S7 #7 Golden records and real governance: Enterprise AI at scale

[00:00:00] INTRO: Welcome to The Power of Digital Policy, a show that helps digital marketers, online communications directors, and others throughout the organization balance out risks and opportunities created by using digital channels. Here's your host, Kristina Podnar.

[00:00:18] KRISTINA: organizations often talk about enterprise AI as though the principal challenge is selecting the right model or finding enough use cases. But the technology is usually the easier part. The real work really involves connecting data, clarifying authority, managing risk, getting different parts of the organization to cooperate as one enterprise. Our guest today has been doing exactly that work. Thomas Hill is senior director of digital and commercial AI enablement at Danaher. He is a global science and technology leader with more than 15 operating companies across life sciences and diagnostics. He's built Danaher's enterprise commercial AI program from the ground up, no small feat. Assembled a cross-functional team of data and AI specialists, helped unify eight million contacts and three million customers into a governed golden record, and he also co-designed the company's agentic AI framework and data privacy architecture.

That is a lot. His work spans the full enterprise AI stack, including GDPR compliance, cross-company data sharing agreements, governance, risk, commercial enablement, and customer experience. Before moving into his enterprise role, Thomas led digital customer experience and e-commerce at Pal Corporation, where his team drove more than one billion dollars in digitally influenced revenue over three years.

Thomas, that is a lot. You've been very busy. Welcome.

[00:01:41] Thomas: Thank you. Yes, when you say it like that, there's a quite a road there, and it was a lot of fun, a lot of learnings along the way. Thank you for having me on today.

[00:01:49] KRISTINA: And more to be done, I bet. So you built Danaher's commercial AI program from the ground up. That seems like such a huge undertaking. When people hear that, they might picture a technology roadmap, collection of use cases, perhaps even a new AI team. But when we say you built the program, what does that actually involve?

[00:02:10] Thomas: Exactly what you said, actually. We had to start from scratch completely, so that meant defining the roadmap and the vision and the outcomes. So starting with business partners and the different operating companies and the business commercial sales and marketing is where we started. Finding out exactly what they wanted to do, and our goal was not typical for AI. Our goal was to actually drive revenue and growth. Most of the time you see in an AI program, there's a big goal on cost out and in efficiency and productivity.

But we were trying to drive incremental growth for the business. So we had to build use cases and kind of work backwards from there. So the roadmap was business outcome oriented, and then we put together the plan around what kind of data pipelines, what kind of data sources, and what kind of architecture we would need, and then also what kind of team we would need to build and support that. So it was the team, the roadmap, the architecture, everything. And it took at least six to 12 months to really get all of that organized and shaped.

[00:03:14] KRISTINA: If you say it really fast, it sounds really easy. I know it hasn't been. What did you learn early on that changed your original planning, and how did that process kind of change from day one to where you could have really got to it eventually?

[00:03:28] Thomas: I think there's a few things. You have to do a lot of education and alignment of stakeholders and leadership to understand what they're getting themselves into. There's a bit of the counting the cost with this, where you need to know that a lot of this work going into it is not gonna be glamorous. You're not gonna get the you know, AI just sprinkled on top of everything like magic dust that just makes everything beautiful. So there's a lot of that politically complicated work, getting data ready, getting processes redefined, figuring out what the human teams are gonna do. But you've gotta do that groundwork and build those foundations. And I think that the teams that are closer to the work understand that, but leadership and executive patience isn't always there with you from the get-go, and also your stakeholders that want product don't always understand the work that goes into baking that end product. So you have to be prepared for that. And also , there's a phrase that I, put together thinking through this where AI is sort of like an organizational MRI. It's going to uncover all of your issues, your skeletons in the closet, your situations where your tribal knowledge is running the process.

And it's gonna illuminate a lot of that dysfunctional stuff, and you have to be prepared for that, but also be prepared to deal with it as part of your roadmap. So if there's technical debt, if there's data infrastructure problems, that has to be part of your investment plan, part of your roadmap, and you need to educate leadership as to what to expect. And there's a lot of roadblocks that you run into that you need to be prepared to tackle.

[00:05:07] KRISTINA: You've been dealing with 15 operating companies, and each one obviously has its own customers, its systems, commercial priorities, institutional knowledge, and certainly culture. How do you create an enterprise capability without flattening the differences that make each opco effective? Because that seems like a challenge.

[00:05:26] Thomas: Definitely a challenge. The good thing is, is there's a lot of commonalities across our businesses, so there's a lot of the same customers, a lot of the same outcomes, a lot of the same even sales processes, so that, that was good. But of course, everybody's on a different tech stack, and they've got their own products and so you have to think about that. What we've tried to do is deploy things centrally as much as possible that are scalable but then allow for customization and sort of bespoke implementation. So if there's a use case that a business really wants to make it their own, make it special, then we can adapt it from there. But our goal is to try to make something enterprise-grade, and you wanna build once and deploy broadly if possible.

So there's there's always challenges. You don't wanna get too customized, and a lot of times if there is something super custom, we just let that business do it. It's not, not... And when you think about how to build an AI team, there's this idea of centralized or decentralized, and when you've got a huge enterprise organization where you're trying to potentially do something central some things are just not designed to be central. You let the smaller local teams build and a lot of times they have their own resources to do that

[00:06:38] KRISTINA: And one of the things that's hard to figure out sometimes is shared accountability, which tends to be centralized, but sometimes it can be decentralized. And it's an issue, obviously, when no single team controls the entire environment. How have you dealt with that? Because you really are working at scale.

[00:06:55] Thomas: There's a lot of different ways we could go with that. I don't think we've figured it all out yet. I think there's a accountability around data and data governance that , we could kinda dive into that for a minute. That's something that we- we've had to sort of organically build. So th- this idea of data governance and privacy and frameworks didn't really exist in the operating companies at Danaher before we started. So we've had to build frameworks, and we've had to build a plan around how we're going to build trust around this data and govern it and make it ready for AI. And as the enterprise organization has evolved, they've brought in leaders that are specializing in this and doing this. So it's, bit's becoming more mature as we go. We're bringing in legal teams and putting together sharing frameworks and all of this. And so it's sort of a, you know, process. But we are trying to just lead from the front by building trust around what we're doing, and being accountable to where we're strong and where we're weak and where we need help. And so there's... I guess there's accountability on that side. When you get into a kind of accountability on business outcomes, it becomes really important to have strong partners in the operating companies or the businesses.

So you wanna make sure that when you're designing an AI team or a data team, that you've got some people in sort of a forward deployment role that are close to the business, and that the business has shared goals and shared outcomes with the AI team. So for our example with commercial, we're trying to drive revenue. In our example, it's how much of the growth plan for the year is going to be attributed back or dependent upon these AI use cases. And so do they have it in their plan, and are they expecting these AI use cases to deliver? In which case, are they gonna provide the resources and the adoption and the support that we need to be successful?

'Cause we can't just build things and hand them off. It has to be a shared initiative. So there's accountability from the business to make sure that they're being a good partner, and there's an accountability on our side to be transparent with how strong our data is, where our gaps are, and what we're gonna do around governing that. So there's other directions we could go with that accountability question, but it's not an easy task

[00:09:18] KRISTINA: One of the things that I think accountability is hard to define around really is this thing of a single customer identity. It's rarely simple, and one of the foundations of the program that you created is this notion of a golden record covering approximately eight million- contacts and three million customers. There's obviously the technology behind it, but when we think about the business problems you were trying to solve, what were the drivers? 'Cause that's a big task.

[00:09:44] Thomas: Again, so everything we're doing with sales and marketing is all around the customer. So pretty quickly we realized we're gonna have to get a pretty good handle on our customer data. And in our example at Danaher, we've got at least 15 operating companies. Each operating company has their own tech stack where all of this data lives. And when we think about customer data, it's a combination of the account and the customer itself, and also the contacts that are within that. And when you get into the world of contact data and contact sharing, then you get into this whole world of GDPR and privacy, and are we allowed to share?

So there's a whole legal and privacy framework you have to build into that. But the premise behind what Danaher was trying to do is move towards more of a cross-selling and enterprise selling approach and have a better experience for our customers, but also reach more of our customers with more of our brands. So we wanted to open up this Golden Record so that all of our sales and marketing teams would have access to the entire pie. In- instead of each, today, each operating company only sees a small piece of that pizza pie. So this idea of building this one cohesive Golden Record, it's also a really fantastic product in and of itself because one operating company might only know a little bit about the customer, but then all the other operating companies know other bits about the customer. So you bring all that together, you master it, you create this data product. By itself that's really powerful, and that's gold. And that creates something that's meaningful even if we all gave up on AI in the future and said, "Look, , we're done with AI, it's not working." The fact that we've built such a powerful data product for context is important.

But that's proving to be really critical as the kind of the new trend this year I'm seeing in different conferences is this idea of context intelligence and semantic layers and having your data ready and available so that you have good context for all of your agentic use cases or whatever you wanna do down the road. But this for us, this idea of a customer Golden Record has been probably our biggest success story and something that the businesses are the most excited about. And now we're trying to think about, how do we turn that into something that's profitable for the business?

[00:11:51] KRISTINA: I was gonna ask you about that. You mentioned semantic layers. What role have taxonomy and metadata and common definitions played in this whole story? Because that seems like a lot of work to get done in preparation for the Golden Record.

[00:12:05] Thomas: Yeah, that's something that we're still early on. I wouldn't say that we're super mature. I think the idea of building out a data dictionary, a data catalog, metadata tagging, that's critical. The problem we have right now is we've done all this work, but our downstream users don't really know still what's available. So we've gotta do some work around educating and building that library, building out that sort of data product marketplace. I think there's a combination of process with this. You gotta build some new processes, but also there's technology. There's a lot of technology out there that does this and makes it easy, and it's, you can bolt it on to your data platform, or you can bolt it on to some of your other tools. So then it becomes a question of do we wanna, you know, buy a, a new technology just to help us with this metadata cataloging?

But I think that's critical. That's part of this idea of a, sort of a governance layer around your data. Need to provide the right context. You need to provide the right metadata so that agents can sort of move into this autonomous world. But I would say we're still very early and aspirational about some of those things. That's that's a difficult thing to tackle. Oftentimes you have to get a lot of people in the room, redefine some processes around common definitions. And for a large enterprise organization, sometimes this is the first time people have ever had that conversation if they've not spent a lot of time with data governance. So you have to educate and bring everybody along at the same pace.

[00:13:27] KRISTINA: I think that's such an important point, the education and bringing people along, especially when you talk about confidence in the data and confidence in a record. How do you measure that confidence from folks? How do you actually understand what type of education to roll out to them? How are you defining sort of the onboarding process for them, because everybody, I'm assuming, is at different levels?

[00:13:51] Thomas: When you think about measuring confidence, there's just a very simple, do we have the data? Do we not have the data? Do we have agreement and frameworks around sharing the data? That's part of it. The tricky part is when you get ready to start deploying into a business, how many nos do you get versus how many yeses , do you get? And if you are getting a whole lot of questions repeatedly, then you haven't done enough job upfront to kind of build that trust.

So at, a lot of what we've done at Danaher is early on we kind of created this, I call it, like, a gamification approach. And this a lot of what we did was- was kind of unlocking our data around privacy for contact sharing. This is one of our early wins. But you wanna get operating companies comfortable.

And this could look a lot of different ways for a lot of different enterprises. But for us, these businesses had never shared data before. So the first thing to do was get them on board with the idea of sharing data and the what's in it for them, how this is gonna be profitable to them, how them sharing their data with the rest of the business would be beneficial. But they're also, they're gonna get data back. They're gonna get the full picture back. We created kind of this nine-step plan of all these different things to build that trust. So everything from data-sharing agreements to having PIAs in place with privacy, to having infrastructure and IT architecture reviews done, sort of like this checklist. And it was gamification approach to saying, "Okay, how can I unlock new use cases, and how can I unlock new even countries?" 'Cause if you think about contacts and GDPR, how you handle contacts in Europe is different than how you handle them in the US. So the ability to scale to different use cases in different countries, you have to kind of go through this plan. So building confidence and building the the buy-in was kind of this combination of, have we progressed through our plan? Have we checked the boxes? Have we unlocked different things? And have the stakeholders actually signed off on those data-sharing agreements, and are they agreeing to share data? And then once you get past that, then you're pretty good.

[00:15:56] KRISTINA: How do you make these requirements understandable to commercial teams rather than leaving them entirely with legal and privacy specialists? Because you mentioned the important part of stakeholder sign-off, which is a pretty significant task.

[00:16:10] Thomas: We brought both to the table. So usually when we're working with the legal and privacy, they don't necessarily know everything we're doing, so we've had to very clearly define the use cases and the business processes that we're impacting, what kind of data is being shared and processed and why. And one example of where we really had to have, like, commercial sales business partners at the table along with legal is this idea of data sharing. In our use cases we wanna share data across our businesses, but some of our contracts don't allow that with our customers. So our commercial sales team said, "Look, you cannot share this, this, this, or this type of data because that's confidential information that our customers have said in the contract is not allowed to be shared."

And legal, you know, they didn't know what could and couldn't be done. So you you have to bring the different stakeholders together because then our job as a technology team that's building is to make sure that we enforce and build the systems in place to do that. We need to be able to build the governance logic. We need to be able to build the snowflake and semantic layer around that with the right controls around data. So just because you bring the data in doesn't mean you can always share it out and use it. But you need to have the right people around the table to define that. We're still, again, that's another thing that we're kinda a little bit early on which is how do we build that, that standard working governance around some of those things? But it, these are the things you run across that you have to solve for.

[00:17:34] KRISTINA: I think a lot of folks who are listening to you are probably going to be fascinated by what you've accomplished because one of the most important distinctions in digital policy is the difference between what the system can do and what the organization has decided it may do. Thinking about it, how do you keep that distinction visible as the technology becomes more capable? Because certainly we can do a lot with technology, but it doesn't mean that we should. How are you kind of balancing those two things out?

[00:18:04] Thomas: Gosh, that's a good question. There's no shortage of ideas with AI and things that are possible. The first thing is what's the organization ready for? And that there will be two constraints on that. There's the maturity of the organization to adopt certain use cases, and there's just, and so depending on what you're... our world of sales and marketing, there's just very basic level use cases that you have to get off the ground. So you can't do all the advanced, you know, everything under the sun use cases. So organizational readiness and maturity and willingness to change is one of your things that might hold you back. The data readiness is another thing that might hold you back. So you can't do every idea under the sun there as well. I think the what should you do versus what could you do question I think it comes down to what's right for the business and what's gonna move the business forward and what's right for your customers. So it's a unique decision I think each company should go down. I wanna answer a different way or take it a different direction a little for a minute. One, one thing that's important with thinking about policies and frameworks is you, and this is something, again, we're thinking about and actively trying to design into our model, but policy can't just be on paper. So policy on paper is not really governance in practice. There's another analogy that , I've heard recently that I like, which is in this world of agentic governance is like in a soccer match, a football match, where it's like an active referee that's running down, up and down the field alongside the players, blowing the whistle in real time when things go wrong. So it has to be dynamic. It has to be real time. So the business in the world of agentic, it's all gonna move so fast, and you're gonna have these swarms of agents, multi-agent orchestration, making decisions, and that's one of the big leadership educational things you have to do to change too. A lot of leaders still think that AI is about chatbots, and that an agent is a chatbot, and chatbots are very different than agents. Chatbots will just answer a question, but agents are gonna start doing things autonomously and making decisions, and then the human has to orchestrate that. But it has to all follow the right governance and boundaries and rules and things that the human defines. So the policy has to be embedded into the orchestration layer. And so we don't have time to, to schedule a steering committee meeting every other day when there's a new use case or a new decision that has to be made. It has to all be embedded into the process so that the agents can reference it. So there has to be this idea of a risk governance policy layer- built into it so that the policy gets off the paper and gets into the system. I think that's a shift that every organization's got to be made, otherwise you're not gonna be able to go fast, you're not gonna be able to make your agents fully autonomous. And there are ways to do that, but I think people have to think through that.

[00:20:58] KRISTINA: Yeah, I think you're you're rallying the troops to get away from shelfware. As I like to say- Yes ... the s- the static PDFs on SharePoint that nobody's looked at in years and probably won't for a really long time, hopefully. But that's a really hard thing, right? Because you're having to define a policy upfront and then trust that you've done the right definition, which isn't necessarily that a lot of legal and audit folks lean into naturally. That's just not their nature. So how are you working through that adjustment and getting everybody to say, "Look, obviously we can think through things, but we can't account for every use case"? So how do you ensure that, you do what is commercially meaningful in terms of use case definition, define the risk framework, and then kind of get folks just comfortable with being uncomfortable?

[00:21:48] Thomas: You, you leave it open-ended in some cases. There always will be this role of a steering committee a forum, a council, what have you, and you have to meet on a, a regular cadence. So your policies that you write on paper need to always be refreshed. But like I was saying, they need to be embedded. So whatever that, that governance and rules and structure is, it is it needs to be embedded into the control pane of AI and the agentic. So that role of the human in this governance committee, it, it starts to, to shift. And the human in the loop is providing the culture and the setting the values and the judgment, and you almost have to define and teach, just like you would teach a human your ethics and your policy you have to teach the agents what kind of company do we wanna be? What kind of-- How do we wanna treat and serve the customer? So if you have agents that are starting to do things like making a decision, for example, on credit approval or all sorts of different things or how to answer a customer inquiry that might come on the website based on how you want that experience to go, define that, but then be flexible to then go in and change it over time and know when to throw it back to the human. So there needs to be the right safeguards in this whole process so that when, you know, when does the agent stop and send it back to the human? I don't think any organization has figured this out.

[00:23:07] KRISTINA: Do you feel like you've gotten at least to the other side of pilot deployments? Because I know that's a struggle for a lot of folks. Actually getting co-pilots designed, getting a pilot launched seems to be reasonably straightforward for folks, but it's a completely different ballgame in terms of let's actually launch an agent, and let's let it go to town. And interpreting a lot of the boundaries that you mentioned, and sometimes maybe even understanding or not understanding when escalations are needed. There's just a lot there. So how do you overcome that barrier? Have you had a lot of experience yet with overcoming that, that big wall?

[00:23:46] Thomas: We're still in the middle of it. I think , in our case, the scale is, is really difficult. You can scale across multiple vectors. For us, you've got operating companies that we can scale use cases to, and even with, and then regions. So a lot of times we go US first, and then we'll go to Europe after that, and we'll kind of go region by region. But when you start with an operating company, we don't go to, let's say there's 100 salespeople in that organization that we're gonna impact on process. We don't start with all 100. We only start with, you know, the 5 or 10 early adopters. But it is very difficult to break past that first pool, and there's a lot of reasons for that. Sometimes it's you've got conflicting priorities. Sometimes you've got leadership changes, and you have to have that top-down leadership to continue to drive the momentum. This, so this gets back to that whole part that, you know, the hardest part of AI is not the AI. Like, we can build a working tool and get it and know that it's working, but you need to have people willing to engage early on and give you that feedback of what you need to fix. And then you need to do the harder part, which is redefining those sales processes or the workflow. Because if you're going to go in and say, "Look, this is the new way to do work, and here's what the AI is gonna do, and here's what the human is gonna do," you sort of have to go to the drawing board and completely redesign all of your flows.

Every organization does it a little bit differently. In Danaher, we have the Danaher Business System, which is kind of our approach to driving change and driving business process. Every organization's probably a little bit different. But I think you, you almost have to go in as a prerequisite and think about work redesign, and that's gonna be the new skill set of the future, is how do you design a future where, you know, you've got coworkers who are AI agents? And you may be managing a team of agents, and the work that you used to do as a a knowledge worker is completely gonna change. You have to do that work before you can scale to, like, 100% adoption on some of this stuff. Otherwise, all your stuff just might be really basic, off-the-shelf GPT chat type things, which are good, but , they're not gonna give you that big step change.

[00:25:54] KRISTINA: And hopefully we've gone a little bit past the chat mode or it feels a little bit- Yeah ... we should have. You mentioned people, and you certainly assembled a strong cross-functional team of data and AI specialists to build the program at Danaher. What capabilities did you know had to be represented from the beginning, and what are some of the harder roles that you found to fill?

[00:26:18] Thomas: So we brought in a mix. We knew we needed people that understood the business, so we had a few, we called them forward deployment or product owners people that had sales and marketing experience, that had even business experience in our market, so life sciences, that type of thing, but that had a passion for AI because we knew whatever we built, we knew we would need to translate it to the end user. So we brought in a few people that would be sort of the product owners and the use case deployment people. But then obviously we have to build a lot around data, so we brought in data architect, data engineering teams just to basically one of each kind of thing to get started, one of each type of person to, to build out our data and then the typical AI engineer, ML engineer, data scientist, those types of things.

We worked backwards. We, started with the menu of use cases that we knew the business was asking for, and then we worked backwards to what kind of team would be required to pull that off. We also brought in some data governance. So we had a couple data governance, one, one person more on the tactical MDM side, which is how do we make sure that we're thinking about this idea of building the golden record and organizing our data in a unified way? But then also we, we hired somebody that was a, a lawyer with legal background that was going to have the role of clearing those roadblocks and clearing the roadblocks and then also addressing data privacy, legal risk issues. And in some cases you have to phone a friend and call external partners and consultants, and that's when you and I first met, Kristina, a long time back, which is how do we get advice from people that have done this in other places in the industry? And sometimes having somebody from outside the company give a voice helps clear the way and clear some of the noise internally as well, that sometimes when people don't listen to their own, they'll listen to somebody from the outside. So that's a strategy to follow as well.

[00:28:08] KRISTINA: We're trying something new today, Thomas. We're it's called Fill in the Blank Lightning Round. I'll begin each sentence and have you complete it with the first answer that comes to mind. Are you up for playing?

[00:28:20] Thomas: Let's try it. Go for

[00:28:21] KRISTINA: it. All right. Enterprise AI fails when blank.

[00:28:28] Thomas: Leaders don't count the cost

[00:28:34] KRISTINA: A golden record is only valuable if

[00:28:40] Thomas: You turn it into a semantic context layer

[00:28:49] KRISTINA: An AI agent should never be allowed to do blank without blank

[00:28:55] Thomas: Act without guardrails

[00:29:00] KRISTINA: And the most underrated human capability in an AI-enabled enterprise is

[00:29:07] Thomas: The human touch. I don't know. That's, that's a good one. Just the being human maintaining relationship and understanding how people think and work

[00:29:18] KRISTINA: Out of curiosity, why do you think that will become the most important capability rather than less?

[00:29:25] Thomas: In a world with AI, people are gravitating towards things that are more human and more authentic. We're hearing and seeing that left and right. Everything is becoming so digital and AI first and people can spot it, and they feel like it can be fake. The quality is amazing, but sometimes, you know, there's something to be said about being inhuman, being human and being a little imperfect. The funny thing with AI now, people can spot writing with all the em dashes, and before AI, I used to write and use em dashes. So now I have to I have to think about, okay, well, I don't wanna be accused of writing with AI, so now I have to write differently. But I have to write in a more human way, and I need to be able to pick up on the softer cues and manage relationships. You wanna set the ethics and the culture around your organization and make sure the human always stays on top of everything and is valued above the AI,

[00:30:21] KRISTINA: Thomas, I feel very seen. My husband called me out the other day and said, "This is all AI written because you're using em dashes." So I had to actually go to CMS Wire and show him a bunch of stuff I wrote in 2019 just to prove that em dashes have been in my vocabulary for a while. They've been

[00:30:36] Thomas: around. Yeah,

[00:30:37] KRISTINA: yeah. Yeah, it's good

[00:30:38] Thomas: grammar.

[00:30:39] KRISTINA: There you go. As you look at the next several years, what do you think enterprise leaders are still underestimating about AI?

[00:30:49] Thomas: I think we're learning a lot this year. You're seeing a lot of large organizations fail pretty fast and setting some examples for the rest of us. When I said counting the cost earlier, I think what I meant by that is you, you've got this, this big corporate peer pressure to realize savings really fast and go fast being AI first, and the elephant in the room is usually that's associated with some kind of cost reduction or headcount reduction or, or people workforce reduction. And then you see examples like with Uber and Microsoft where they go in and they have burnt through their token budgets and for the full year within the first three, four months of the year and realizing this is actually expensive on the other side of the fence too. You've, you can't just replace a human with an AI agent and expect it to be free. So I think this this learning that everybody's gonna go through is that you've gotta be smart about, Where you redeploy your human capital and your knowledge workers and how you use the human plus AI to accelerate and grow your business. There's a really good example I saw recently of IKEA, the furniture store, who they just took 8,000 people and replaced them with AI agents. But they did not lay off the 8,000 people. They repurposed them and retooled them to higher value customer service, knowledge working type things to help grow the business with supporting the customer. So I think that's gonna have to be sort of a, a shift in corporate expectations around what you're gonna get out of AI. It's gonna be a shift towards kind of a balancing of all this hype that's been going around, the hype and the buzz around AI, and a lot of the fear that's been in the market around it replacing jobs. I think that'll start to die down, and we'll see a little bit of a leveling off once people have a bit of a reality check.

[00:32:39] KRISTINA: That's a really great note to end things on. Thomas, thank you for coming on to have this conversation and for giving us a practical view of really what it takes to build, not just sort of enterprise AI across data privacy governance technology, but also the people and the business itself. I think at the end of the day, enterprise AI does not scale because an organization has found the right model, but rather, when you get to the point where the organization has created the data and the authority and accountability and all of the operating discipline that you've talked about, which really allows us to trust the model and to move forward. So appreciate you having this conversation.

[00:33:16] Thomas: This was really fun. Thank you so much.

[00:33:18] OUTRO: Thank you for joining the Power of Digital Policy; get access to policy checklists, detailed information on policies, and other helpful resources, head over to the power of digital policy.com. If you get a moment, please leave a review on iTunes to help your digital colleagues find out about the podcast.

‍