How Do You Prove AI ROI When The Return Is Only 1%?

0:01Most AI coverage tells you what happened. AI Signal tells you why it matters. It's a free, five-minute weekly briefing that helps people leaders make sense of the shifts, trade-offs, and decisions hiding behind the AI headlines. Less keeping up, more knowing what to do next. Subscribe to AI Signal at peoplemanagingpeople.com forward slash AI dash signal.

0:27Only 17 % of organizations have deployed AI agents at all. The multi-department number is around 3%. JLL has stopped counting theirs. On today's show, I'm talking with Carl Empower, Global VP of AI and Platform Technology Enablement at JLL, about what the other 97 % are about to learn. Because JLL isn't experimenting with agents. They're running them in production across facilities management, brand communications, and corporate functions company-wide. and the lessons from that aren't mostly technical. The first thing that broke wasn't the technology. It was the ROI conversation. How do you justify the cost of table stakes?

1:03How do you sit in front of your CFO and say, we might get a 1 % return on this? Power's answer was to put the value conversation back where it belongs, inside the P &L that's making the investment decision. If a business leader is trying to grow share of wallet with top clients by 20%, AI becomes a part of that conversation instead of a hammer looking for a nail. The early career question deserves some focus in this topic, too. Stanford Research draws a hard line. When AI automates, entry-level hiring falls. When it augments, employment holds or grows. JLL made a deliberate choice, and the interns they've sourced recently presented a capstone work so sophisticated that without being told, Power wouldn't have guessed that they weren't there for three or four years into their careers.

1:43So today we're covering what separates agents that survive from the 40 % Gartner projects will be canceled by 2027, why ROI framing matters more than the technology itself, how to build AI fluency across 80 countries without losing local relevance, and what a chief people officer with zero agents in production should do in months one through three as they roll them out. I'm David Rice, and this is People Managing People. And if your organization is still figuring out where to start with agents, this conversation gives you a practical blueprint from someone who's already well past that question.

2:13So let's get into it. Welcome to the People Managing People podcast. As always, I'm your host, David Rice, and today I'm joined by Carlin Power. She is the Global Vice President of AI and Platform Technology Enablement at JLL. A lot of organizations out there are experimenting with AI agents, but you've actually deployed some successfully. So we're going to talk a little bit about that and some of the lessons learned through that. So the first question I wanted to ask you was around sort of, I'd seen like a Gartner CIO survey that was saying, you know, only 17 % of organizations have deployed agents at all.

2:51Sure. And I think the multi-department number is like 3 % or something like that. It's really low. Yeah. So for a lot of people, they are at the beginning of this. And you've released, the last number I saw was like 34. I think you're at more than that now.

3:06David Ri: Yeah, we're growing. I think we've just stopped counting. So my first question is, what do you know now that the other 97 % are about to learn? Sure. Well, I think it actually kind of goes back to some decisions that we made initially. And we had this investment in what we call JLL Falcon, which is a fancy brand name for a collection of AI capabilities and APIs. that all of our software developers could utilize cost-effectively to deploy AI capabilities on technology that they were building brand new and existing technology that they already had in the marketplace. And so because of that and because we were able to decentralize innovation across all of our engineers, they started upskilling a lot faster in how to build and deploy things like agents.

4:04David Ri: So we have agents in our facilities management business that are optimizing work orders and how technicians get deployed to sites to do their facilities management work. We have brand agents that kind of brand our PowerPoints and our tone of voice for communications that are both internal and external JLL. We just deployed, you know, about a month and a half ago, AskJLL, which is a collection of agents that all are domain experts across our corporate functions that help employees get support across all of our corporate functions. So because of that decentralization across all of those engineering teams, we're just building agents for bespoke business problems, you know, against very deliberate ROI outcomes.

4:56Is that part of why you think it's been, you mentioned the bespoke nature of it. Is that part of what's made it successful? Is it you're not sort of generally trying to solve something? You're being very specific.

5:06David Ri: Yeah. I mean, we do have general agents and we do have general purpose AI tools. So we have like a general AI assistant that we call Pulse, affectionately, that looks across all of our chats, all of our emails, all of our shared documents. And so it kind of acts like an administrative assistant that really takes a lot of the burden of the administrative tactical things that you have to do in a knowledge worker job. And so we did intentionally deploy some general capabilities. But yes, the real ROI, because that's what I would call table stakes. Right. So that's stuff that all of our employees need in their hands just for us to be able to compete.

5:46David Ri: Right. Marketplace. all of the agents that are being deployed by our software engineering teams that support a particular business line or a corporate function are doing, are solving bespoke business problems that drive our business outcomes forward. And that's aligned all the way up multiple layers of the organization. So take me back to like agent one, you know, like what broke that wasn't technical? Was it like permissions and ownership or someone's job description essentially? Agent one. Wow. So that's actually Pulse. So also known as like the JLL work tools, because now we deliver that through, you know, an MCP or the fancy word for a connector right into an agent into structured data.

6:35David Ri: So so I actually think that it wasn't really a maybe a technical whole thing that broke, it was more of how do we justify the cost of this against like ROI? And so I was talking to you a little bit about table stakes, right? It's a pretty hard conversation to have when like, you're really only seeing, you know, efficiency numbers and like the one to 3 % like range, right? Where it's like kind of hard to like sit in front of your CFO and say, hey, so I think we should do this thing, but we're, we might get 1 % return on it. So it was initially trying to figure out what is the, how do we articulate the value of table stakes?

7:19David Ri: How do we articulate the value of a particular agent? And is that going to be a standard that we set for building, right? So how do we make sure that every single builder in our organization can say, I'm going to deliver this agent to solve this problem that before cost us X amount of labor or, you know, deliver it X deliverable. but now what it's going to do is we're going to be able to be at 5x throughput or we're going to be able to generate a new line of revenue or we're going to be able to increase our market share or our share of wallet for a client and once you start getting your teams thinking about it in that way the the ROI proving out is a lot simpler to justify but that was the first thing where it's like where they asked us like hey like you know when should people use this how should people use this like what's what's their ROI and we're like good questions well that's interesting you mentioned sort of uh making that you know normal like normalizing that it was there like a standardization of language that went along with that or sort of like understanding like what qualifies as ROI for the people who wanted to do these things like that's because I could imagine people's imaginations might go a little wild you go you got to go like hey that's not actually like a real business value.

8:37David Ri: Yeah. Thankfully, we have some great finance professionals in our organization. And that's actually, though it's a centralized function, right? Like within finance, it's decentralized in terms of how they sit within our business lines and our functions. So our finance team has spent a lot of time thinking about how do we give our teams frameworks for actually coming up with an ROI conversation and, and I think, um, or an ROI equation. And I think, um, that's maturing now. I wouldn't, I wouldn't say that, you know, we're, uh, I have the answer to that because if I did, I would just quit and start a billion dollar, you know, consultancy overnight as what everybody else, but, but I do think that it is, um, putting the conversation about value back in the place where it belongs, which is in the P &L that is making the decision about the investment.

9:34David Ri: And so if a business leader is saying like, we need to increase our share of wallet with our top 10 clients by 20%, AI is now part of that conversation instead of us being a hammer, like looking for a nail. I think it was Gartner projects like 40 % of agentic AI projects will be canceled by the end of 2027. I guess my question to you is like, when you look at the agents that you've built, what makes you confident that they'll be resilient and that they'll be, they'll go on beyond that? Again, it kind of comes back to our, our decision to deploy JLL Falcon, right? And then the decentralized decision-making in our engineering functions that are now tied to business line KPIs.

10:23David Ri: We don't build anything, or at least we're trying not to build anything that doesn't have an attributable ROI or a bankable savings now. And so it makes the conversation really easy. Like when you're looking at the AI agents that are doing well or doing not, are they meeting their targets just like an employee would be meeting their business targets, Right. Did we is this technology, is this agent doing what we expected it to do? Is it realizing the benefits that we expected to realize? And if not, why? And should we continue to invest? Right. So a stop, start or continue conversation needs to be happening periodically over any of the agents that you're deploying in your your organization.

11:11You mentioned the decentralized decision making. Was that something that you all built in from the start? Was that part of your processes before you started all this?

11:18David Ri: The foundation of our AI strategy is our enterprise data layer, right? So we've spent the last five to six years under our CTO, Yao Morin's direction, in building data fluency and accuracy and health and relevance in all of our data, basically in our enterprise data warehouse, right? And so every single piece of data that we have in JLL goes into our enterprise data warehouse. So then there's a set of governance capabilities that define what that data is, what it's relevant for, who it's relevant for, and how people get access to that information. So we have a really robust data layer that sits at the bottom.

12:09David Ri: And then we have JLL Falcon that sits on top of that data layer. So, you know, when we get into the MCP or the connector conversation, right, that's just going straight into our enterprise data warehouse and pulling data out like that's relevant to the user consuming it. So that's our, you know, that's our scaling function, right, plus a cost effectiveness and just from a data relevance perspective. And then we have a citizen-driven AI strategy and we have a centrally driven AI strategy. And our centrally driven AI strategy is where the decentralization that I've been talking about takes place.

12:45David Ri: So there are initiatives for each business line and corporate function technology team against that strategy to deliver the differentiation that we need in the marketplace. And the citizen-driven strategy is how we're enabling people internally as an enterprise to be able to use our tools and to develop an internal AI fluency. That's like our enabling function, right? So you've got the foundation, which is our data. You have the scaling function, which is our JLL Falcon platform. we have our differentiation function, which is that centrally driven. And then we have the enablement function, which is our citizen driven.

13:32David Ri: And then that all rolls up to our free business outcomes across like delivering, you know, client value and unlocking on excellence in our operations and innovation. So I think that is core to why we've been successful. But to your question, which was, was that decentralization decision deliberate? Yes, it really was. Did we make it right away? I can't remember. But is that where we ended up? Absolutely. One of the questions I want to ask you is I've seen some Stanford research. It kind of drew like a hard line that, you know, when AI automates tasks, entry level hiring falls. And when it augments, employment sort of holds or grows.

14:18So across 34 agents or even more now, which side of that line do you think you've landed on? And sort of, was it a decision or an outcome?

14:29David Ri: Sure. So I think that if a company is focused on efficiency and productivity and thus and outcome is to reduce, you know, labor, early career labor. I think that's more of just an immaturity in their overall AI strategy and engagement model. And so I don't, I don't know, it was never our deliberate for us to, to go in and say, okay, we're going to have all of this cost out in the organization and we're going to go eliminate 40 % of jobs and, and all of that. I think that was a lot of, you know, at the keynote this morning, they were calling it fear mongering. Right. Like, I just feel like that's probably too extreme of a position.

15:17David Ri: And again, I think it's short lived. I think it actually can be really detrimental to an organization's success to completely cut off your early career talent. Yeah. And I actually just this last Monday was sitting it with our sourcing and procurement interns. And they were presenting their capstone projects, a lot of which had used our internal tools and some of the advanced labs that we have partnerships with. And if the head of sourcing and procurement wouldn't have told me that they were interns, I would have told you that they were three to four years into their professional careers because of the depth of knowledge they had about very complex sourcing and procurement problems.

16:07David Ri: problems, how close they were with the client accounts that they had to work with to improve like our, you know, how much assets we have under management, how much of the contracts we have under management. And I was, I mean, I learned things. I was, I was blown away by both what they were able to do in such a short period of time. And I was even more blown away by the fact that at the very end, our sourcing procurement chief basically said, you all have jobs here at JLL if you want them. And I think that is a really good example of we shouldn't be cutting these early career jobs. We should be setting completely different expectations about what we should be getting out of early career talent.

17:02David Ri: And so we've completely re-architected our early career strategy around kind of these project-based capstones that really create value out of these internship programs that we hold inside of JLL. And so, again, I just think that it's probably an immaturity thing and that I think that they will eventually get there. And I would caution firms against trying to eliminate that early career ladder, because I just think that that's actually where a lot of your innovation is going to come from in the next three to five years. Well, I love that you said that, because like I've long been a proponent of, you know, like the capstone thing reminds me, it's almost more like an apprenticeship.

17:44we think of like internships and it's like i mean i remember when i did one in college and it's like it's pretty boring there wasn't a whole lot of responsibility that went along with it you know

17:53David Ri: i think i answered phones and and did filing exactly it was like it was not that in depth that i went out of there going that was that important yeah put it on my resume yeah i did that but yeah the apprenticeship model which i think that kind of gets at is like you actually do. You are learning. You are doing the work and doing it in a depth that you can, like you said, like they sounded like there were three, four years in. Yeah, it blew me away and it helped me set a new standard and educate the organization on what we should expect out of our early career talent. And I hope that also sets an organizational level competency around.

18:36David Ri: We should probably also be expecting a lot more out of our professional workforces. And we've been educating our leaders on what I would call kind of evolving your value proposition. And we take them through like a multi-module training workshop where, and you would be surprised, or maybe you wouldn't be, but it was very hard for them to articulate their current value proposition to the firm. And so going through like, hey, when your entire team is augmented, enhanced, replaced by AI, what are you going to do with that value? Right. What do you how are we going to recapture that value, whether that's higher throughput, better quality deliverables, more time with clients, like more time to to build future lines of revenue?

19:28You know, like people first, I think it needs an instrument or it's just words. right? It's a slogan, basically. I'm curious, you know, what is yours? Like, how do you distinguish adoption from compliance from people who are just sort of routing around the thing?

19:42David Ri: Yeah. That was another very early intentional decision. We actually spent a good chunk of our budget on engagement and enablement professionals that were helping with, at the very beginning, helping all of our employees establish a level of AI fluency. So it was everything from introducing them to new tools, kind of giving them fundamentals around how generative AI works and how we're protecting their data and our clients' data, getting them really comfortable with all the questions that were coming up at the very beginning, which is like, what happens when I, when I, you know, upload a client income statement into, you know, into a GPT.

20:27David Ri: It's like, well, here we've built it so it's safe and it's firewalls, but, you know, use your best judgment. And so we were also helping them with the balance between what you can trust a generative AI to do and what you can't, right? A trust but verify methodology, right? And getting that ingrained into people very early helps them with the acceleration of adoption of other more advanced capabilities as they were coming out. So we had, you know, we were building tools. We were rolling them out. We were training people on it. We built a champion program that could take all of those trainings that we had built and tailor them for the individual roles and personas within all of our business lines of corporate functions because we're a massive company with, you know, operating in 80 countries, right?

21:17David Ri: There's a lot of things that have to be considered at the local level that a central function isn't designed to support at scale. So we had to enlist the help of the experts within those domains to help us establish the next level of fluency that our teams needed to be able to start innovating on their own. So we're running up against time, but before I let you go, I want to give you a hypothetical because I love a hypothetical. Sure. Um, so you have a chief people officer, let's say they have zero agents in production at this point. Uh, but their board is starting to put pressure on them. They want to see some progress.

21:57They're asking why that's the case, right? So they got 12 months, they got a budget.

22:03David Ri: Yeah. What do they do in months one through three to get themselves, get the ball rolling? And what's the thing that they'll be tempted to do first that maybe you think, don't do that. That's a little bit of a waste of time. Yeah, I think the first thing is to establish a group of people who you're very comfortable with experimenting across the organization. I'm not sure if I would go full tilt into FDs, but like domain experts in your business line and corporate functions that are technically literate, that can start to help you define what are some real use cases that even experimentation is going to be net ROI positive.

22:40David Ri: I'm trying to go really wide and really broad all at once is going to incur a lot of cost. And if you don't have people internally that are educated on your different domains and your business or aren't educated on the AI side of things, you're just going to balloon costs. And it's basically what you've outlined through these questions is going to start happening to you. Right. So building success in a few use cases and then scaling those and then using that ROI to reinvest in more or the more horizontal, what I would call horizontal or general purpose tools delivered to your organization, I think is key.

23:25David Ri: And the other piece is that enablement function, that fluency. You have to get your people really comfortable leveraging tools and you have to incentivize them to use it. I think something that I continuously reference and am just so grateful for is our CEO, Christian Ulbricht, who really set the vision and empowered people to use AI, right? And set global KPIs around fluency and adoption and usage. So there was no punishment for using these tools, right? It wasn't like, oh, you figured out how to automate your department? Cool. We're just going to let everybody go. That was like never, that was never a narrative that existed inside of JLL.

24:10David Ri: It was, we need to be able to leverage AI to best serve our clients and better serve our people. And so all of our executives, one had to build KPIs around the achievement of that and also role model those behaviors. Like our CEO was very straightforward around, you know, every single one of you needs to be using AI tools and I'm going to measure it. Right. Because if your people don't see you using it, they're not going to feel comfortable or empowered to use those tools. And he wanted the innovation to come from the bottom up just as much as he wanted the role modeling to come from the top down.

24:53David Ri: So in giving a, you know, any C-suite leader advice, I would say you need to role model that behavior as well. And you need to align incentives to the behavior that you want to see in your organization. Well, thank you for coming on the show today, Carl and Power, ladies and gentlemen. Until next time. Thank you so much. Well, listeners, if you haven't done so already, head on over to peoplemanagingpeople.com forward slash subscribe. Get signed up for the newsletter. you'll get this podcast and everything else that we create straight to your inbox and until next time