The Human Architecture of AI Transformation

Ben Jones: Hello everyone, welcome back to the Powered by Data Show. I'm Ben Jones, your host. Real excited about today's show. This month, we are joining Patrick McGarry, Federal Chief Data Officer for ServiceNow, coming out of DC. So hi Patrick, how you doing?

Patrick McGarry: Good, thanks for having me. I'm really excited ⁓ to take part in this conversation. It's exactly what I'm looking for.

Ben Jones: Perfect. Okay, good. Yeah, a little bit of backstory, folks, for those of you, how do I know Patrick? Well, Patrick and I go way back to when he was running communities for the data.world team. and I really felt a lot of kind of alignment with them. This is when I was running the Tableau Public platform, and a lot of what we were doing was building tools that anyone could use in the public domain. And so Patrick and I got a chance to meet, we got to know each other, and we've kept in touch ever since. Of course, you know, ServiceNow.

⁓ acquired data.world about a year ago. And so Patrick has grown into that role. but the reason why we want to talk to him right now is because he is going to join three other authors as the fourth author in the data literacy press family of authors, coming up here real soon. We're a matter of weeks away from the launch of his book. In fact, I have a pre-release copy of it right here. It's called The Adaptive Organization: Leading Change in the AI Era.

So gonna be hot off the presses here in a matter of weeks, already available for pre-order. But ⁓ so welcome again, Patrick. And you know, thanks for writing this book and for going with us on it. I'm just curious, you know, kind of talk to us a little bit about, you know, your career and how it led you to this place where you thought this is a book that needs to be written and I'm the right person to write it.

Patrick McGarry: Yeah, I mean this sort of happened almost by accident. ⁓ you know, I hate to say it that way, but like, you know, spent a long time in the data and and sort of tangentially AI space even before it was, you know, the chat GPT era, you know, when when AI ML and deep learning and like all of the different flavors weren't lumped together as just whatever LLM was the flavor du jour. but you know, being in that space for a long time, it's been interesting to see how the

industry has evolved, and then how it kinda got lit on fire by the advent and explosion of LLMs. but really what triggered writing this book was, you know, like as you mentioned, data dot world got acquired and there was sort of a little bit of a blackout period where we weren't doing a lot. We were kind of pencils down. and I I just figured, hey, you know, I'm gonna go through and write a bullet list of all the things that I help our customers and prospects and and

you know, community to think through, to figure out. and it ended up being like a thirty six page bullet list. And I was like, Hey, you know, if I add a few connective words, that's probably a book. so that that's why I say it kinda happened by accident.

Ben Jones: Kind fell into it. I love it. Yeah. an accidental author. Well, that's perfect. So talk to us about the central argument here in the book and the framework that really the book, I guess, is built around this catalog framework, which I think is appropriate given your time ⁓ working for a data catalog and semantic layer company. but talk to us about that framework, talk to us about the domains that build up to it, and also I think kind of.

interestingly and really connected to what we do at at data literacy, that first letter C being all about culture and talent. So kind of give us an overview of the catalog framework and how you kind of develop that and what it's all about.

Patrick McGarry: Yeah, I mean the I'll work my way backwards. So like the catalog framework, you know, I'll say the framework itself is less important. Like I didn't make this framework because it's going to be a rigid thing that everyone should use and deploy as is, but it's supposed to be sort of a loose guideline set of guidelines, right? For the things that you need to think about when you are trying to modernize data and AI. So the sort of central argument of the book.

Right, is that most organizations are approaching AI and even before AI, the data transformations sort of the same way that you approach a renovation, right? They're gonna pick the most visible thing to fix, they're gonna hire specialists to fix it, and they're gonna assume that the rest of the house will just sort of adjust, right? but it doesn't. The walls are connected, the plumbing runs through everything, and the people live there actually, you know, want to change how they work, or the you know, the renovation sort of sits unused, right? So the book

makes the case that transformation fails not because the technology is wrong, you know, we have the right tools, we have the right data, but because organizations treat it as a technology problem only. The the real problem is much more systemic, right? It spans culture and talent and governance and strategy and operations and what have you. and

You know, I like to say and I'm not the first one to say this, but I love to borrow this phrase that it's a socio technical problem, right? You have to have the people and the technology all rowing in the same direction. ⁓ and so there's kind of two parts to this. Like first is the governance is infrastructure, not overhead argument. you know, organizations that embed governance into the design of their AI systems from the start are gonna outperform those that bolt it on later. for the same reason

you know, Deming historically argued that building quality into manufacturing processes costs a fraction of inspecting and correcting defects after the problem. Right. The constraint in almost every organization is not access to models, it's the human architecture around the culture that either sort of enables or resists change, right? The talent strategy that either builds the right capabilities or doesn't. and it's the leadership that sort of

creates that clarity or generates confusion at this point.

Ben Jones: Yeah, I love that connection to Deming. You know, having been kind of in the Lean Sigma change management kind of world for many years earlier in my career, this idea of you know slowing down to go faster, where, you know, if we build in detection and prevention up front, we can avoid some of the costly errors down the line. Even, for example, the Toyota production systems, popular andon cord

system where anyone on the line could pull the cord to stop things. And then as you mentioned, the culture of root cause analysis and continuous improvement allows them to fix once and for all a problem that will never surface again before the production line even really gets running. So, ⁓ you know, it doesn't take long for that kind of a culture to produce better products faster at lower costs. And the same thing, I think, is

Your argument here is the same thing as applying to AI, that, you know, organizations that spend that time to get it right are going to see the benefits down the road. But what if there's an executive out there, let's say, who's thinking to themselves, well, you know, governance maybe to this CEO with a board of directors breathing down their neck. That's a dirty word. That's going to slow me down. It's going to cause me to lose to a competitor who's a little more reckless that moves faster and such. So let's address that.

pushback perhaps. What if there are listeners out there who say, you know, governance is just gonna slow you down and you need to run as fast as you possibly can and just wait for the models to push you forward and just adopt the latest and greatest and it'll all fall into place. So talk about or you know, address that maybe counter perspective to what you're arguing in the book.

Patrick McGarry: Yeah, I mean the whole, you know, governance versus speed argument, you know, it's a fair challenge. and I I wanna take it seriously rather than sort of dismiss it. But I I also like to challenge myself, right? And this is where I would say that my advice here ⁓ is either wrong or at least incomplete, right? in a genuine exploratory phase, ⁓ the place where you need to run fastest, you're gonna be doing things like rapid prototyping or

you know, contained sandboxing and and the stakes of a wrong output are pretty low, right? And the goal really is learning. So rigid governance in that context is the wrong instinct, right? If you're running a ninety-day proof of concept with a small team with an internal use case, you don't need that weight, the boat anchor of enterprise governance. You need just enough

guardrails to sort of learn quickly and enough documentation to sort of industrialize it if it works. So the problem is that organizations are really bad at sort of maintaining that boundary, right? What starts as a contained experiment has a way of becoming operational infrastructure before anyone has sort of made that decision formally. And that's when the governance debt becomes really acute. Like you are now running a consequential system that was never designed to be one.

So I guess the more precise version of my advice should be governance should scale to consequences rather than calendars, right? Low stakes experimentation warrants a light process. High stakes production deployment warrants rigorous process. the organization that sort of conflates the two in either direction, either sort of governing prototypes as if they're nuclear facilities, or running production AI with prototype level oversight.

is gonna be one that gets into trouble, right? So speed and governance are only in conflict if you sort of apply them at the wrong scale.

Ben Jones: That's brilliant. I mean, I love that framing. and you're right. I mean, even if I think back on my career, you know, plenty of instances of some Skunk Works project that some executive got wind of and just pushed out the door, and that would be an example, and then the whole thing goes awry, which is the example, of course, of the experiment that gets out of the lab, you know, without the right you know, kind of system in place to maintain it.

And of course, have also been in regulated environments where you know everything grinds to a halt. I mean, even just getting drawings done for a prototype just causes everything to really, really slow down. So it seems like then, you know, an organization almost needs to have two identities, right? Some sort of ability to fail fast in experimentation without needlessly slowing things down, but also the ability to ⁓ scale something responsibly and carefully.

So, how do you have both as an executive? How do you kind of cultivate that?

Patrick McGarry: I think the interesting thing here is to go back to the the sort of deming logic, right? And I really want executives that are thinking about this stuff to sit with that for a minute because the challenge here, the danger here is that they read this and they think of it as an ideology. And it's not ideology, it's economics, right?

Deming showed in manufacturing that the cost of building quality in is is a sort of a fraction of the cost of inspecting and correcting. And you know, the ratio isn't subtle. It's an order of magnitude, sometimes two orders of magnitude. And governance applied after an AI system is deployed, after biased outputs have been circulated, after a regulatory inquiry is started, after public trust has eroded, which is the most costly of things, is going to cost you ten times what governance would have been to build.

into the design process would have cost you. What looks like speed when you skip governance is actually just deferred expense with interest. the organization that ships fast without governance is not actually moving faster. It's just borrowing against a debt that will come due at the worst possible moment. It's a lot like the balance that a good CTO has to play with technical debt when you're writing code we can get it done fast and dirty.

But eventually we've got to go back and pay down that technical debt. We've got to clean up the code. We have to make it, you know, robust with you know with respect to the security and with respect to scalability and things like that. you know, I think one of my favorite examples of this was actually in the federal government, Palantir built a system called Maven Smart System. It was designed for the military and very quickly they were able to build a prototype.

They were able to show how it was really cool and did amazing things and everybody got real excited. And then they spent the next three years trying to scale this thing because they didn't build it the right way to start. And they had to sort of re-architect it on the fly with sort of angry customers whipping them along the way. So like bringing this back to sort of that skeptical ⁓ executive, ⁓ governance done well does not really slow decisions. ⁓ it should just clarify who makes them and on what basis, right?

The organizations that I've worked with that treat governance as overhead are also the organizations with the most decision paralysis. Because nobody's been given clear authority and every contested judgment becomes sort of a political negotiation. So governance isn't the thing that slows you down, ambiguity is the thing that's gonna slow you down.

Ben Jones: Mm-hmm.

we can't talk about leadership and AI and implementation without the question that seems to always come up, which is head count. And so, you know, we see lots of headlines these days of CEOs bragging about laying off X 30, 40% of their workforce, et cetera. So I want to talk about that a little bit because that's a real elephant in the room, isn't it? You know, this idea that if you're in leadership and you're trying to help your organization adapt, like your book title suggests.

Question's going to come up. Do we have the right people? Do we have too many? Do we have too few? How do we get that right? So displacement of jobs. Talk to me about that, Patrick. What's your idea? and what's your advice for executives facing maybe pressure from boards to reduce headcount? I hate that word, but what does the book say about that very dilemma that I think a lot of executives are wrestling with right now?

Patrick McGarry: Yeah, I mean I didn't go into a lot of this explicitly. I tried to keep it more on the constructive side, but you know, I'll tell you there's there are sort of two places where that higher value work story tends to break down ⁓ that I don't want to sort of talk around. First is sort of the transition cost, right? So even if it's genuinely true that AI creates new categories of work.

⁓ you know, the people whose current jobs are displaced are rarely the people positioned to take those new jobs, right? The nurse, to use the ⁓ example that we were talking about before the call, which is, you know, with 15 years of clinical judgment, like can grow into a role that incorporates AI. The data entry clerk who processes insurance forms may not have a clear pathway to the work that AI creates, right? So it's

It's more thinking about it in terms of augmentation rather than replacement. And I I think that is particularly important and and empirically defensible at a sort of sector level over the long-term horizon, right? I know it's cold comfort to a specific person in a specific zip code in a specific decade, but I I also think that a lot of this is performance art, right? On the terms of the CEOs where they

Have done these layoffs and they haven't done them with the nuance that I think is required. Right. And and so when you look at, you know, sort of the math, what I would do if I were a CEO, and I think this it's funny that this hasn't happened more, you have these people that you have built the organization around that were the right people, ⁓ or you wouldn't have hired them, to that have done the job and provided value. Now suddenly you can make all of these individuals

Some amount more effective, 25%, 50%, 75% more, more efficient, effective. and rather than take this as a overall net lift to your business's ability to execute, you want to fire a bunch of people and maintain the same ability to execute? That just doesn't make sense to me over the long term, right?

The other sort of side of this coin though is that the math in some of these contexts actually doesn't work, right? So there are entire categories of work. When you look at you know, some of the examples like back office or document process processing or types of customer service, where when you look at that cost benefit calculus ⁓ for an organization

it's a little bit more straightforward. You know, fewer people, same or better output, because that part of the business doesn't have to scale in the same exponential way that you might want your profit centers to scale. So a leader who pretends otherwise and ignores the question entirely is also not being kind. They're being evasive. ⁓ so I would say, look, if the math doesn't support keeping headcount, the ethical obligation is to be honest about it.

Earlier rather than later, but invest meaningfully in transition support and engage with sort of the policy conversation around what institutions need to exist to absorb the displacement ⁓ and then kind of plan accordingly with the idea that you should still be trying to grow your business, not just maintain the status quo with fewer overhead.

Ben Jones: Yeah, I remember the same thought in Lean when we were implementing Lean, the Toyota production system. It was the same idea, right? Is that, you know, you're going to be reducing cycle times and lead times and processes and the amount of work people need to do is going to go down. So then what do you do, right?

with that

workforce that you have. And I like your answer. I mean, you know

The question I think for the executive first and foremost is how are we going to do more? How are we going to get even more out of ⁓ all of the inputs, technological, human, and otherwise, that we put into our overall company in order to produce value. And so if the focus is on value and producing more value, then most times there is another place for someone to land to do some great work in another area of the business. And to your point, maybe it isn't in the same area because some technology might

reduce the amount of people needed to perform certain functions. But there's lots out there. And there's and especially if your mindset is one of growth. And, ⁓ you know, if possible, make that, make that happen and help people make those adjustments. and if it isn't possible, then, you know, again, do the responsible thing and figure out how to help them land outside of the company in a place where they can continue to grow and thrive. And that's, I think, a way to invest in.

goodwill and sort of an overall strategy that's going to pay dividends might be hard to count the hard benefit of such an approach to leadership, but I think in the end it's the right one. And we're not seeing a lot of that. Like you say, we're seeing that performance art CEOs trying to brag about how much they've been able to save ⁓ for one reason or another, whether that's entirely honest and accurate or not is another question. So such an interesting time to be a leader, you know, in tech especially

so I think your book comes at a very kind of opportune time. So let's wrap up with this, Patrick. If you could kind of give a couple answers. One being sort of if a leader is out there, let's say a C D O or a C A I O, someone who's, you know, responsible for helping their organization adopt and implement new technologies and the change associated with all of that. Give them something concrete to try today, to start today and to

think about implementing and doing differently. And then help them envision or paint a picture of what it could look like a few years down the road if they do all of this well.

Patrick McGarry: So I I think the single sort of highest leverage move right now is to run an honest assessment of where your organization actually is versus where it says it is. And and I'm I'm going to hurry to say this is not a maturity survey, right? This is not a paper process that you might get, you know, one of the big five consultant companies to come do for you, right? This is

A real conversation with the people that are doing the work, asking where governance is theater, where talent strategy is disconnected from the technology roadmap, where AI strategy means a vendor relationship rather than an institutional capability. So the leaders who are sort of furthest behind, I think, are almost universally the ones who think they are further ahead than they are. The assessment

like this is not glamorous, but it is sort of a prerequisite for everything else. Fast forward five years out, the leaders who take this seriously, I think will have built organizations that ⁓ you know, where AI capability compounds rather than plateaus. I think the distinguishing feature is not going to be which tools they chose. it's gonna be

you know, that they made the investment in human architecture early enough that their people know how to adopt new tools when they arrive, which at this point you have to assume will be constantly. you know, the goal is not to have deployed a specific AI, because it's gonna change by tomorrow afternoon. ⁓ the goal is to have built an organization that learns that what choose to lead actually means in this era, right? You're choosing repeatedly

to do the harder work of building institutional capability ⁓ instead of trying to buy the shortcut.

Ben Jones: It's so it's so true. I mean, it isn't about adapting once. It isn't like a change transformation that they're helping to navigate as a one time movement or adjustment. It's like your title suggests, that the organization is capable of continuously adapting. and you know, people are afraid right now. People workers are afraid. I think leaders are pretty intimidated. So I think a lot of it is emotional as well, and you know, and that comes down to culture, that comes down to rewards and recognition.

That comes down to support and the kind of leader that's you know is able to help a group of people face the unknown and embrace it. Like I talk about in the preface to my book, AI Literacy Fundamentals, there's a huge wave and the leader is the one saying, Let's swim toward it. And, you know, and recognizing that the next wave is coming, they're just going to keep coming. Like you said, it isn't a one time.

adjustment that needs to be made. So I think that that's a great mindset. I think mindset is a big part of it. And also, you know, these processes and really the playbooks I think you refer to in the book itself, super practical ways to roll up their sleeves and start doing things a little differently. I think that's a great resource that you're you're putting out there. So all right. So thanks a lot Patrick. Really appreciate it. Folks look for the book. It's going to be coming out on June 8th.

Again, you can already pre-order it in paperback, hardback, and ebook. And we're gonna get out there pretty soon. ⁓ stay tuned. There's going to be ⁓ maybe some events happening here in the DC area to to kind of celebrate the launch of the book. And Patrick, thanks again for joining on the show. Thanks for the book. I wish you all the best. It's been great chatting with you and yeah, I hope hope that this first time author experience for you was rewarding.

And looking forward again to seeing the fruits of of all your efforts as the book hits the shelf.

Patrick McGarry: Yeah, definitely. I mean thanks. I appreciate you having me on, but also taking me on this, you know, journey. it's been one of a lot of learning. So I've appreciated every step.

Ben Jones: Awesome. Yep. It's a lot to learn and and a lot and you know, teaching, I think, you know, helping people learn as well is a really rewarding thing. So you're gonna love it. So thanks again, Patrick. Okay, everyone, stay tuned for the next show coming up soon, and we'll talk again real soon. Bye now.