Panel · 2:20 to 2:55 PM · Wednesday, May 13, 2026
So let's have a quick name, rank, and serial number from our panelists, starting over with you, Leslie, on the left.
Thank you. I'm Leslie Peeler, I'm the COO of Cenlar FSB. We're a wholesale bank that specializes in the mortgage industry in the United States.
Terrific. Adi?
Yeah, hey everyone. I'm Adi Shetty. I work for Visa. I think I'm the only HR representation here on this panel, but again, I'm responsible for all the operations and the systems and the workforce infrastructure for the enterprise. Terrific. Yeah.
Hi, I'm Mary Lacity. I'm a professor in the Sam M. Walton College of Business at the University of Arkansas, and I live and breathe and teach young people.
Mark Hodges. I'm with Acresis. I've been studying outcome-based pricing for 20 or 30 years. It's very hard to do. I'm based in Los Angeles, and most of our clients are smaller channel partners trying to sell in the enterprises different AI solutions.
I'm Vijay Vijayasankar. I lead AI at Genpact. And I'm sorry.
Just, you know, Adi, Mark was one of the original protagonists of HR outsourcing back in the day. I'm sorry. Anyway, HR, by the way, needs to be deep in the AI conversations. So we're going to go. First question is breaking out of this cost trap, how do you reframe the AI investment conversation without losing the CFO, shifting that conversation to outcomes without derailing the business case that got you here? Let's start with you, Vijay, Genpact.
So I would start by saying lowering cost is an outcome. It's not a bad outcome, right? But the rephrasing that is needed is what do you do with that reduced cost, right? So I generally rephrase it as how do you compound outcomes over time as opposed to how do you reduce cost? Because reducing cost usually equates to, yeah, you take some headcount out now and then, okay, then what happens, right? There is not an end to that conversation, which is usually where projects go to die. So reframing that conversation to how do you keep compounding the outcome, which is actually a lot easier with AI, because AI, thankfully, has the ability to keep getting better over time, so you can compound outcomes, right? So that's how I like to think about that.
Terrific. And Adi, to you, I mean, do you report to the CFO in your company?
No, I report to the vice chair, but I'm friends with the CFO.
Oh, okay.
He's one with short hands and deep pockets, so you've got to make sure. So let me, I think the equation with our AI office is actually, I mean, the CFO office is, is not too bad. I think they've come to terms with what AI can do to us and what it can do to the organization. The challenge is it's not the problem of the investment. I think it's the pitch, right? Every time we go in as an HR or any other function goes in and says, I can reduce, I can bring efficiency by 30%, the CFO is hearing your budget's going down by 30% next year. That means congratulations, we've officially defunded ourselves for the next year. And I think that's where the challenge is, that how do you bring the CFO along in having that conversation with him about, yes, I do have a short-term goal which has the stuff with results. So whatever I can do to gain efficiency, improve time to hire, all the work that we can actually have points on the board. But there's a bigger story to tell, right? I think we need to stop selling the shrinkage because we've been doing that for the last 18 months and now it's starting to hurt us. Because every time you say efficiency, they're thinking lower budgets. That means we're getting less money to work with. I think we have to stop selling the shrinkage, and we need to start selling capacity. I think, as you were saying, that we need to start talking about conversations where you can ask for the important work that never gets done because our people are buried in the wrong tasks, right? If you get AI in picture, and then you get them to surface that out, that's where you create capacity. I think that's the conversation or the shift that we need to have. And our CFOs are coming to terms with that conversation now. Yeah.
Mary, you must have the answer to this question.
I like to say that every business, and this is going to tie into what you just said, I think every business case needs to have three sources of value. One is for the enterprise, so that's the language the CFO wants to hear, typically cost savings, ROI. One is value to your customers, and one is value to your employees. And since I'm a professor, I always like to tell a story to illustrate a point. And I'll give you an example of a business case for an AI computer vision for a large retailer on checking out. Who really wanted it was the people in charge of the customer experience. They didn't like that people had to go stop on the exit, have a human being, look at the receipt, count the things in the items. It's not a good customer experience. But how do you create a business case for that? So when they went to the chief financial officer, they sold it on a shrinkage play. We'll have less shrinkage because we'll have more accuracy. And also we don't need to have so many human beings up at the front to do that. But the customer experience now after implementation, 85% now exit with absolutely no friction at all. And the human beings who are the employees, they didn't have to be the cops saying, I'm counting the wrong number. They're getting a ping on their app that's saying, oh, let's see, our AI just wants me to question something. So it took the pressure off of the employees. And, of course, the customers had a great experience, and they got a positive ROI. So every business case should have three sources of value.
Thank you. Three sources of value. I love that. So, Leslie, we're going to move to you now. And why are so many AI pilots stuck at productivity? What actually unlocks the move to higher value outcomes?
I like Mary's answers. So can you? You want to? That was really good. I love storytelling. So I actually think it relates a lot to the first question about the position of the CFO and the decision-making process. You know, if the CFO is focused on, you know, expense containment, then your projects are typically going to line up with cost containment and reduction elements, which typically line up then with productivity or labor cost improvements, right? So I think the two questions are very much related. I like what Mary said. I don't typically ask for three. I do ask for two, right? It's got to be something that improves my ROI as a business, but it also has to improve our customers' experience or client sat or reduces risk or does something else. You know, I think the other element that is important is, and when you add that second element, you're then kind of shifting away from this, you know, sole focus on productivity. So if my team brings to me something that's purely a productivity gain, I don't do it. I don't prioritize it. And I think the second thing is really centered or focused on the talent question that we've talked about a lot throughout the day is, you know, who's actually controlling the funnel of AI related projects? Are they people who are really thinking about how to reimagine a process? Are they people who are playing around the fringes of productivity? So I think that in both of those areas, kind of focusing on the why question better and bringing in the right talent in order to run these, you know, run your AI pilots, you know, those are probably the two leading factors for me.
Yeah, and I'm going to flip this one to you, Saurabh, because you've been having a lot of conversations with enterprise clients dealing with this. Do most start with a cost mindset and they make this shift as they start solutioning? Is it a flip?
I think we're obsessed with cost. And I don't think cost reduction is a bad thing, like Vijay, you mentioned. But being obsessed only about cost is what's driving this. And I think whether it's two, three, four, there should be something more than one. And I think where a lot of pilots are getting stuck, Phil, is where we are obsessed by productivity or cost saving or efficiency. Can we do this slightly cheaper? Can we do this slightly faster? And that's basically it. But we are leaving aside the bigger gains of either some business outcome, something else. We at HFS call it productivity, performance, personalization, and prediction, the Four Ps of AI value. And the reason is those things are harder to measure. You can measure number of people. You can measure dollars and cents saved. How do you measure customer experience? It's so easily said, but it's a very touchy-feely thing. And all CFOs are accountants, right? And they're left-brained, not right-brained. So I think that's the issue. And I think if we have to go to that outcome ambition, we have to start creating balanced scorecards. Is where cost has to be one of the dimension. I think if you're not reducing cost, then what are you doing? But it needs to have some other dimensions to it. We at HFS are looking at these four Ps. Maybe different industries or different clients or different organizations will have their own balanced scorecards. But every AI initiative should have a balanced scorecard. And I think that's where we'll, some qualitative measures, some quantitative measures. If you have four things, two will, you'll get two at least, right? So then it looks good.
I would also add that a lot of those pilots deserve to be in purgatory. And it's usually because of cost. If it's not carefully designed, the costs are astronomical. We talked about tokens earlier, and there's so many layers of the AI solution that compound the cost that it's quite shocking how often you get invoices and clients are pulling the plug on those.
Maybe I could add an HR angle to this because I'm the HR representative here. So it's like the saying, right, operations successful, patient dead. I think that's what the pilots are. Usually you get the pilot right, but sometimes it doesn't scale. And sometimes it's not the technology. It's the organizational design around it. It's the culture. It's the people. It's the training that lacked. I think that's where we kind of disconnect the two. Our focus on budget spend is on the, I think there was a study by BCG that said that in order to gain AI value, there's a 10, 20, 70 rule. 10% is on the algorithm. 20% is on the data and the tech. And the 70% is on people, process, and culture. And we invest 95% of our budget on the 30% that we care about. We're not talking about the remaining 70, which is like the success driver. I think that's where the disconnect is, where the points of failure is that. It's truly the culture that we have. And if you look at why does it fail, right? If you start to unpack that and peel that onion, there's like, since we're going in threes, I'll give you a reason, the three primary reasons I have observed from my seat, right? One is, again from an HR standpoint, there is no true owner to the result. I have a vendor that has the tool, I have IT that has the model, I have functions that have use cases, but nobody owns the results. When nobody owns the results and it's failed, then it's nobody's responsibility and everybody's surprised. I think that's where we don't have an outcome-based owner for some of these pilots that we put out. The second thing that we see is then, you know, when we put these projects together, we're always trying to just, I think, redecorate the current process with AI. We're not really re-engineering it. I think that's where there is no fundamental shift in the gain, right? I think that's where we're saying the pilot works because we took a process and we identified sweet spots where AI could do some agentic work, and that's what kicked in. We need to scrap the whole thing and start to redesign the whole model in order to really get true value. And there's studies over this time and again that tells you that the only way to get true gain on AI is if you redesign the entire process and not replicate the process and decorate it, that won't get you any value there.
But that's hard, right?
It's extremely hard because you don't know where to start.
There are only 10% who get to that next stage.
Isn't this about figuring out what data do we need to make quick, slick decisions internally and externally? And how do we get there? And then you wind back from there. You build back. You don't just start doing little tweaks here and there. And it shouldn't matter if that data's coming from PeopleSoft or SAP or Workday or wherever you're looking to get it. It's that new control layer that you're building, right? So isn't this more about figure out the data? That's the outcome, isn't it?
It is quite complex, even in boring back office operations, right? So if I take an example from, say, finance, right, CFO. You can take an example of a vendor sending you an invoice and then saying something like, if you pay in the next 10 days, I'll give you a 10% discount. The vast majority of the Fortune 500 companies that I know of have policy-based decisions on how to take a discount. We are living in a world where, at real time, we can go check what other obligations exist for the companies. There are a big bonus payout waiting or an IRS payout waiting and so on and take a much more informed decision as opposed to letting agents do accounts payable and take, say, a 40 headcount out, it is a whole lot easier if you can manage the treasury better. But who is measuring that outcome? So when we make the business case for the pilot, usually the business case is about, okay, prove whether AI works. Yeah, the pilot kind of it's easy enough AI works. It's a decent technology, it will work. But then if that's all you're measuring and that's all you're aiming to solve, that's not going to scale. What's the point? So just measuring that business or thinking through what else can happen as value add is what unfortunately the industry lacks for the most part.
Let's go to Question 4, I think, beginning with that. And Leslie, I'd like you to start here. But when we talk about a deal structure that prevents the transformation it was supposed to enable, what do you do? I mean, do you go back to the drawing board or do you just persist with it?
Yeah, we're actually confronting this question right now with one of our largest spend suppliers. And so I actually want to attend one of your labs soon to see what I can learn about it. We have many of our contracts. I've been with this company for a year and a half, a year and three quarters. So many of the contracts predated me coming on the scene. And they're almost always some kind of labor-based agreement where there's this perverse incentive around innovation and what the vendor partner earns, right? So we've got to, you know, the conversation I'm having with them right now is going to more of a, you know, gain share model. They've established some partnerships with some really sophisticated AI partners. We want to take advantage of them. So how do we introduce some of these concepts around, you know, outcomes that have economic value for us that they share in, they still make, you know, gobs of profit, but they are going to have to give up, you know, give up some revenue so that we have some expense reduction. And that's kind of the, you know, that's the, you know, the item that we're grappling with. I don't think, you know, charges per compliant transaction, like compliant transactions, that's just table stakes for me in the industry I'm in. So I, you know, we already have, you know, controls around that. But how do I link, how do I link their performance to my average cost per unit or, you know, or client sat or leads generated or different things like that is probably the next level. But again, the reason it's so hard is because they all still want to preserve revenue while having labor-based agreements.
I mean, Mark, there's a lot of this as well around the mindset of the buyer. So a CPO and a CFO often like predictability. And they don't want to have to write a huge check to Genpact or somebody in a gain share. I mean, from your perspective, what do you think needs to change?
I think for larger AI rollouts, if you come across a corporate culture, anybody that's managed large or global accounts which knows which ones are really risk averse. It could be regulatory. It could be discrete manufacturing that's just squeezing every cent out. That's just a disaster for something as radical and innovative as AI. So one intervention is to just stop. Another one is to try to find a progressive business owner with enough juice in the organization. There's a couple executives here from corporations earlier that had that kind of juice. I heard it. I heard some F words go back and forth. And then there's the opposite personality that does really well, that are really elegant, and no one knows that their cheese is moving and chess pieces are moving, but the project is moving along. But as soon as procurement gets in there and they're driving it, AI is too early for them to understand how to price it accordingly. So I would shut it down.
All right.
I think if I can just add to that, at least from my perspective on this one, is that the way I see this is it's like you walk into a gym and you've hired a personal trainer and he's charging you by the hour and not by the loss of weight that you're trying to get to. And then you're surprised that he's calling you for longer workouts and repeated days to show up. I think that's how we feel with our AI contracts. It's more input-based. How many seats can you take? How many tokens can you buy? How many hours are you spending in consulting with them? How many meetings do you get? It's all transaction-based. So that's not a vendor relationship. That's like a gym membership that you can't cancel. I think it's one of those things that's happening to us.
AI is there.
Can I just say historically, because I know some of you people for three decades in here, we've been talking about outsourcing-based contracts. And research has shown that you need four things in order to get outcome-based contracts to work. First of all, they have to work for both the client and the service provider. And they work where there's relationships, have high levels of trust and transparency. You have to also be able to measure the outcomes. You have to have strong baseline measures. And most important is that the service provider has to control most of that value generation. You cannot hold a supplier accountable for outcomes if 90% of the effort is affected by what the client is doing. So the question in the AI world is how many of your AI contracts can meet those four criteria? And are those four criteria still relevant in the world of AI? I think they are. I don't know.
I think the deal structuring is just archaic right now. You know, the whole intellectual property of deal structuring was created in the 1990s. And ITO and BPO was a new thing. You know, offshoring was the AI of 1990s, right?
Yeah.
We've added bells and whistles to it. But I look at RFPs, etc. They're the same damn things. You know, we're asking the same questions. We're still doing the dog and pony show for six months. Now, if you're taking six, eight months to select an IT service provider, what are we doing? The world is changing in six months. It's almost become like you're trying to subscribe to Netflix and you're counting DVDs. It's just not going to work.
I haven't seen it for an AI solution provider, whether it's from a larger contractor or not. But for other radical innovations that this group talked about earlier, whether it's going to cloud or offshoring or business process outsourcing, there was one or two service providers that recreated the contract. Salesforce recreated how you bought software. That took seven years to realize how to buy SAP, you know, what the SAP pricing was, right? And so if any service provider has a better mousetrap, I haven't seen it yet, but that's what I think it takes. Because procurement gets very fearful when someone has a better contract. Lawyers get very fearful when there's a better contract. And they will stampede it if a market leader says this is the new standard. I haven't seen that yet.
I 100% agree with you, Mark, but my point is let's at least be cognizant of the fact that we need a new contracting structure.
Sure.
Let's not try and buy something new with the 1990s structure.
I completely agree. It's hard because if you look at the service provider revenue, all their comp systems are driven by that. And so it's not easy. I just think there's going to be innovation in the negotiation, the deal structure, the contract. And if I had to bet, I think it's going to come from, I think there'll be one from a legacy outsourcing provider. I think there'll be one that calls themselves AI native. And they may apply to different things. But I think that's where the source of the innovation is going to come from.
But the procurement function, right, of all the functions I mean, I also have been the bane of my existence, right? Yeah. There's a philosophical problem in enterprise. What is procurement's job? Procurement's job, if you ask a chief procurement officer of a good-sized company, would be that of a risk manager, right? They want to minimize the downside risk, which is a laudable goal, right? I have zero problems with that. My usual question back to them is, who is in charge of capturing the upside? Because you have vast majority of the challenges, right, on talking about outcomes and any creative deal structure. It largely depends on this balance of downside versus upside. If you over-index on downside, then you're consciously giving away the upside. Because a vendor also then has no incentive to share the upside. Right. So this, still that philosophical change happens, and I also have some sympathy for my friends in procurement functions, because they are brought in really late into the process. There are very, very few business leaders who, when they have a bright idea, the first call is to the chief procurement officer to say, come brainstorm this with me.
Right.
Doesn't happen. But if you make that happen, if that cultural switch happens, I think a lot of these challenges will go away. I agree.
I think the one thing that I would add to just the deal structure is also how to get the hell out of it, right? I mean, once you're in a deal, you are so thick into the vendor and the commitment that if you don't have good portability clauses, you don't have data migration rights, you don't have exit architecture laid out, you're going to be stuck with them whether you like it or not. I mean, that's the problem that we see when you're going into more AI deals. You really want an exit model also defined. And the vendors who challenge you are the ones you absolutely need it with. I think that's the one that you need to kind of think through as well.
So your message is we need prenups.
Kind of, yeah. Yeah.
Okay, so let's get to tokenomics. And how do enterprises get their heads around this consumption-based AI economics? Maybe, Vijay, you want to kick that one off?
Yeah. I'll acknowledge this by saying this is a new challenge, right? So I'm rather convinced that nobody has a perfect answer to this. But I'll offer two opinions on this. One is the financial aspect of this. Rather than think about it as a cost, it should be, well, cost as in an overhead, right? You know, like overall IT cost or something. If you think about it as a variable cost of revenue, the thought process changes very quickly. As in, overspending on tokens assumes that there is no extra value created. But if in the result you can say that, and hence, every 10% increase in tokens correlates to 1% increase in revenue, or a 1% decrease in customer churn or something like that, then tokenomics gets a whole different meaning. So I think that's one. The second aspect is a lot of token spend happens out of sheer engineering stupidity, right? Maybe not the most PC way of saying it, but as an engineer myself, right, there are tools fit for every job. Not every last thing needs to be done with AI. There are many, many other ways of doing things. So we waste tokens a lot. It happens with me. It happens in my team. It happens all around at my clients. Till we get out of this obsession, I know Jensen Huang said that every $500,000 engineer should pay $250,000. Yeah, of course it helps his business model. He should say that. I respect that as the CEO of NVIDIA. That should not be taken at face value. Use it for things that add value. Don't just measure yourself on token consumption. That's pretty dumb.
There's a lot of really good token dashboards that don't really align with the underlying business value. Someone earlier mentioned it measures activity really well and that results. And that video you showed earlier was hilarious.
Yeah, you know, nobody knows what tokens are. That's the fundamental issue. Ask any CFO or anybody, how many tokens do you need to match an invoice? What does this even mean? At least rate cards and hours of people you know is working for one hour. Okay, I need to pay him a beer or something. But in tokens, there's nothing. You know, it's just such an obscure techy metric that you can't really convert it into anything real and then you're trying to budget for it. It's just impossible.
You've got leadership which have no clue what they're doing with AI. And so they're going to their staff saying, use AI. It's bullshit driving bullshit. Yeah, it's basically that.
You know, Uber bought a bunch of tokens, a huge bunch of tokens, and because of the token-maxing culture, they blew through their budget in four months. So what are they going to do? So they can't do AI for the next eight months.
Their revenue and profit did not increase materially either.
They were creating motivational agents.
So a final question. Let's just go down the line to finish this off. One thing, one concrete action that every leader in the room must do differently in how they buy, govern, and structure AI to get to outcomes. We'll start with you, Leslie.
So I would say use your AI labs to develop the proofs of value in order to then lead, actually let me take a step back. We use our AI lab, we don't have that count against anyone's projects from a budget perspective. It's just investment funding. So the one response would be allow for experimentation and investment in your AI labs without harming your actual then implementation choices.
Okay, good answer.
I think I'm going to keep it simple so that this is how we challenge everything, right? So this is more of a habit than a template. Every time we have an AI opportunity, we just ask the four questions, right? Like how, what, who, and under what circumstances. So I'll just explain what that means. We ask first about what is the business outcome you're buying because you're either deploying AI or buying AI or implementing AI, that's what. And then who is accountable when this business outcome is not met? And how do you know that this outcome has arrived or you've reached a result in 90 days? So you're not churning and you're not doing this again and again and you're trying to find gold when there's nothing there. Or under what circumstances do you pull the plug? If you get your team to answer just these four questions every time they come in with a proposal, if any one of this is not solid, you know that you don't want to go forward. I think that's how we've started to keep it very simple. Start with the four questions, just how, what, who, and under what circumstances will you kill it? And just try to understand the scope there. And I think that helps us at least govern it in a way that is relatable to how we operate. Yeah.
Well, you kind of stole mine, Adi, because one of the ones I was going to say is that every one of your AI initiatives has to have a business outcome owner. So you've said that several times. So I'll give just another possible thing to experiment with. Focusing on how do you get to the outcome base, let's go pick one of your service providers that you trust today. And if you meet the criteria of some aspect that is measurable and that you have benchmarks and that they can control the value chain, just experiment with part of your live contracts on an outcome base. And if you don't know how to do it, we know where there's hope.
Thank you for the sales pitch. Always appreciate it.
Thanks, Mary. So earlier we talked about innovation in the contracts and procurement. I would suggest to have two or three service providers if you're buying, maybe an incumbent, maybe a disruptor, a challenger, but really evaluate them based upon the lack of fear of failure that the account team you're working with is not going to get fired or terminated if they don't win the larger project or get enough account share because if you have an account team that's just trying to cover their ass, you're not going to... AI breaks things, particularly if you're trying to scale an AI solution. If you're trying to stabilize an AI solution, that's probably safer with more legacy service provider compensation schemes. But if you're trying to scale something, there's a lot of things that are gonna go wrong. Indicator lights are gonna be red, yellow. So it needs to be the founder's buddy or someone who's a high potential in the organization that's not gonna let disaster wreck their career.
Thanks, Mark. And to finish up, Vijay. 10 seconds, I would say. Don't let procurement and legal set your AI ambition. Bring them early into the process, problem-solve with them.
One thing?
Do something.
Thank you. There we go.
If you don't have an account, Register here |
With the exception of our Horizons reports, most of our research is available for free on our website. Sign up for a free account and start realizing the power of insights now.
Our premium subscription gives enterprise clients access to our complete library of proprietary research, direct access to our industry analysts, and other benefits.
Contact us at [email protected] for more information on premium access.
If you are looking for help getting in touch with someone from HFS, please click the chat button to the bottom right of your screen to start a conversation with a member of our team.
