Summary
Notes
Transcript
All right, so my name is Mike Leppitt. I am a faculty member in the school of engineering and for student sustainability. If you haven't been over to our side of campus, it's on the other side of the country club. Beautiful new buildings as well. And please feel free to walk over and explore. It's for our building tonight. And go over and check it out. So on that side of campus, I have a number of. All right, so.
I run a research center. I guess I'm a professor with a research group that focuses on multi-scale design that can save us from those environments, that goes from nanoscale, where we develop new construction materials and things, all the way up to smart cities and smart urban systems. Also our research center called the Center for Sustainable Development and Environmental Health where we have for the past 30 years applied advanced computing to helping organizations all around the world, particularly in developing countries, become more global and competitive economically while also not sacrificing or damaging local environmental social ecosystems.
a very large executive education program over in the School of Engineering, including our flagship product management program, where we have trained thousands of product managers or aspiring product managers all over the world How to engage in that practice.
Product managers all over the world and how they are using some of these tools, what works, what doesn't. A lot of this content is also based on a lot of conversations and research that I have done with one of our Center Fellows, Jeff Wong. Jeff ended his career after a variety of different staffs in Silicon Valley as Chief Innovation Officer for EY. where he built their AI practice within UI. And in that role also saw the technology adoption and AI practice of many other firms.
that have tried to do many of these things. And so from a lot of our research, talking to those folks, Fundamental work, talking to lots of product managers, trying to come up with this content. I'll talk about that in a minute. And then-I also run the School of Engineering's Venture Capital Investing class.
And most of this content is also pulled from a forthcoming book that is being reviewed by editors right now that Jeff and I are writing together on a lot of this topic.
But I want to talk about how, I want to start off by kicking off about how big of a deal this is. So how many of you are familiar with the Haber-Bosch process? The most important process that you can never forget. Early 1900s, 1909, Haber figures out that You can fix nitrogen From Sydney. Now, up until this point in human history, Which like, that's a big state, right? Up until this point in human history, Nitrogen as a nutrient or fertilizer is the limiting agent on population growth.
We fight wars, we dig up cemeteries, we do whatever we can as a species. to get nitrogen. And What happens over time, and you can take this back, 10,000 years-ish? We got a slope line. And then... We figure out we can fix nitrogen out of thin air. And then Bosch, two years later, industrialized it. And this is the actual population This is what it was expected to be. without that process happening.
We have an extra 4 billion people on this planet because of this one discovery and its industrialization. And every single one of you, most of the nitrogen in your body, including mine, comes from that process. just by the fact that you ate food that was grown since 1909. You can see the tape. But he's a big deal in the history of humanity. This little chart shows population. 10,000 BC? To 2000. And it's rare that you see a chart with a timeline that long.
We were looking at global brain power. Four times. And you see two very distinct curves. I'm Superhero King. Haber-Bosch. And the swamp. This is a big deal. in the history of humanity. How big of a deal is this? For the first time, we know no longer in humans to increase the supply of brain power. We can convert energy directly into intelligence. That's crazy. That's why we see this amazing king cabinet.
with only 3.8 billion.
That's a 142 times reduction. in the amount of resources needed to solve the same problem in two years. That's not kind of a video. That's EarthShower. And it's also getting better. They're getting cheaper. And better. So in 2023, Sweet bench verified. Actual GitHub issues that were correctly solved by AI. And in 2023, it was about 4%. That same test. In 2025, 2026, when to 70% or 80% success. So less resources.
more accurately at solving the American problem. But at these rates of improvement,This is remarkable. And yet, 85% of big data and large projectsThis is failed. This is for Gardner. MIT also had a study that said 95%. I don't care if it's 85% or 95%. That's way too high. So we're taking this amazing technology Which is truly world changing. And the vast majority of applications are failing. And when I say failing, I mean, like...
88% that have adopted AI, great. A lot of folks are adopting it. What exactly that means, eh, we can talk about that. What I'll bet on is 6% report a significant Ebit's impact. So everybody's doing it, but nobody's benefiting. Nobody's creating value, nobody's capturing it. And that's where we start. What is that? That word significant Yeah, so... measurable to hit their ROIs, Now, this brings up a great question. What do they consider failure? What do they consider significant?
There are a lot of problems with those definitions. One problem. Is that a lot of existing expectations around ROI related to AI adoption are wildly out of reality.
So where did we find-It's telling you why it's not in the technology itself. It really is in the law.
So let's start with a planning framework for AI implementation. I'm down here now, Phil. I said I like frameworks, so you're going to get boxes and things like that. Like this. One of the things that we study right now in my areaYeah, public research. is high performing organizations. And the reason we are connected with this is because When you're doing construction or when you're building things, You do it once Everything is a prototype. We're dumb enough not to do the same thing twice.
In a regular business, you make a prototype, you get it perfected, then you scale it and sell that same thing to a million people. Okay. In construction, we're going to make one of them and then we're never going to do it again. So how do you get that? Well. You have to get better by not including the product,Put it in the bag. Organizations, processes. And so the three levers that we talk about in high performing organizationsParked.
Organization. And process these for workflows. This is what we call a pop culture model. I'm an engineering professor. I don't know anything about pop culture. But this is the closest I can get. We also thought about calling it product people execution. That's key D-E. After COVID, no one got it out. And these are really the three major levels. that we have in any organization or entity. to try to get to a high performance.
There are three basic things that we want. The first is what is its function? What do we want to do? The second is its form. What does it actually look like?
And the third is The behavior or performanceThat we're going to use to measure. Whether or not thatThe role was actually achieved. So in today's This is a culture model.
How will we measure whether or not We achieved Is it the number of hours that people are actually just using it? Maybe it's ROI, but maybe it's not ROI. And your point is well taken. Sometimes we're doing stuff for reasons that don't really fit with... The targets are-were measured. And so therefore, our signaling is going to be off. And this is where most funds Stop. Get the tools. They pilot the tools, they implement the tools.
And it stops there.
And look, they're going to get sent back anyways. We might as well have you spend next to zero time actually writing the initial submission, right? Then just fix it when it comes back. We're just going to try to save your time. So she started to do this, and he just popped up. Top it and paste it into the Oracle. Amazingly, None of them are coming back. Yeah! This is neat. So... A month later I was up.
We have some folks in financial management services and they're like, "We also did, we programmed up an agent to read."And compare them to Stanford rules. Great. Classic. And because of this, we're getting them through faster, which is wonderful. I love that. But now, what are we doing? I am paying Lee to copy. From Claude into Oracle and then somewhere in Redwood City is getting paid to come. out of Oracle into whatever they're using.
It's like, huh? And amazingly, he could sell Parabox. Oh yeah, I did a good job writing this. Okay, so this is the big day. Oh, look, we tried to, Introduce the new products. I introduced This new product internally to my team. We're going to use An LLM to take care of this work. We did not change the organization at all, did we? Absolutely not. So probably three before we get out of here. And we did not change the promise.
What we did was we tried to bolt A couple of new tools, one in financial management services, one over here in engineering. We tried to bolt them These two new products onto the existing set of organizational structures and workflows and think that it was gonna be great. Does it seem to be a little tight? Is it getting us the kind of amazing improvements that we should expect from AI? Absolutely not.
So the first impairment that we found to be successful, and also not at all disappointing,Commissioner, a clean sheet redesign of your three highest value processes for workrooms.
And think about a clean sheet redesign of how AI changes that workflow.
So it starts-and so we've got four stages of AI implementation. per unit of human thinking on a log scale. If you have humans doing everything, and this is where we have been for many, many, many It's a one-to-one ratio between what humans do, and the output per unit of human being. Okay. We'll all kind of walk through. Stage two is what we call electrification. That's where you get about 1.5 to 1 leverage ratio on the use of AI. That's what we're doing with me. Thank you. We're getting a little bit of benefit.
She's not spending as much time. As she used to. Are we getting amazing returns? No, we're getting 1.5 to 1 leverage out of admin. That's not... Earth-shattering for any kind of farm. But this... Is bolting The electric motor onto the line shaft. This is introducing the product here. without thinking about anything here.
And this is where, honestly, that 88%Of companies, we were talking about that gap. That's what the 88% look, right? They said, look, we're not getting any significantAnd why? That's not significant. That's a little better. And so to get to your question of significance, Let's watch.
This is where we see as we look at a lot of firms, about 5% of firms, and this is that roughly 6% that are actually seeing real benefit.
Please follow theSo here's where we start to see real leverage. in companies that have implemented an AI first strategy. This is where you start to see 10 to one output When one word of human thinking has been lost.
And what do I mean here? This is where humans are primarily judging.
The output of AI.
This is Lee uploads all of the raw receipts and then looks And says, oh yeah, that looks okay. Or she just uploads it and someone in financial and management services looks at it and says, yeah, that looks about right. or we've had a lot of fraud detection, also take place using AI. In the most And so in this case, you're using humans to judge the work.
We haven't got the organization signed yet. And where you really get leverage is what we call AI movement, where humans simply set the systemN.A.R. then runs.
This is where I upload my receipts. We're 15 minutes later. It's in my Stanford credit union account as a university.
So one of the organizations that we're working with is Silicon Valley Company here live at the high. They make holograms for AI.
But what they're moving forward with is a complete AI native Workflow. Where every single man on the farm comes in, AI handles the workflow. And then maybe there's a review order. We'll stop at that. Also, Insurance and loan provider, China PN. They dropped their Cost to loan issuance ratio by 700 times by going to a full AI system doing small microlobs. Stop big stuff. Because you're still the man who wins. Can I ask you a question? I'll be the first person to say it. Most folks are here. You're trying to be cute.
Is your metric of success X percent faster or Y percent cheaper?
We're talking down here, right? If the answer is yes, you're electrifying your line shaft. Stop now.
Because you should not be measuring real improvements in AI acquisition in percentages. Multiple. Because if you are... We're doing this and not this.
This is success method exercise faster, wide recension. If so, you're electrifying. You're building a new order on your old line shape. But if you're truly changing things, Focus on it and be skeptical of just electrifying what you have to do.
The reason, and this is kind of a broader comment on change management. Aussie. No one is... predisposed Two against change. Change half a goal. And when you've seen fear change, This, by the way, is something you can be somewhat introspective on, too. The reason that people are afraid of change is because we're always afraid of losing something. Something we have. It could be the fact that I'm an expert in this area and now I'm not.
It could be the fact that I'm going to have to learn a new tool and therefore I'm not going to be able to be at home for dinner with my family because I'm going to have to spend more time working.
And so whenever you're engaging in change management, your role as a leader is to try to get down to what exactly is it that you're afraid of losing?
And how do I mitigate it as much as possible? Now, you can't always mitigate it. But if you can really get to the core What it is that folks are afraid of losing. And look, We're all afraid of AI. Every single one of our graduates is afraid of AI. What are they afraid of? They're afraid all of this time, effort, money that they've spent is completely worthless. That's a loss. five years or four years or whatever of their life.
But if you truly think about what is the role of our organization and is part of that role psychological safety? is part of that role A place where you're going to work and be proud and things like this. And yes, that's not socialism. That's having a job that people want to go to and work at. If those are the goals, How do you intentionally design these things to hold on to that site? So if we ask The first question from your..
So let's now talk about the organization. In many regards, a lot of our structures and organizations is a pretty old thing. So this is the structure of a Roman legion. And it was built on a seven-person tent because that was about the number of people that a single individual could manage. And if you hear a lot of-leadership coaches, how many direct reports you should have, it's probably right about that same number.
And then yes, it goes up to roughly 100 men led by a centurion and all this other stuff, right? And so why do we do that? Well, we do that to have effective control over large numbers of people. so that we could focus on process and scalability. And this is the root of modern middle management. And it was incredibly effective. We're all having a good run. But AI changes the scarcity. of Animal capacity, information processing, and coordination family.
It depends on what you're trying to go fast on. Especially if it requires that, because you've got people in the mix. Because we're slowing down. So, Roman Legion. Thank you. Looks about right. What we're seeing nowHave you been talking to a bunch of folks in the US phone center? is a transition from that What we call an hourglass model. of organizations.
You need people here. To build trust. to interact with customers, to be those frontline folks. We need folks at the top. Because That's vision and judgment up there. That middle layer of coordination He's really good at that.
Have you seen in the 88% of the companies that haven't get anything that they delegated to the community?
Yeah, that's right, and that's where it goes to die. Why? Because that's the part that smells fragile, right? to this whole thing. So they say, hey, let's get these tools to middle managers and see what they do with them. I can tell you what's going to happen. It's not going to work very well. There's two hundreds of scale.
Yeah, so these are gonna be like your SM, right, your subject matter experts, depending on exactly what your field is, right? But you're gonna need folks that ultimately are gonna apply that judgment to whatever it is that's going on. But folks that truly have expertise in that sense. Tell me how many of those you need in four. Now, one of the things I want to say is this is not a recipe for elimination of jobs.
That's not the goal here, honey. Rather, this is an opportunity for redesign of work. Absolutely you won't vote for this. Does this person's role add judgment? Or does this person just add information? AI is a way to the information. It's not very good at judgment. If the answer is primarily information, aggregating data, relaying reports, translating strategy into tasks, The rules are good candidate for redesign.
either pushing up or pushing down. Not elimination.
And the answer is yes, you're gonna have to do some of it. What a good week. So, pattern number three, audit every layer of your organization with a judgment or information test. And it doesn't mean everybody gets fired right away, but you start out that way. What does this layer of management do?
And it was growing faster than Netflix by far.
So much so that Reed was ready to sellThe blockbuster. Because he's like, look, we cannot compete.
And then Carl Icahn comes along. Carl was, had built up a lot of shares and was on the board. Allah. And he's looking at blockbusters income statement, that's got $800 million in late fees.
That's a good cash. And John Mancioco, the CEO at the time, knows that total access is the future.
And so he gets the rest of the board to kind of go along.
And so then they hire a ex-CEO of 7-Eleven who has no experience in media, and they kind of go back to the blockbuster model, and three years later they're back. 66% of board directors have limited or no knowledge of AI. Six percent have a dedicated AI community. This is the return on equity spread. between companies with an AI savvy board, which has at least one AI savvy member on it, and a committee, and those without.
That's a big, big difference. And clearly number four, reconstitute your board for the AI era. You're going to need people. on that board that understand that these things will have to change as well.
AI transformation demands someone with energy to pursue completely new transformation in their life. So James Quincy of Coca-Cola stepped out and said, this is not my thing. I can't do this. Those that are deeply engaged with AI are 12 times more likely to leave companies than top 5% of innovation. And the CEO's got to personally and data-driven real work and have 30 days of making strategy-based decisions on second-hand or third-hand information.
This goes to, do I have to do it? Yeah, you got to do something. You as CEO or leader, you have to build your personal AI practice with your hands on the keyboard. If you're scared of using it, get on the scale of using it. Because you have to know what people are talking about. You have to know what it can do and what it can't do.
And you will only become comfortable with it by doing it yourself.
Don't get a briefing on it. Don't have somebody tell you this is what I can do. Mm.
What we found is that you have to start to think about a 24 month Transformation timeline that even loops again and again. Because that starts the foot. with the speed and scale at which these things are changed.
Form that competitive mode because you're going to be out in front. Should we move fast? and help that organizational learning Lock in. You're going to want to think, and we're talking about somewhere between a 1 to 3% of revenue investment into this. with roughly 2% of it. You're doing 3%? Well, if the 2% are going to the technology, 1% going into all the order change and process change. Because it can't just be the technology.
And it takes resources. We get to know each other. So instead of 24 months,I'm gonna make it a board level commitment.
And finally, when it comes to behavior, So we design great things. So we have 10 questions that are really based on all these imperatives. Three questions for the organization. One strongly agree, the five strongly agree. You get 40 and 50, you redesign it. 45 to 39, you're kind of electrifying with intent.
What 25? You're just putting electric motor under the line shaft.
You're going to get all these slides. But these are all basically the things that we just talked about. Our three highest-value business policies have been examined from scratch. More than half of our advisors target fundamentally different workflows. A single leader owns organizational redesign, not just technology deployment. Don't make this the job of the CTO. CTOs don't tend to know about this stuff. Why? Because CCOs tend to come from my afghan. We're geeks. You should do it then.
So it should be someone who reports directly to the CEO or the CTO, but not the CTO.
All of these basically come out to what we had just talked about. And screw yourself up. Where are you? So score yourself honestly. Because It's up to you to line yourself if you want to. Thank you. So those are our seven original big contractors. Are everyday organizations, is this judgment or information? Thank you, Mr. Gore. Is it a high performing work by these individuals who know what the future of these technologies are going to look like?
The other thing I did say, Is that... Without that layer of middleman, to help with that reduced learning management, to help train our young people, the role of mentorship. at firms will become much more important. Because honestly, the only way that you learn how a firm does its work and ethics and all the things that we're so worried about.
New hires using AI for that's always come through mentorship. And sadly, this technology followed a pandemic where most mentorship programs got decimated. And most in-person contact between senior folks and junior folks got pushed to the side because work from home is easier. But not better.
So the first thing we're doing is we're going to put a paper that summarizes a lot of this up on SSRN, so it's on the research network, because that's where this will go.
That will sort of put a stake in the graph with the publication out there and the data that they have and all of the necessary stuff. So we'll put that up, and I'll be sure to let... Please. It's where you can put papers before the free prints and stuff like this. Is there any source that you use like podcast?
which is why some of these simple tools like use it yourself You don't have to implement the redesign. Think about the redesign. So then when the time comes that you're gonna have to do it? Okay. There's a variety of things that you can do if you can't pull the trigger right now. Does that help, Carl? No, thanks. All right. Thanks for staying. Thank you. Is that a reporter? No problem. Thank you very much.