicate some of the same experiments or similar experiments.

Starting points are really important. That determines whether you're going to succeed and how quickly, but they rely on their own memory or other people's memory to kind of find, find starting points for this. If they have a new idea that they want to explore, they can take It can take weeks or months to try and find the data and compile it and then the data quality are poor and so it may not even support testing their idea.

Even if the data kind of flowed through the system at some point, it's not captured in a way that they can use it for this. And if they want to try something new, like, hey, we want to try machine learning in our, in our process, you know, it's a major ordeal to get the data. They probably don't have suitable data for the reasons I just sort of described or alluded to.

The data are inconsistent. It's scattered. It's not suitable. It's not standardized and so on. So there's no, and then there's no straightforward way to deploy those models in production. So those are kind of examples of like, they don't have good digital DNA. They're trying to do something new. They're trying to use the data they have to, to, to capitalize on an insight.

And they really can't. If you have digital DNA, the data are automatically collected, they're high quality, they're clean, they're complete, they're in a standard form for secondary analysis it's easy for end users who have ideas to access it. It's suit, and it's suitable, right? It's easy for them to use for secondary analysis.

The tools they're using hide the incidental complexity. They don't have to know about AWS server configuration. They don't have to have a database admin skill set. They can get the data they need for the, for the uses they're doing. And so, so they have, And they can try things out rapidly and they can do it again with negligible expense.

Right. So kind of this very rapid learning process and failing fast and all that. And then if they want to incorporate a new tool like machine learning, it's easy to get set up. It's easy to get the data. It's easy to explore. It's easy to maintain the models. It's easy to deploy the models to their colleagues.

And then people have those. I mean, part of it is obviously people skills as part of this digital DNA. They have the skills to make use of data, to access the data, to analyze the data, to make decisions off of the data. And that is a different skill set. I mean, chemists know how to think about chemistry, but they don't necessarily know how to think about digital chemistry unless you provide those skills.

So that's kind of, that's kind of a contrast pair there in more, Detail, which is in the one case they can, they can do new things. They can bring new tools and they have an idea. They can try it out all because the digital world, the digital infrastructure and tools and stuff are set up for them to be able to have insights, try things out, make decisions in the other case.

It's not set up for that. You know what I mean? They have some digital elements, but it's not set up for them to be able to do this kind of rapid decision making, learning, exploration. And so, you know, I guess, taking it to a more familiar realm, the taxi company versus Uber. Uber's got all the data on all the transactions of all the people.

A taxi company, if they're paid in cash, has no record. There's no data about that ride that they gave someone. Who was it, you know, I mean, they probably tracked where they went and stuff, but they, they don't know anything about the person, whereas with Uber, they've got the whole profile on everyone, and so they can start to look at the data and say, Hmm.

You know, people don't seem to take rides that are less than a mile long. And so they come up, what can we do to do that? So based on those data, they come up with an idea for scooters, right? Where you can rent a scooter for short trips. And they say, ha, you know, some people are on the same route. And I wonder if some people would like to save money by carpooling.

They have all the data on where everyone's coming from and going to. They can create a new business line, essentially, doing carpooling or Carrying packages and maybe combining those two things where I'm carrying a package and along the route, I'm picking someone up and dropping them off nearby. So there's these incredible efficiencies that they can come up with and new business lines.

You can't do that with a taxi company. You don't have any of the infrastructure, the data, the, you know what I mean? So that's kind of the, I hope that makes it kind of clear what the, you know what the elements of a sort of good or healthy digital DNA are.

Tim Butara: It sounds like pretty much the most important thing to a good digital DNA is data efficiency.

Mike Connell: Well, data is at the center of it. I think it's, it's necessary, right? It's, it's, it's a critical element that you have to get right, but it's not sufficient because you have to have the tooling and the access patterns and people have to have the skills and stuff. Yeah. And it has to like. Barnes and Noble went against Amazon.

All they did was layer digital technologies on top of their current inventory system, for example, and so all they could do is the same thing that they could only have the same insights and business decisions that they had before in the manual process. Amazon, on the other hand, had like all the credit card data of everybody, and they were able to analyze the data in a different way.

So they could look at the, what they were selling and they realized that they were selling to the long tail. They were selling books that you couldn't get at the main, at the stores that were the popular books that were on the shelves. So they could do that kind of analysis and pivot how they were marketing to people to say, Hey, get your, you know, get your rare books here.

Whereas Barnes and Noble, because they didn't do anything to Basically to make use of the data once they had it. Do you know what I mean? So it's more than that It's the processes around the data and it's the tooling and it is the skills of the people They all have to work together in order to sort of make that work.

Does it make sense?

Tim Butara: And are there any any industries or any particular use cases where it's Particularly important to to have a really good digital dna and why so?

Mike Connell: Yeah, it's another good question. So I think it's coming probably to all industries, but one area where certainly we work most in these scientific domains like you know, drug discovery, development materials.

I do think that that is an area that, especially with the rise of generative AI. It's becoming more urgent for them to, to kind of clean up their, their digital core, their digital DNA. And I can cite a couple of reasons. One is because in some ways science has been, that kind of science and R and D has been resistant to other forms of AI because it's too complex, right?

And it's too, it's too complex. The rules are too obscure and we don't have enough data to really kind of map it. The chemical space, for example, of how you know plastics. If you want to make a new plastic, you just have to go do trial and error because we don't know the logic by which these formulas produce plastics with certain certain properties that's changed with generative AI.

Basically, what LLM show is that we can take. We can take these very complex spaces like English language or Japanese language and we can actually get those models to essentially, they extract the logic, the grammar of those languages. Well, chemistry is a language too and so is biology, right? And so that's what we're going to see is that for the first time we have the ability to capture the logic of these scientific domains.

And I think that's going to be a radical game changer and we're already seeing that happen. And so what that does is it makes it so that you can now treat biology or chemistry in the same way that you treat, you know, a database or the Internet. Where you can search it, you can analyze it, you can do synthesis on it.

You can optimize things across it. That's never been possible before. It was just always a manual process. So that's a huge change. And we're already starting to see people who are. Are kind of pursuing that and they're, they're upgrading their digital DNA around that idea. They're starting to pull ahead and I think, you know, it's, it's going to become very, very urgent because there is a bit of a winner take all phenomenon here.

Like if you're by the time the case studies come out from those people, it's going to be too late for others to jump on. The other one is that the process of doing this kind of modeling is accelerating and we have some open source or publicly available models like alpha fold. Or protein folding, which is very important for drug discovery and other applications.

And so what that means is that even with the rise of kind of generative AI, the larger companies that had the money to do training or building their own models, or they had a lot of historical data, Right. Those two, again, data is critical to this. And so they had a lot of data. They had a lot of resources to kind of build the models.

That's changing where the ability to do this, this modeling is accelerating. We're starting to see open source versions of these models. And so that's kind of cracked open. I think the competitive landscape to all kinds of new entrants that are much, much smaller and leaner and, you know, digitally native and stuff.

And so the playing field is changing, I think, quite dramatically there. In real time right now. And again, I point to that because because of this change where for the first time, I think, in history, we can actually make the scientific realm digital and start to apply these advanced computational capabilities to it in a way that was not possible before.

Tim Butara: I like how this is all. This is also a little bit connected to the discussion that we had with your colleague, Alex, that we published back in December of 2023. It was exactly that. It was basically how AI can help unlock innovation in this sector, which has had different challenges. To other industries, maybe in, in catching up with all this technology, innovation, and all this digital transformation.

And now it's kind of time for, for this sector to also reap the rewards of all the digital and AI have to offer. I think that's right. Yeah. And so Mike, besides everything that we already mentioned today and to kind of start rounding off the conversation, do you have any other key tips and best practices?

For, you know, improving and building a healthy, good digital DNA.

Mike Connell: Yes. So if you're a startup, you definitely want to do that from the start. And part of what makes that hard is that startups are moving quickly. And their main goal is to sort of build things and get them over the line to get them to market, do product market fit, that sort of a thing.

But then they end up with all this sort of lost opportunity. If you kind of build it right from the start in the ways I was talking You're capturing the data in a way that you can use for secondary analysis. That's extra work a little bit, but it's not, but the value that comes from it is, is much bigger than the sort of investment there.

Typically we work with established companies and they have a different set of problems because they have a business that's functioning and they don't want to mess it up. And so I think in, in both cases, but especially in the latter case, people have trouble figuring out where do we start or what should we prioritize?

And really you should look at. Sort of the business value, like what's going to come out of this for you and start with that. And then, right, because we need those business outcomes to justify this to all the stakeholders and shareholders and keep and if you want to keep this process going of digital transformation and getting and building that digital DNA.

So you're going to need multiple opportunities, right? Different projects to do that. And the way we do it is we kind of start small, but we go deep. Because part of what you need to attend to is not just introducing the technology. This is not primarily a technology initiative. It's like 20 percent technology and 80 percent change management, people change, you know what I mean, and process change and stuff around that.

And so what, so you could do things like. You know, layer technology on like, like Barnes and Noble did onto their inventory system, and that might be valuable to them. There's a business win there, right? But if you're going to do something like that, you want to go deep. And so you want to change all the way down.

You want to make sure that you're collecting all the data that you're, that the data are going to be available to all the people who could use it to have insights and make decisions in any part of the business. And so it's really about. Let's do a project, but let, and let's get the value of that project.

We kind of do the project for the business value, but then at the same time, we're doing the project to build the muscle, not the muscle, but using the metaphor to build our digital DNA. And each time you do one of these projects, the digital DNA gets, you know, broader and deeper and sort of more coherent and stronger.

And so then what you'll see is the new capabilities ultimately emerge. And so that's kind of this dual thinking, or at least the way we do it is you get short term value, and then you're building this long term capability. Yeah, and so again, you have to attend. So when you're doing these projects, that is a good way to think about it.

I already stated it, but I'll restate it here as a tip. Don't just look, where can we get wins in our current process by speeding things up or making them cheaper? Look at it through the lens of if we were to rebuild this business process today with the same desired output, whatever it does, it's valuable.

For your business, you want to retain that. But if we were to rebuild it today, given the computational capabilities and digital technologies that are available, what assumptions would change, like how would we build that from the ground up differently? Now you don't have to go and throw the old one out and do that.

We also kind of have a way to. You can kind of take an agile approach where you do it piece by piece, but you know, you have to have a roadmap and a vision for where you're going to end up. And the whole ends up greater than the sum of the parts because you get those incremental enhancements along the way.

But if you do it right, you end up with something different at the end, which is sort of your digital DNA fully functioning. Does that make sense?

Tim Butara: It makes sense. And great tips right here at the end. Great strong notes to finish on a great conversation, Mike, just before we wrap things up. If anybody would like to connect with you and learn even more from you, what's the best way to reach you?

Mike Connell: You can email me. I'm mconnell, c o n n e l l at enthought. com or certainly find me on LinkedIn, Michael Connell. And, you know, you can, you can search for enthought and you'll find me.

Tim Butara: Okay, great. We'll make sure to put everything in the show notes alongside the episode with Alex for reference. And Mike, thank you again for a great conversation, for being our guest today.

It was great having you here.

Mike Connell: Oh, it's my pleasure. Thanks for having me, Tim.

Tim Butara: And well, 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.