TRANSCRIPTIE 🔗
WEBVTT
00:00:00.160 --> 00:00:03.680
Hey everyone, welcome to another episode of Beyond the Case.
00:00:03.919 --> 00:00:10.640
This is a podcast where global leaders from Howard Business School's OPM community join in a personal capacity.
00:00:10.800 --> 00:00:16.480
They share their real life lessons, mental models, and the decisions they've taken to build enduring companies.
00:00:16.640 --> 00:00:19.280
Today's guest is Dennis Campbell.
00:00:19.440 --> 00:00:22.079
He is a professor at the Howard Business School.
00:00:22.239 --> 00:00:25.120
He is a professor for the OPM community.
00:00:25.280 --> 00:00:31.600
He taught us AI and organization design uh earlier this week as part of my unit too at uh OPM67.
00:00:31.760 --> 00:00:33.200
Dennis, it's a pleasure to have you here.
00:00:33.280 --> 00:00:33.759
How are you?
00:00:34.079 --> 00:00:34.719
Thank you, Saran.
00:00:34.880 --> 00:00:35.280
Great to be here.
00:00:35.600 --> 00:00:39.600
Pleasure to uh be on the podcast and to interact with the broader OPM community.
00:00:39.840 --> 00:00:40.159
Thank you.
00:00:40.320 --> 00:00:44.960
Dennis, just uh for the benefit of the listeners, could you give a brief introduction on yourself and your background?
00:00:45.280 --> 00:00:45.679
Sure, yeah.
00:00:45.759 --> 00:00:47.679
I'm a professor at Harvard Business School.
00:00:47.759 --> 00:00:52.240
I've been here for a long time now, about uh 23 years on the faculty.
00:00:52.399 --> 00:01:03.600
And uh during that time, I've spent a lot of my kind of research and teaching around what you might broadly think of as analytic technologies and organizations, how you can leverage data science analytics to make organizations work more effectively.
00:01:03.759 --> 00:01:11.680
And of course, now we're in a world of artificial intelligence, and this has been a major shift in what we're teaching, how we're teaching at Harvard Business School.
00:01:11.760 --> 00:01:20.319
And I've been involved for the last three years and in our new course that we're actually requiring for all MBA students, all 900 of them on data science and artificial intelligence.
00:01:20.480 --> 00:01:27.120
I'm leading many of our executive education kind of efforts uh with large and small companies around the world in this space.
00:01:27.280 --> 00:01:31.840
Everybody's grappling with it, and I think, you know, we're trying to stay at the forefront of it here at HBS.
00:01:32.079 --> 00:01:32.719
Great, thank you.
00:01:32.879 --> 00:01:34.879
Let's start with uh with your journey.
00:01:35.040 --> 00:01:39.200
What sparked your interest in academia, in business and technology?
00:01:39.439 --> 00:01:41.920
And how did you find your way to Howard Business School?
00:01:42.159 --> 00:01:49.359
Yeah, you know, I um I was when I was an undergraduate, I was uh I majored in mathematics and economics, and I also had a fascination with mathematics.
00:01:49.439 --> 00:01:55.920
And I got interested in kind of mathematics in the business context, and I ended up, you know, I knew I wanted an academic career.
00:01:56.000 --> 00:01:58.799
I really just loved the freedom of academic exploration and research.
00:01:58.959 --> 00:01:59.760
I enjoyed teaching.
00:01:59.920 --> 00:02:04.400
So I spent a little bit of time working outside at the Federal Reserve doing some economic research.
00:02:04.640 --> 00:02:09.120
But I knew I eventually wanted to go get a doctorate, and I ended up at Harvard Business School.
00:02:09.199 --> 00:02:19.840
I was really looking for a place that I could do, you know, get PhD level training and do research, but also in a way that could have impact on real organizations and understanding how real organizations work and can be improved.
00:02:19.919 --> 00:02:26.560
And I always thought Harvard, from the time I was a new doctoral student to the time I stayed on the faculty, was a great place to do that kind of work.
00:02:26.639 --> 00:02:40.080
And, you know, it's been a place where I feel like it gives us the freedom to really invest in understanding the technological frontier around analytic technologies and also really do leading, cutting edge research, but also keep a foot in the real world.
00:02:40.319 --> 00:02:41.280
Harvard really values that.
00:02:41.360 --> 00:02:47.439
And it's been a great, great place from my perspective in my entire journey from doctorate through being a faculty member.
00:02:47.759 --> 00:02:53.520
How did your research interest evolve and what continues to excite you about teaching and working with business leaders?
00:02:53.759 --> 00:03:01.439
Well, if I think about the current moment we're in, even just teaching in this course on data science and AI, this is a fast-changing technology, right?
00:03:01.520 --> 00:03:03.840
And it is it's hard to keep up with, I think, for anybody.
00:03:04.000 --> 00:03:08.000
And it's not just about, say, the mathematics and the technology underlying it, right?
00:03:08.080 --> 00:03:12.960
It's actually just about what's working, what's not working, how organizations are adapting and should adapt.
00:03:13.120 --> 00:03:20.319
And I think, you know, as faculty members, we have a responsibility to stay on that curve, ahead of that curve, but also in a rigorous way, right?
00:03:20.400 --> 00:03:22.960
To really help people understand how to navigate it.
00:03:23.039 --> 00:03:33.840
And I think for me, that's what's most exciting is that teaching in this space and doing research in this space is really forcing me to stay at the forefront of it and really grapple with how the world is changing around it.
00:03:33.919 --> 00:03:40.400
So, and I feel like we have a responsibility to do that in a way that is, you know, not just chasing trends, but is actually rigorous, right?
00:03:40.479 --> 00:03:47.759
Taking a real look at it, understanding what's working, what's not working in terms of how leaders and organizations can adapt to it, how it's affecting work, right?
Load full transcript
00:03:47.840 --> 00:03:57.039
I mean, you see a lot of uh kind of uh discussion, right, about the negative impacts of this on work, but then you actually see that, you know, some of those, you know, dynamics aren't playing out the way we expected, right?
00:03:57.120 --> 00:04:03.280
And so I do think it's a it's very complicated space and but it's also very engaging to be kind of in the middle of understanding.
00:04:03.520 --> 00:04:07.039
So Dennis, we've now had four sessions on AI at this unit.
00:04:07.199 --> 00:04:08.240
Two of them were with you.
00:04:08.400 --> 00:04:13.120
And I think most participants are convinced that AI will transform their business.
00:04:13.439 --> 00:04:22.720
But many of us are still uh are still asking, how do I uh I get it, but how do I actually go ahead, go to office on Monday morning and implement this?
00:04:22.800 --> 00:04:27.600
Because there's such a big gap between understanding AI and implementing, implementing it.
00:04:27.759 --> 00:04:29.120
Why do you think that gap exists?
00:04:29.360 --> 00:04:29.519
Yeah.
00:04:29.839 --> 00:04:31.439
I think I think for many reasons, right?
00:04:31.519 --> 00:04:37.360
I think there's there's so first of all, just think about it's like most of us find a way to use it individually, right?
00:04:37.439 --> 00:04:42.639
For our own individual productivity purposes and probably get great, great sort of uh benefits out of it.
00:04:42.879 --> 00:04:46.079
I think it's really different though when you say, how do I use it for my organization, right?
00:04:46.160 --> 00:04:47.680
That's so that's a very different thing.
00:04:47.920 --> 00:04:50.319
And it's very different than every individual just using it, right?
00:04:50.399 --> 00:04:58.560
Because the idea is it has to work with, to make this work in a way that it leads to productivity for your organization, that leads to maybe competitive advantage, right?
00:04:58.639 --> 00:05:01.600
You actually have to think deeply about your processes.
00:05:01.680 --> 00:05:13.600
And processes aren't just about individual work processes, it's also like your culture and how people interact with each other and what the barriers to coordination in your company are, what people have agency or what they believe they have agency to do and not do, right?
00:05:13.759 --> 00:05:18.959
I think uh those things have to be really worked out in conjunction with implementing AI in an organization.
00:05:19.040 --> 00:05:28.079
So I think that's number one is it's like using it for your individual purposes is much easier, I think, than if you're a leader trying to use it in a productive way for organizations.
00:05:28.160 --> 00:05:32.800
And, you know, when you start to look at it that way, it gets into things like just even adoption, right?
00:05:32.879 --> 00:05:36.480
So, like do people understand where to apply it in their work, right?
00:05:36.560 --> 00:05:46.079
What are you doing as leaders to create the conditions where
people are don't just feel free to experiment, we're also experimenting in the right direction so that they're seeing uh, you know, potential productivity gains for them.
00:05:46.160 --> 00:05:47.279
It's making them better off.
00:05:47.439 --> 00:05:47.759
It is.
00:05:47.839 --> 00:05:49.920
So it's about adoption and people's resistance to it.
00:05:50.000 --> 00:05:55.759
It's also about how do you, as leaders, get a sense of a feeling that it is safe to implement, right?
00:05:55.839 --> 00:05:58.800
That it's uh reliable to implement in your particular setting, right?
00:05:58.879 --> 00:06:03.680
That many of you maybe you have sensitive data, you don't want that getting exfiltrated from your organization.
00:06:03.839 --> 00:06:09.519
You have uh maybe operate in regulated sort of spaces, right, where there's a lot of risk if things go wrong.
00:06:09.600 --> 00:06:19.279
And we have to remember that this, the, the sort of technology underlying all of this, you think of most companies are using LLMs in some way or trying to, those are inherently probabilistic.
00:06:19.360 --> 00:06:21.439
And so, how do you make them behave in reliable ways?
00:06:21.519 --> 00:06:32.079
And I think that's where you have to get into both the technical architecture of what you're implementing, but also governance around people and around the technology, what it can and can't do, can and can't access, for example.
00:06:32.240 --> 00:06:42.720
So, long story short, I think you start to like look at all of those things, it's much more complicated to drive this in organizations than it might be for any one individual to use it in very productive ways.
00:06:43.040 --> 00:06:52.000
We have all types of participants here from small businesses, some of them as small as just the founder themselves with no employees, and some hundreds or thousands of employees.
00:06:52.319 --> 00:06:58.000
How do they all go about identifying the right AI opportunities and prioritizing them, in your opinion?
00:06:58.160 --> 00:07:02.560
Is there a framework that you think could benefit uh these business owners?
00:07:02.800 --> 00:07:15.120
Uh I don't know that there's a specific framework for that piece, but I do think that what I would say is this is what one of my colleagues actually put a term to this, which he called the, his name is Remcon, and he put it this term to it called the AI mapping problem.
00:07:15.199 --> 00:07:17.199
And I think this is probably the way to think about it, right?
00:07:17.279 --> 00:07:28.639
Which is like, I think what's really different about this technology than prior technologies that may be built for, you know, largely a single purpose or for well-understood purposes, this isn't really like that, right?
00:07:28.720 --> 00:07:35.040
This is a technology that we might think of as more of a general use technology that could have any number of applications, right?
00:07:35.120 --> 00:07:41.040
And so when you let this loose for people, how
do they know the direction in which to experiment and which are what are the best use cases?
00:07:41.120 --> 00:07:44.879
And I think for the organizations in, say, OPM, it's the same thing, right?
00:07:44.959 --> 00:07:48.319
It's that they're like, what is where do we start with this technology, right?
00:07:48.480 --> 00:07:51.199
Where's the best, where are we gonna get the most value out of this?
00:07:51.360 --> 00:07:58.399
And I think it's really important to answer, not just because it's about the value question, it's actually about the capability building around it, right?
00:07:58.480 --> 00:08:04.480
So that the what we often see is large and small when organizations identify those those best use cases, right?
00:08:04.639 --> 00:08:07.279
You start, they build capabilities that then cumulate.
00:08:07.439 --> 00:08:12.000
And I think that the answer to this is different, I think, for a small versus a large organization.
00:08:12.160 --> 00:08:17.519
I think a large organization has to be much, much more systematic about this than a small organization, right?
00:08:17.600 --> 00:08:22.560
And so you think about, you know, an organization with thousands of people, you know, or hundreds of thousands of people.
00:08:22.639 --> 00:08:23.519
And I can give you an example.
00:08:23.600 --> 00:08:30.319
One of the examples I did not get into in class is uh a group company called DaVinci Group, which is one of the world's largest construction companies.
00:08:30.480 --> 00:08:32.480
They have like 4,000 business units.
00:08:32.639 --> 00:08:35.440
They have um 300,000 plus employees, right?
00:08:35.519 --> 00:08:38.480
This is a huge, huge company, very dispersed for good reasons, right?
00:08:38.559 --> 00:08:43.679
They need to be very focused, very localized, very agile in all the markets that they're in, all the products that they're in, and so on.
00:08:43.840 --> 00:08:50.320
But that creates a lot of challenges, you can imagine, for what is where where do we focus on for AI in an organization like this?
00:08:50.480 --> 00:08:58.480
And they've they've developed a really nice kind of lightweight structure, basically a platform that sits alongside the corporate structure called Leonard.
00:08:58.559 --> 00:09:06.879
This is like their creative unit that's focused on raising awareness of AI and also um incubating interesting AI ideas that come from bottom up.
00:09:07.039 --> 00:09:09.840
But you can imagine, unless you're gonna have a massive unit, right?
00:09:09.919 --> 00:09:14.240
How do you actually deal with all of the ideas that might be coming up from 4,000 business units?
00:09:14.399 --> 00:09:18.720
You have to have some rational basis on which to select the ones to put limited resources behind.
00:09:18.799 --> 00:09:27.679
And so what they do, I think, in a clever way is if you think of like any organization, large or small, what you really want to pick are the use cases that will be like locally profitable.
00:09:27.759 --> 00:09:35.919
What I mean is like they can lead to products and they can lead to like processes that actually will be beneficial, either in productivity or revenue growth, whatever it might be.
00:09:36.080 --> 00:09:40.480
But you also want the ones that are going to scale, right, and impact more parts of your organization.
00:09:40.559 --> 00:09:42.879
That's a more important one you're in a larger organization, right?
00:09:43.039 --> 00:09:47.200
So if you imagine this Vinci case, right, like the first part is pretty easy, right?
00:09:47.279 --> 00:09:59.440
Because why would a local business unit even volunteer an idea, send a, you know, a person out of that business unit to spend time developing and innovating something, unless they believed it was going to be beneficial for that business unit, right?
00:09:59.519 --> 00:10:00.480
So that's kind of easy.
00:10:00.639 --> 00:10:03.840
The part that's hard actually is how do we pick the ones, right?
00:10:03.919 --> 00:10:09.600
If I'm sitting in this unit, how do I pick the ones that are likely to be most scalable and impact the broader organization?
00:10:09.759 --> 00:10:14.080
And there, what you want to think about is what kind of capabilities do we really want to build?
00:10:14.159 --> 00:10:17.440
And that should actually map to how you think about use case choices.
00:10:17.600 --> 00:10:18.720
And let me give you an example of this.
00:10:18.879 --> 00:10:32.639
If you look at uh in that particular case, there it highlights how one enterprising engineer that was sitting in a business unit that uh basically bid and designed, bid on and designed the Springbok systems, right, for fire prevention in buildings.
00:10:32.720 --> 00:10:36.240
And you can imagine this is super different from project to project, right?
00:10:36.399 --> 00:10:37.440
Buildings are different.
00:10:37.600 --> 00:10:42.399
And you have to come up with a very complex design, and then you have to price it, and then you have to try to win the bid.
00:10:42.559 --> 00:10:43.840
This could take a month to do.
00:10:44.000 --> 00:10:45.200
And they might not win, right?
00:10:45.360 --> 00:10:51.840
This guy had a good idea, which was now that I've seen what AI can do, we have a bunch of blueprint data from past projects.
00:10:51.919 --> 00:10:53.759
Some of those have been successful, some haven't.
00:10:53.840 --> 00:10:59.600
Well, he saw that as an opportunity to say maybe we could build some model that could help us speed up that process, right?
00:10:59.679 --> 00:11:02.080
And bid a lot faster and actually at a lower cost.
00:11:02.320 --> 00:11:06.559
So he get he gets picked, right, by this unit says, go off and create this.
00:11:06.720 --> 00:11:08.240
He creates it, it's successful.
00:11:08.399 --> 00:11:17.200
He builds this thing that actually called Sprinkler AI that basically design optimizes those designs and it brings down that design process into less than one day.
00:11:17.360
--> 00:11:21.759
So it means that they can bid faster, they could bid at lower costs, more likely to win the bid, right?
00:11:21.840 --> 00:11:31.440
So this is a big deal locally, but the question then becomes why would that unit select this project and maybe the hundred plus that get proposed for all these 4,000 business units?
00:11:31.600 --> 00:11:42.320
And the answer is the way they thought about that mapping problem is we want to have upfront a strategy around which tech underlying technologies have the most potential use cases, right?
00:11:42.399 --> 00:11:43.919
So you have to do that upfront mapping.
00:11:44.080 --> 00:11:47.360
And in this case, they didn't really care that it was like the sprinkler AI system.
00:11:47.440 --> 00:11:52.000
What they cared about was it leverages an underlying idea or technology called generative design, right?
00:11:52.080 --> 00:11:59.519
That idea that you generate a design, right, out of this technology using some kind of machine learning kind of methodologies and optimization techniques.
00:11:59.679 --> 00:12:11.360
And the interesting thing is then what happens with that person, he goes out back into the business units, he he sells this sprinkler AI system internally and so on, but he's also in charge of a new what they call center of excellence.
00:12:11.440 --> 00:12:15.440
But the center of excellence is not about sprinkler AI, it's actually about generative design.
00:12:15.600 --> 00:12:16.720
He becomes the expert in that.
00:12:16.799 --> 00:12:22.240
Now, somebody that's for HVAC, right, that wants to do this, they come to that person, not to this small unit.
00:12:22.480 --> 00:12:26.799
Somebody that wants to do it for wiring or for roads in front of the buildings comes to that person.
00:12:27.039 --> 00:12:28.240
So they do this over and over.
00:12:28.320 --> 00:12:39.759
They have like this, these kind of an upfront, well-defined set of technologies they believe will have the most potential use cases, and they prioritize selecting those projects and seeding those capabilities.
00:12:39.919 --> 00:12:46.559
And this has led to them developing this idea of generative design spreading way beyond just sprinklers into all these other areas as well.
00:12:46.639 --> 00:12:49.600
And that's happened over and over for them in a number of other technologies.
00:12:49.759 --> 00:13:05.039
So I want to raise that just to say that that mapping problem and thinking deeply up front about not just profitability or productivity, but this idea of scalability, which kinds of technologies or use cases have the potential to impact the broadest parts of our organization.
00:13:05.200 --> 00:13:10.960
The more you as a leader could specify that up front, the more focused and I think impactful your AI direction investments will be.
00:13:11.200 --> 00:13:11.360
Yeah.
00:13:11.519 --> 00:13:22.960
A lot of the business owners, I feel, are struggling with even understanding what the right questions to ask are of their business in terms of understanding what could be automated or where where they could benefit from AI.
00:13:23.039 --> 00:13:25.360
I feel they lack the technical know-how themselves.
00:13:25.519 --> 00:13:28.480
And so they don't know where to start with this whole process and the endeavor.
00:13:28.799 --> 00:13:29.919
So here's what I think is interesting.
00:13:30.159 --> 00:13:40.960
This is it sounds like it's going to be contradictory to what I just said, but I actually think it's it it they come together in a nice way, which is in cases like that, you might want to ask yourself like who in the organization does know the right questions.
00:13:41.120 --> 00:13:43.919
And sometimes that's the people on the front lines, right?
00:13:44.000 --> 00:13:49.360
It could be the front lines of your processes, it could be, you know, people that are seeing the work every day and closest to the work.
00:13:49.519 --> 00:14:01.120
And I think that this is what you sort of see happening at so I I where I'm saying it's contradictory in a sense is I'm telling you, you as a leader need to upfront specify the technologies that have broad use cases.
00:14:01.200 --> 00:14:09.919
But what I'm arguing here is the people that are closest to the work are the ones that probably have the best idea of the use cases and and where it could be applied.
00:14:10.159 --> 00:14:14.399
And I think if I think if you really look at that example I gave you, it kind of shows both happening, right?
00:14:14.559 --> 00:14:24.720
They've specified a broad set of technologies that could actually have a broad potential use cases, but it's these people at the front line that are seeing the potential and of where this could really impact their work.
00:14:24.799 --> 00:14:37.360
And the more you can allow that to happen, right, for people to understand that you raise awareness of the technology, what's possible with it, you allow people to then have ideas about it and even to experiment and build in their own work, right?
00:14:37.440 --> 00:14:42.960
I think this is where you actually might and then where you actually can see things that are impactful, right?
00:14:43.039 --> 00:14:47.679
Now you as a leader can then say, okay, these are the kinds of ideas I'm seeing have impact.
00:14:47.840 --> 00:14:50.320
Now we might want to build something broader across the company.
00:14:50.480 --> 00:14:52.240
And you actually saw that in the Who's case.
00:14:52.320 --> 00:14:56.559
If you look at that, that was not the CEO that came up with this new architecture, right?
00:14:56.639 --> 00:15:01.600
It was like one software engineer that wanted to improve his own productivity.
00:15:01.840 --> 00:15:11.679
He builds this kind of like a like an MVP of this broader system you saw implemented there with his own memory layer, his own agents, right, helping him do coding much faster.
00:15:11.840 --> 00:15:16.799
And then he brings out to a CEO, the CEO sees the potential of
that to impact the broader organization.
00:15:16.879 --> 00:15:27.279
So I think that's a perfect case where you see this interesting juxtaposition of people closer to the work, but then a willingness of the leader to really say, okay, what can I take from this that could work across the organization?
00:15:27.519 --> 00:15:33.039
And how do these business owners measure whether the AI investments are actually delivering business value?
00:15:33.279 --> 00:15:48.240
Because I think after a point, if it's if they're not uh measuring these metrics or trying to track them, I think they could be tricked into believing that this is just a tap they've left on and they're pouring money into an endeavor which is not giving them any returns.
00:15:48.480 --> 00:15:50.240
I'm a I'm a huge proponent of measurement.
00:15:50.480 --> 00:15:52.960
I'm I'm technically in a group called accounting and management.
00:15:53.200 --> 00:15:59.360
I'm more on the management side, but we think a lot about how information should should flow in organizations, right?
00:15:59.440 --> 00:16:00.480
How do we measure performance?
00:16:00.639 --> 00:16:03.759
How do we use in data sources and new analytic techniques to do that well?
00:16:03.840 --> 00:16:12.240
So I'm a big proponent of measurement, but I would say I would be careful here and don't don't uh don't over-index on on measurement here.
00:16:12.399 --> 00:16:18.799
I it's important to do some of it, but I guess the the point of this is to say like measuring ROI in this space, I think is super difficult.
00:16:18.879 --> 00:16:21.840
And I think it's a worthy endeavor to figure out how to do it better.
00:16:22.000 --> 00:16:32.159
But I think the the thing I'm scared about with this is that uh this, if you focus too heavily on it, it could lead you too much towards much more narrow use cases that may have some productivity benefits.
00:16:32.399 --> 00:16:33.919
Those are the things that are easiest to measure, right?
00:16:34.159 --> 00:16:35.519
Okay, we could do this faster.
00:16:35.679 --> 00:16:38.000
It helps each individual be a little more productive.
00:16:38.159 --> 00:16:38.720
That's great.
00:16:38.879 --> 00:16:47.039
But it may, if you focus too much on those things, it may leave off the table or it may get you to move away from things that could be truly transformative, right?
00:16:47.120 --> 00:16:52.559
That actually like those, so think about some of this idea about like, look, if we free people's time up, right?
00:16:52.720 --> 00:16:55.600
That's fine if they just keep doing the same tasks.
00:16:55.759 --> 00:16:57.200
They just do more of them, right?
00:16:57.360 --> 00:17:02.799
But if actually it means they can go off and do more high value tasks, well, can you measure what else they're doing?
00:17:02.960 --> 00:17:06.000
Can you actually trace that and say, what other opportunities are they now pursuing?
00:17:06.079 --> 00:17:07.359
Where's that value created?
00:17:07.519 --> 00:17:08.960
That's not as easy to measure, right?
00:17:09.039 --> 00:17:10.480
But it still could be very real, right?
00:17:10.559 --> 00:17:14.960
The idea that people can be the the nature of as you free people up, right?
00:17:15.119 --> 00:17:19.519
The nature of roles can actually shift towards where judgment is more valuable, for example.
00:17:19.599 --> 00:17:25.680
And that may show up in more customers, it may show up in more innovation and products, it may show up much further downstream.
00:17:25.759 --> 00:17:32.880
And so I just raises to say I think ROI could take us into the world of focusing mainly on cost reduction and productivity gains.
00:17:33.039 --> 00:17:37.440
That may leave on the table a lot of gains that we could have that are much, much harder to measure.
00:17:37.680 --> 00:17:47.279
That said, I actually think if you want to think about measurement in a broader sense rather than just financial kind of ROI, you could think about this idea of capability building.
00:17:47.359 --> 00:17:47.599
Right.
00:17:47.680 --> 00:17:58.400
And I think if you look at uh Vinci, that's an interesting one where you the idea there is not just that the AI incubation center leads to like this one-off product, right?
00:17:58.559 --> 00:18:05.759
It's that there are other groups that benefit from that same underlying technology that you just seeded in that product, right?
00:18:05.839 --> 00:18:13.759
And so I think one measure for especially larger companies or companies that are doing this over time might be to really measure those capabilities and how they're spreading.
00:18:13.839 --> 00:18:31.200
And at Vinci group, this could be as simple as like how many other units are actually implementing products or improving their processes using that same underlying generative design technology or same underlying, you know, kind of contract language modeling technology or whatever else that they are building, right?
00:18:31.279 --> 00:18:35.440
You can actually see a lineage of people building on those capabilities, right?
00:18:35.599 --> 00:18:42.559
You may not see the financial benefit for a while, but that would be to me a leading indicator that you're building capabilities that are impacting the broader organization.
00:18:42.720 --> 00:18:53.680
So, long story short, I would just say that you could think about measurement more broadly than just ROI and think of it in terms of also the idea that you are really trying to build capabilities that help you in the future as well.
00:18:54.000 --> 00:18:59.519
If all the businesses have access to the same AI tools, where does competitive advantage come from?
00:18:59.599 --> 00:19:06.880
Do you think it's from data, from people, from processes, culture, or is there uh another answer?
00:19:07.359 --> 00:19:08.319
That's such a great question.
00:19:08.400 --> 00:19:09.920
And we we grapple with this all the time.
00:19:10.079 --> 00:19:11.359
I think you know you're absolutely right.
00:19:11.519 --> 00:19:14.160
Like the underlying models everybody can access, right?
00:19:14.240 --> 00:19:16.319
So you so it cannot come from that, right?
00:19:16.400 --> 00:19:24.240
Uh I think in my view, it actually comes from, I mean, we we talk about this in the course, but I think in my personal view, I think there's like three broad things, right?
00:19:24.400 --> 00:19:29.200
One is just the ability to execute and stay ahead of the curve, I think is a big deal, right?
00:19:29.279 --> 00:19:31.440
So even if you just focus on productivity, right?
00:19:31.519 --> 00:19:39.920
If if you're if you're in your industry, right, or staying ahead of the curve and integrating this in ways that are allowing you to be like I think of the Vinci thing I just mentioned.
00:19:40.000 --> 00:19:44.400
If you could bid faster, get get be more likely to be successful, move faster than others, right?
00:19:44.559 --> 00:19:45.200
Of course, right?
00:19:45.359 --> 00:19:49.119
If you could do that at scale, right, there's just an execution advantage that can happen, right?
00:19:49.200 --> 00:20:01.200
And if you if you believe like we do, right, that there is this growing gap between the capabilities of the underlying AI models and the organization's capacity to really absorb those for a typical organization.
00:20:01.359 --> 00:20:04.319
They're organizations that can bend their own curve and stay ahead of that, right?
00:20:04.400 --> 00:20:07.279
They should have an advantage just purely by execution alone.
00:20:07.359 --> 00:20:13.119
So I think that idea to execute at scale and to do to really be the forefront of that, I think that that's one source.
00:20:13.279 --> 00:20:15.680
The other, um, as long as that gap exists, that's a source.
00:20:15.759 --> 00:20:23.759
I think the other source of this is what you're hinting at, which is, you know, look, do you have some fundamentally unique data, right?
00:20:23.920 --> 00:20:31.839
And I mean by this like real data on your customers, data on how workflows happen in your company, all those things, right?
00:20:31.920 --> 00:20:35.200
That that the frontier model firms don't have, right?
00:20:35.279 --> 00:20:37.519
So that that's I think, or others don't have, right?
00:20:37.680 --> 00:20:42.720
Then that's something you could apply AI to and actually get competitive advantage on, potentially, right?
00:20:42.799 --> 00:20:53.440
So if you think about um, you know, I think there's so one person in your class I just found out about that she's her company is developing their own uh language modeling for language translation, right?
00:20:53.599 --> 00:20:56.799
And so, and I think, but it makes sense, their whole business is about that, right?
00:20:56.880 --> 00:21:01.039
And so they they will take the expense to train a frontier model doing that, right?
00:21:01.119 --> 00:21:11.839
And so I think that that is if they have unique data, right, on on uh, you know, translation nuances, for example, that happen on their platform, for example, then maybe that there's a benefit from there and you can go to market with that.
00:21:12.000 --> 00:21:26.640
If you're a cybersecurity company and you have a lot of unique data on, you know, what uh on cyber breaches, what's been successful, not successful, and you could train a frontier cybersecurity model because you have unique data others don't have, then again, this is something you can imagine could create value.
00:21:26.799 --> 00:21:33.279
If you have data on your own workflows that can be applied to help you drive those workflows more efficiently, that's great, right?
00:21:33.359 --> 00:21:38.960
So it's a unique data that that AI models can intersect with that you can maybe capitalize on in unique ways.
00:21:39.039 --> 00:21:39.920
So I think that's one.
00:21:40.079 --> 00:21:44.640
I think the other one that's more difficult and more interesting actually is the idea.
00:21:44.799 --> 00:21:47.680
It was this memory layer we talked about with the goose case.
00:21:47.839 --> 00:21:55.039
And that's the idea that every organization that's been around for any significant period of time has what we might call tacit knowledge, right?
00:21:55.119 --> 00:21:59.440
It's like the stuff that the veterans, your best workers, understand about the way things get done.
00:21:59.519 --> 00:22:04.480
It's a Stuff that's not just written down in SOPs and things like that gets passed from person to person.
00:22:04.559 --> 00:22:07.519
It's on the job learning, it's for through culture and that kind of stuff.
00:22:07.680 --> 00:22:12.319
I think these AI architectures that we're starting to look at in enterprises, you saw one in Ooh's, right?
00:22:12.400 --> 00:22:19.359
This idea of shared memory, that's where you could start to capture that stuff and it could become more usable more quickly by more people, right?
00:22:19.519 --> 00:22:27.680
I think that's the that's really the more transformative thing in a sense, because that's the stuff that's like really quite unique about your organization at some level.
00:22:27.839 --> 00:22:35.599
If your organization has had a historically, historical competitive advantage that rests on you operating in some fundamentally unique way, right?
00:22:35.680 --> 00:22:39.440
That's something that you can now speed up and capture with these kinds of architectures.
00:22:39.599 --> 00:22:41.680
I think that's really what the Who story was about.
00:22:41.839 --> 00:22:42.960
So that's there.
00:22:43.119 --> 00:22:54.960
On the data side, though, I would say like another place to look at this would be if you look at some of these AI native startups, this is like an open question because a lot of those AI native startups are not building their own models, right?
00:22:55.039 --> 00:23:00.480
But what they are doing is building architectures around these models that do tasks really well, right?
00:23:00.559 --> 00:23:09.920
And so you could think about one
would be well, again, my colleague Rem, he has this great case on this company called Gamma, and they are a startup, uh for startup that builds slides for people, right?
00:23:10.000 --> 00:23:12.640
And you might say, well, how can they be in business when anthropic?
00:23:12.720 --> 00:23:15.839
You could do that in anthro in Claude, you could do that in Chat GPT.
00:23:16.079 --> 00:23:17.519
Question is, can they do it better, right?
00:23:17.680 --> 00:23:25.359
By calling on the right models at the right time, by an architecture that has unique data because they work one of the first in this space, right?
00:23:25.519 --> 00:23:33.759
Can they do they actually understand the choices, right, that work when building slides for people in different contexts, right, better because they have this data, right?
00:23:33.920 --> 00:23:36.960
Can they build agents that then learn more and do a better job?
00:23:37.119 --> 00:23:46.160
That's an open question, I think, for any of these kinds of um, you know, startups that are effectively wrappers around the uh large language models rather than building their own models, right?
00:23:46.240 --> 00:23:56.720
Is it like what data do you have that would allow you to allocate intelligence, right, or tasks to intelligence, right, to the right model at the right time, given cost quality trade-offs and the like.
00:23:56.880 --> 00:24:02.480
Do you have some unique data that lets you do that better than the um large or large frontier labs could do?
00:24:02.559 --> 00:24:03.519
I think that's an open question.
00:24:03.680 --> 00:24:06.400
So anyway, if I'm stepping back, I would just say the three things, right?
00:24:06.480 --> 00:24:09.039
It's the ability to execute and stay ahead of the curve.
00:24:09.119 --> 00:24:10.559
That's one, and do that at scale.
00:24:10.720 --> 00:24:19.039
Two is again unique data that allows you to harness and leverage these underlying models in ways that competitors and even the model providers can't.
00:24:19.200 --> 00:24:29.519
And the third, which is really the unique thing, I think, is this idea of tacit knowledge, the way things are done in your organization, your culture, your unique ways of doing things that have led to competitive advantage.
00:24:29.680 --> 00:24:36.000
To what extent can you codify those things in a system that makes these outside models work better for you uniquely?
00:24:36.160 --> 00:24:37.519
I think that's the third thing.
00:24:37.839 --> 00:24:39.519
Thank you very much for going in such detail.
00:24:39.599 --> 00:24:41.279
It's gonna help the listeners a lot.
00:24:41.599 --> 00:24:42.000
Great.
00:24:42.240 --> 00:24:45.039
I want to move away from AI to case research.
00:24:45.200 --> 00:24:56.240
I was curious how you go about identifying topics to research on, companies to write cases on, and what that process looks like without giving the secret source away if you're comfortable talking a little bit about it.
00:24:56.400 --> 00:24:57.119
So case selection.
00:24:57.279 --> 00:24:59.920
So the question, how do I select cases or think about that?
00:25:00.160 --> 00:25:00.400
Yeah.
00:25:00.559 --> 00:25:02.880
And what does the process of writing a case look like?
00:25:03.200 --> 00:25:04.720
So case selection is interesting.
00:25:04.799 --> 00:25:06.640
You know, I think there's several ways you could do that, right?
00:25:06.720 --> 00:25:10.319
One would be look, there's a topic we really need to understand, right?
00:25:10.400 --> 00:25:16.079
And so, you know, we need to go out and try to get context of the company or write a public source, right?
00:25:16.160 --> 00:25:19.119
So again, this might be we we need a good case on AI safety.
00:25:19.200 --> 00:25:20.480
Let's look at what anthropics are doing.
00:25:20.559 --> 00:25:27.039
Let's figure out like, you know, kind of then and you do know what you're you know, kind of like you need to go to a particular company, try to write the case.
00:25:27.200 --> 00:25:33.519
I think the alternative is, and this is what I've tended to do, is you really want to think about uh synergies with your own research.
00:25:33.680 --> 00:25:35.519
Like, what are you trying to understand in the world?
00:25:35.599 --> 00:25:44.319
And you might be doing like traditional research with large sample data asking a research question or even with company data, but you know, you could only go so far with traditional research.
00:25:44.400 --> 00:26:00.559
And if you want to understand mechanisms, you want to understand like the open questions, the kind of boundaries of a particular theory, like getting out there and really doing a case and intersecting with an organization, understanding how leaders are making actual decisions, how those decisions might be showing up in the data that we're looking at, right?
00:26:00.640 --> 00:26:02.480
That kind of thing, that that's really helpful.
00:26:02.559 --> 00:26:08.960
And so, so I've I have a number of my cases have been built around particular research projects I was doing.
00:26:09.039 --> 00:26:20.799
So, for just for example, even prior to this AI moment, you know, I was quite interested in, you know, how how leaders can build systems to enhance collective human judgment in their organizations, right?
00:26:20.960 --> 00:26:25.440
So this gets into things like empowerment and how do you measure risk of empowerment?
00:26:25.519 --> 00:26:33.039
How do you actually create the right motivation systems and the right systems to give people the best information to use their judgment and that that sort of thing?
00:26:33.119 --> 00:26:34.319
So how do you build that at scale, right?
00:26:34.559 --> 00:26:43.920
I was looking at a lot of those questions, and you know, I I spent a lot of time doing cases on just identifying companies that were more distributed in their decision making than their peers, right?
00:26:44.000 --> 00:26:48.720
I'd pick banks that are radically decentralized
relative to others in their industry, for example, right?
00:26:48.880 --> 00:26:53.920
Or, you know, could be hospitality, could be manufacturing where people are more radically empowered uh than others in there.
00:26:54.079 --> 00:26:56.880
And then I would want to look like, how do they make a system like that work?
00:26:57.039 --> 00:26:57.200
Right.
00:26:57.279 --> 00:27:06.160
And then you can end up with a lot of really interesting choices and a lot of controls that look very different than traditional ones, incentive structures that look different than traditional ones.
00:27:06.240 --> 00:27:10.720
So you actually learn a lot by if you have a question in your mind, right, about some phenomena.
00:27:10.799 --> 00:27:16.640
And in my case, again, this was this human judgment at scale thing, how do leaders build systems that harness that and enable it?
00:27:16.880 --> 00:27:28.160
So I did a number of cases looking at that in detail, and I ended up doing a number of research projects around that too, looking at various like small pieces of it to see how those choices could actually impact performance.
00:27:28.319 --> 00:27:33.039
So, for example, it might be how uh systems that capture what we call soft information, right?
00:27:33.200 --> 00:27:42.240
Soft information on how people actually make decisions, why they, what trade-offs they're actually making, how though that can actually enable others in the organization to make even better decisions, right?
00:27:42.319 --> 00:27:43.759
Once you start capturing that data.
00:27:43.920 --> 00:27:44.480
Things like that.
00:27:44.559 --> 00:27:51.920
I look at things like organizations that had much more structured, systematic ways of selecting who fit at the higher stage than didn't, right?
00:27:52.079 --> 00:27:55.440
How did that play out in terms of processes and control going forward?
00:27:55.680 --> 00:28:03.839
So I look at very specific things like that, but I was looking at cases in a as a way to understand what are the various levers that leaders are using to make organizations like that work.
00:28:03.920 --> 00:28:06.319
Uh by the way, that's informed some of my AI work now.
00:28:06.480 --> 00:28:15.200
One of the things I really spent a lot of time, as you might, I you may have heard, right, in the Vinci case is kind of thinking about what are the possibilities of reorganizing around AI, right?
00:28:15.359 --> 00:28:26.000
That AI has great potential to give speed up people's access to expertise, great sort of uh potential to speed up the information flows inside of organizations.
00:28:26.160 --> 00:28:31.359
But none of that matters, right, unless organizate leaders just redesign their organizations to capitalize on that, right?
00:28:31.440 --> 00:28:40.240
And so to allow people more agency to act on their expertise, to build with these tools, to react to you know, work that's speeding up in other parts of the business, right?
00:28:40.400 --> 00:28:42.319
And so this is coming back full circle.
00:28:42.400 --> 00:28:50.079
Much of my work now is looking at some of what you might think of as these older topics of organization design, but actually how they have to be adapted now in a world of AI.
00:28:50.240 --> 00:28:55.680
So I think it's that sense of a phenomenon or open question that then leads you to interesting organizations.
00:28:55.839 --> 00:29:01.200
And final thing I'll say on this since this is an OPM crowd is that OPM has been a great source of inspiration for me.
00:29:01.279 --> 00:29:06.079
I taught uh this was my main teaching assignment for five years in 2015 to 2020.
00:29:06.160 --> 00:29:10.799
And I have interacted with so many people from that, uh, those cohorts still.
00:29:10.960 --> 00:29:17.279
And when I hear about them, like that, I teach about these ideas, and then they'll tell me, hey, here's what I'm doing, or here's a problem I'm grappling with.
00:29:17.440 --> 00:29:20.000
And then I immediately hear like this would be a great case, right?
00:29:20.079 --> 00:29:30.319
Either you did something like really interesting, I don't want to document it and sort of like see what the possibilities are, or you're actually really being serious about a problem that you're facing and wanting to address it.
00:29:30.400 --> 00:29:37.039
And then it helps me write a case that could put our students in the shoes of leaders like you that have to grapple with that and think about design choices around it.
00:29:37.119 --> 00:29:42.640
So this has been an inspiration for me as well is actually teaching exec ed, especially in programs like this.
00:29:42.799 --> 00:29:50.160
And then you're you're presenting your ideas, you're getting pushback, you're also getting insights into what's working and not working around those ideas for people.
00:29:50.240 --> 00:29:54.000
And so that's that's often led me to cases specifically with OPM participants.
00:29:54.240 --> 00:29:54.720
It's amazing.
00:29:54.799 --> 00:29:56.240
Thank you for sharing that insight.
00:29:56.400 --> 00:30:00.559
I think we've all always been curious about how HPS goes about writing cases.
00:30:00.880 --> 00:30:01.599
This helps a lot.
00:30:01.680 --> 00:30:03.759
Uh, Dennis, in closing, I have two questions.
00:30:03.839 --> 00:30:10.400
Uh, firstly, if if you were to sit with a younger version of yourself, knowing everything that you know about life today, what advice would you give him?
00:30:10.640 --> 00:30:11.200
Boy.
00:30:11.519 --> 00:30:12.799
That's that's a good one.
00:30:12.960 --> 00:30:16.960
I'm pretty happy with the way that things have gone career-wise and what I've chosen.
00:30:17.039 --> 00:30:22.480
And I would say I do think that this idea of thinking about think about what excites you.
00:30:22.559 --> 00:30:24.640
And I know it's like it's like tried advice, right?
00:30:24.880 --> 00:30:25.920
What are you passionate about?
00:30:26.000 --> 00:30:33.200
But what I would say is actually act
on that earlier and like really talk to be try to be around people that inspire you in that area.
00:30:33.359 --> 00:30:40.640
I do think the people you intersect with and who you keep as your close circles, I think that has much more of an impact on you than I would have realized at that time.
00:30:40.720 --> 00:30:59.440
And so I do think like it doesn't mean you have to know exactly what you want to do in the rest of your life, but but follow your gut on what you're really curious about, what you want to understand about the world, and try to be around people that are serious about the dealing with the same kinds of challenges that you're interested in or have been successful in areas that you want to explore, right?
00:30:59.519 --> 00:31:00.400
The more you can do that.
00:31:00.480 --> 00:31:08.559
Um and when I look back, I you know, I did, I I did like one of the things that led me to an academic career is I actually made an effort even when I was um in undergraduate.
00:31:08.720 --> 00:31:15.119
I spent, I probably going back, looking back on I spent way more time with my professors than I even did with some of my peers my own age.
00:31:15.200 --> 00:31:18.000
And I think like that was like that was beneficial in a way, right?
00:31:18.079 --> 00:31:22.000
I mean, I kind of like I was fascinated by what they were doing, what they were working on.
00:31:22.160 --> 00:31:23.359
I wanted to learn more about it.
00:31:23.440 --> 00:31:38.160
And I would say I would I would be more intentional about that, I think, if I was going back that really thinking about who curating who you want to really spend your time with and think about that in relation to how at least in the moment, how you're thinking about your future direction, what fascinates you.
00:31:38.240 --> 00:31:41.119
I think that's that's maybe more important than I would have realized at the outset.
00:31:41.519 --> 00:31:42.319
Thank you very much.
00:31:42.559 --> 00:31:46.480
Lastly, is that a book you've read in your life which has influenced the way you think?
00:31:46.799 --> 00:32:01.039
Actually, uh one of my favorite authors is this uh Turkish author named Orhan Pamuk, and he uh I think writes like these really deep novels that are set in a time and place that really makes you feel like you're there and experiencing it.
00:32:01.119 --> 00:32:12.640
And I think it's like uh and it is his his books tend to do that in a way that helps you understand differences between cultures and people in ways that are really interesting and mindsets, right?
00:32:12.720 --> 00:32:30.319
And so I actually like that that that those his series of books actually have influenced me to be like I read some of these when I was younger, to be more open to the world and to very, very different ways of looking at the world coming from forces you may not understand about heritage and history and where you are and all of that.
00:32:30.400 --> 00:32:42.880
So I think that, yeah, for me, not business related, but I think just more general beautiful writer, beautiful books, and gives you a sense of place, but in a way that helps you respect and understand differences in people and cultures and everything else.
00:32:42.960 --> 00:32:44.480
So thank you very much, Dennis.
00:32:44.559 --> 00:32:47.680
These insights are very valuable and it'll go a long way for us.
00:32:48.000 --> 00:32:48.319
Thank you.
00:32:48.400 --> 00:32:52.640
And I appreciate uh the opportunity and I hope that I can interact with all of you uh going forward.
00:32:52.720 --> 00:32:53.279
So thank you.