The Order Is the ROI

Global AI investment is crossing $1.3 trillion, and 95% of pilots are delivering no measurable P&L impact. That gap isn't a technology problem — it's a sequencing problem.

Jackson and Scott unpack why the money isn't following the results and what the CHRO needs to do about it. Five moves, in order. Get the sequence wrong and no adoption dashboard will save your business case.

What You'll Learn

• Why "we bought the AI module" is not an AI strategy — and why adoption metrics measure the wrong thing
• The constraint inversion loop: how vendor demos drive tool selection before any business problem is named
• Why AI is fundamentally different from prior technology waves — and why the headcount elimination instinct misses the real opportunity
• The five-move constraint-first framework and why the sequence matters as much as the moves themselves
• How the CHRO who owns the constraint review cadence owns the AI accountability conversation for the entire enterprise

Key Quotes

"Usage tells you people are using the tools. It tells you nothing about whether the tool is moving a business outcome anyone cares about."

"The CHRO who treats AI governance as an enablement function is handing away their most important capital allocation role in the enterprise."

"Redeployment is a strategic expansion question. Where does the free capacity go? What can the business do with it that it could not do before?"

Sources for Statistics Cited

• $1.3T global AI spend by end of 2026 — Source not verified at $1.3T; Gartner puts 2026 AI spend at $2.5T
• 95% of AI pilots failing measurable P&L impact — MIT NANDA Report, 2025
• 14% of CFOs see clear, measurable AI ROI — RGP 2026 CFO Research Report
• 25% of AI initiatives deliver expected ROI — IBM CEO Study, 2025
• 42% of companies abandoned AI projects in 2025 — S&P Global
• 61% of CEOs under pressure to show AI ROI — Kyndryl 2025 Readiness Report
• IBM tripling entry-level US hiring in 2026 — Bloomberg, Feb 2026

Transcript

AI Spend and the Wrong KPI

Scott: Companies are spending more right now on artificial intelligence than any other technology in history. And the primary thing that most of them are using to evaluate whether it's working is whether people are logging in. But usage is not the metric the board is going to ask you about.

Jackson: I'm your host, Jackson Lynch, and today I am joined by my co-host, Scott Morris. He's a former CHRO with the scars to prove it, the alleged survivor of more vendor demos than any human being should ever be asked to endure, and is now a vendor and the founder of Propulsion AI. Scott, I want to start with a number that I think most people in our audience already know, but has not yet been named clearly. Globally, enterprises are on track to invest over one trillion dollars — with a T — in AI this year. And according to nearly every major research firm tracking this, more than 90% of that investment is not producing measurable business results yet. Now, I don't think that's a technology failure. I think it's an implementation failure. I think it traces back to a single question being asked in the wrong order.

Scott: I agree with you, but before we go any further, I just want to be clear, Jackson — with our audience — that this is not a diagnosis from the outside. You've watched the patterns. I've watched these patterns run inside organizations at every level, and I lead a technology company. This is pattern recognition from being in those rooms when the decisions got made.

Jackson: And that's exactly where this conversation needs to start, Scott, because these are not careless mistakes. They're very logical and rational ones.

A Quick Systems Thinking Reset

Scott: So before we get going, Jackson, I want to say a big thank you — and I know you do too. A shout out this week to Shannon from Plymouth Meeting, Pennsylvania. Shannon, thank you so much for being a part of the Talent Sherpa community. It genuinely means a lot to Jackson and me that you're here. And thank you to everyone else for tuning in. Whether you're joining us from Oakville, Ontario, or Smyrna, Georgia, we're really glad you showed up today. Let's get going.

Jackson: Yeah, it means everything, and we'll do our best to make sure it's worth your time.

Scott: Hey, Jackson, just one more thing before we get going. Let me ask the audience: take a second right now — just take 30 seconds, actually — and think about a performance problem that you're focused on. Now ask yourself honestly: is this about a person, or is it about how the system around that person was designed? The reason I'm asking the audience that is that a recent graduate of your CHRO Ascent Academy described the organizational design module that you run this way: "immediately relevant." Said, "we applied the techniques that same week." What you're building inside of the CHRO Ascent Academy is a program for CHROs who are learning to read the difference before they act, and not after. Seats are limited in this program, and I would highly recommend everybody goes to mytalentsherpa.com and figure out how to get into the CHRO Ascent Academy.

The Trillion Dollar AI ROI Gap

Jackson: Yeah, I appreciate that. And we have the next one kicking off May 15th. So let's get into this episode. Here's where we are today. Global AI spending is projected to surpass $1.3 trillion by the end of 2026. 85% of organizations increased their AI investment in the past 12 months — that number actually feels low to me. The commitment is serious and it's growing. And the board conversation about AI return on investment is no longer optional.

Scott: And the ROI just isn't following — and that's why we're raising this point. Research from multiple firms is landing roughly in the same place. 95% of AI pilots are failing to deliver any kind of measurable P&L impact. Only 14% of CFOs surveyed said they could point to a clear, measurable return on the money they had invested in AI. Not insignificantly, IBM found that only 25% of organizations report AI delivering any kind of expected ROI. And 42% of companies abandoned most of their AI projects in 2025 before the measurement conversation even happened. Now, I know that's a lot of numbers, but here's the take: it isn't working.

Jackson: Yeah, that was a lot of math. And here's what I think is underneath all that data. The spending is real, the technology is genuinely impressive, and the tools probably do what the vendors say they do. But the question that most organizations are asking when they enter an AI deployment is: how can we use this? And that question, asked first, almost always produces bad outcomes. Because it puts the technology before the problem. You end up with a solution before you've named the constraint. I want to put a real example to this.

Scott: I worked in manufacturing, and I watched this happen in a manufacturing company that was excited about AI for workforce scheduling. Genuinely excited. But when someone stopped and asked what was actually constraining the operation, the answer wasn't scheduling — it was unplanned downtime. And the scheduling tool would have made an existing problem marginally more efficient. The constraint, once you named it, pointed to something completely different as an AI application: predictive maintenance. And the business case for that was more than three times larger.

Jackson: I've seen that too. And the gap between the tool being evaluated and the constraint actually limiting the business is where most of that trillion dollars is going to disappear. Because when you start with a tool, you solve the problem the demo describes — which, by the way, is what human capital leaders often do. They try to find the problem in the organization that they can solve with the program they've already built. That doesn't always drive where the business problem lives.

Scott: And I think boards are starting to notice that. 61% of CEOs say that they are under increasing pressure to show some kind of return on AI compared to just a year ago. And that pressure is arriving faster than most organizations have built the infrastructure to answer. The CHRO who can walk into that conversation with a constraint-first framework is operating at a very different altitude than the one who walks in with an adoption dashboard.

Assumptions That Quietly Kill Value

Jackson: And that difference — that altitude difference — is what this episode is going to be about. So let's start by naming the assumptions that get the organization stuck here, because they don't persist because people are careless. Each one makes sense in the room that it's made in. Start with the one I keep running into.

The first one sounds like this: "We bought the AI module, so we are going to be doing AI." I understand where that comes from. The purchase was approved, it might have been in your existing tech stack. But buying a tool is not the same as deploying a capability against a problem. And deploying a capability is not even the same as solving a business problem. Three different things, and most organizations are celebrating after the first one.

Scott: And the signal that this assumption is running the room is when the all-hands announcement about the new platform lands with more energy than the conversation about what the platform is supposed to change. The purchase becomes the proof point, which means that the question about what actually changed in the business never gets asked.

Jackson: So let's move forward. The second one sounds like patience, but in practice is more like avoidance: the return on investment will reveal itself once adoption is high enough. Look, adoption might be a legitimate thing to track — jury's out — but usage tells you that people are using the tools. It tells you nothing about whether the tool is moving a business outcome that anyone cares about.

Scott: I've sat in renewal conversations where the vendor came in with strong adoption data, and everybody in the room nodded — because the counterargument requires a measurement framework that never got built. You can't dispute the adoption numbers if you never established what the business outcome was supposed to be. The assumption protects itself.

Jackson: And that leads directly to the third one: the business case follows the investment. We will figure out the ROI story once we have the data. No one goes into it actually thinking in those terms. Maybe sometimes organizations genuinely believe that, but I think more often it's a signal that the constraint was never named before the tool was selected. And when the constraint's not named, there is nothing to build the business case around.

Scott: You're buying technology to buy and deploy technology. Now, here's one that I think is the hardest to name in the room because it sounds like cost discipline: the goal of AI is to reduce headcount. I totally object to this. I run, as you know, an artificial intelligence company. I object to this argument — get the humans out and get the costs down. And I've been in multiple executive conversations where that was the primary logic, and nobody challenged it because it sounds responsible. I think it's completely wrong-headed.

Jackson: It is, and here's where the logic breaks down. Most prior technology deployments were about making humans faster at existing work. Spreadsheets made finance faster than the abacus. CRMs made sales faster, even though every salesperson complains about having to use it. The mental model that organizations have carried into AI is the same one — but AI is genuinely different. It allows organizations to reimagine the distribution of work between tools and humans, and between full-time, part-time, and gig workers. It allows you to rethink that whole thing. Which means that when you deploy it well, you almost certainly do have additional human cycles available. But the question is: what do you do with those cycles? That's where the real strategic opportunity lives.

Scott: You're going to get zero argument from me on that. Wrong-headed organizations are asking, "How do we get humans out of this work?" Right-headed ones are asking, "How do we take pieces of their work, utilize artificial intelligence and other technologies to handle those pieces, and redeploy those cycles so that we're increasing capacity? How do we double innovation? How do we serve the customer in ways we've never had the cycles to do before?"

Jackson: And there's a real forecast about what happens when organizations choose that first path at scale. When enough enterprises eliminate rather than redeploy, power concentrates in fewer hands. The people running those organizations end up with enormous leverage for a little while and almost no bench beneath them. The people who redeployed thoughtfully end up with organizations that are faster, deeper, and harder to compete against. A real example is IBM — they re-examined all of their entry-level positions, used AI to redesign the work, reset the outcome clarity for those roles, and are now hiring 3x what they had been doing historically. You won't necessarily see that today. But what you will see is decreased cost, increased bench strength, and a more well-developed next rung of leaders three, five, ten years from now. The folks that just pull all of that to the bottom line — they're going to have to spend more money to fill those gaps as we look into the future.

Scott: And I think that conversation is why it's so important for CHROs to be engaged in this discussion. Our job in AI deployment is to enable technology, run change management, train the users, and track adoption — and all of those things matter. But if that is where the CHRO's role starts and ends, then the CHRO is at the wrong level, because they're not talking about the constraints of the business.

Jackson: I agree. The CHRO who treats AI governance as an enablement function is handing away their most important capital allocation role in the enterprise — not only right now, but in the next five to ten years. Constraint identification, business case design, redeployment planning, reallocation of freed time toward other things — that's where the strategic work is. If HR is not in the conversation, somebody else is going to be. Finance will come in with a cost lens. Legal will come in with a compliance lens. IT will come in with a security lens. We're the only ones in the enterprise with the mandate — or who should have the mandate — to look across the entire organization and figure out how to optimize the workforce. That's what this is about. If you don't do it that way, I don't think it's going to go well.

Two Loops That Keep Failing

Scott: You and I are going to get into what to do about this. But before we do, I want to name the structural mechanism that's underneath all of this. Because if we don't, the fixes that people reach for treat symptoms. There are actually two reinforcing loops running at the same time. Why don't you take the first one?

Jackson: The constraint inversion loop. A vendor brings a demo. The demo is genuinely impressive, and it usually does exactly what it claims. Then someone in the room sponsors the purchase based on what they saw — not based on a defined constraint the organization has agreed to eliminate. Deployment begins. The question everybody asks is, "How do we get people to use this?" — which is a legitimate question. But the more important question is, "What business result have we committed to moving?" And that question never gets asked because there's no constraint at the other end to measure against. Adoption numbers come in. Leadership is satisfied because people are using the system. And the loop resets.

Scott: And to be really clear — no one in that loop is doing anything wrong. The vendor has often never even seen the technology they're selling in the way it gets implemented. I think everyone is trying to do their job and trying to do it well. The problem is structural. The constraint was never named. So there's nothing to hold the loop accountable to. And the loop runs clean precisely because it has no accountability architecture attached to it.

Jackson: Let me do the second one — the distribution of work blind spot. This is the one I think explains why the elimination impulse is so present among even thoughtful leaders. Every major technology wave before this one has been fundamentally about making humans faster at existing work. Organizations carry that model into AI, deploy the tool, make the humans faster, measure the efficiency gain, and nobody ever thinks about making the job different.

But AI allows organizations to restructure what tasks belong to machines, what require people, where the human is in the loop, where the human is above the loop. That's a fundamentally different capability than accelerating the existing workflow. When you apply the old mental model to the new kind of tool, you might see the efficiency gain — or you reach for cost reduction. The headcount elimination feels like the logical conclusion. The actual opportunity goes unexamined. And importantly, because the time horizon you're trying to solve for is too short and too precise, you haven't tried to solve for the larger piece: how do I best position my company to win over time — not just how do I do this thing now?

Scott: And the opportunity you're talking about right there is that freed human cycles can go somewhere really meaningful. More customer contact, more innovation, more of the strategic work the business has always wanted to do but said it didn't have capacity for. Those look like the same thing for about 18 months — and then they look really, really different.

The Constraint First Five Moves

Jackson: So then how does the CHRO actually change it? Let me frame it as the constraint-first framework — five moves. And the sequence here matters as much as the moves themselves.

First: identify the constraint. Before any vendor conversation, before any demo, before any budget request, facilitate the conversation with business leadership that answers one question: What is holding this business back right now? Not what would be useful, not where could we find efficiency — what constraint, if eliminated, would change the trajectory of the business? If you can get that question answered honestly, it should govern everything that follows.

Scott: And it sounds obvious until you try to do it in an organization where the vendor is already in and building, and the board is asking why you aren't moving faster on artificial intelligence. The discipline to stop, name the constraint, and insist that the answer governs which tool gets evaluated — that discipline is harder than it looks from the outside.

Jackson: It totally is, because the pressure from the board and from the CEO is real. But let's talk about the second move: vision the unconstrained state. What would this business look like if that constraint didn't exist? How would it operate differently? How would it advance? This step matters because it creates the measurement framework before the tool is ever selected. You know what you're solving for. If you know what the unconstrained state looks like, you can define success — and that allows you to build the real business case.

Scott: I found that to be the most clarifying question you can ask of your CEO — and probably the most clarifying conversation you can have with your CEO. I use that question a lot. In fact, you and I did an episode on whether to take the job as a CHRO — and part of my calculus on that has always been whether you can have that constraint conversation with the CEO in a meaningful way. Because most senior leaders aren't regularly asked to describe their business without its primary constraint. They're so deep into managing around it that the unconstrained state has never been made concrete for them. When you make it concrete, tool selection starts to get a lot more obvious. And the ROI conversation becomes a whole lot more straightforward.

Jackson: I go back to my time in the PepsiCo organization. We would figure out ways to have operational efficiency gains inside of our plants that no one else could ever even imagine. And that was the exact approach. What would an unconstrained world look like? And then: what are the constraints getting in the way? Then go figure out how to get the best minds in the organization focused on that, and deploy it across the company. That's how you ended up with quarter after quarter of productivity gains. When you do it well, this isn't theoretical. I've seen it done.

So let's talk about the third move: select the actual tool. I point this out because it's step three — not step one. Once the constraint is named and the unconstrained state is defined, the question becomes: what capability in the AI tool bag addresses that specific constraint? And here's where AI being a tool bag rather than a single thing also matters. Agentic systems, generative models, predictive systems, automation frameworks — each addresses different kinds of constraints. You need the constraint named before the tool can be selected if you want to do it smartly.

Scott: The CHRO who goes into a vendor conversation with well-defined constraints and says, "I don't want your demo — I want you to tell me how you're going to solve these constraint problems with the tool set that you have. And if you can't do it, that's fine" — that CHRO is having a fundamentally different conversation than the one who says, "Okay, I'm ready for your demo." A vendor who can't answer the question is just selling you a tool. A vendor who can solve your named constraint is selling you an outcome. But you're not going to find that if you just let them drive their demo.

Jackson: I do want to push back on you — there is a value to just going to Home Depot and buying a tool. I have consistently enjoyed buying tools. Unfortunately, my problem is I can't fix anything. So I will spend three times on the same project: once to buy all the tools and try it myself, once to pay someone to fix what I just broke, and a third time to get it right. But that does not take away the optimism and hope I feel every time I go in and buy a brand new tool for something I know is probably going to get screwed up.

Scott: I don't throw stones on fixing things because I'm less handy than you are. I think we'd better move to the fourth one.

Jackson: As I've gotten older, I've realized my superpower skill is writing a check for someone who knows what they're doing. Here's the fourth move: design the business case. Define how you will know if you've succeeded. What measurements will the business take that, if the constraint is truly eliminated, should show movement? Build the business case around that before you go into the deployment. That gives you early detection systems. Without it, you don't have any accountability architecture. You can't prove you moved something you didn't identify and measure before you started.

Scott: That's a really important point — and here's what most business cases are missing: what are you going to do with the freed capacity? When AI — or any technology — eliminates a constraint, human cycles become available. That's part of the return on the investment. You have to think about how you're going to redeploy those cycles before the deployment begins. It belongs inside the business case.

Jackson: Which is the fifth move: plan the redeployment before the tool goes live. You've got to figure out what those freed human cycles are going to do. What is the business going to do with capacity it didn't previously have? Think about it this way — there are things that if you had an extra day a week, you would focus your discretionary effort toward and have an outsized impact on the business. Figure out what those are up front. Whether it's more innovation, deeper customer relationships, or in the human resources area — more time building capability than managing compliance to a process — if you define those well up front, then you can redeploy it as part of the business case.

If you don't do that, the cost savings gets absorbed and next thing you know, you have a lower-cost but similarly capable organization. That's not bad — but you're not actually using this to deliver the growth that might be there. The ones that are actually getting it right are the ones that have a redeployment plan identified and aligned before it goes live.

Five Plays To Start This Week

Scott: Let me run those five moves one more time so everybody's got them front of mind. Constraint first. Unconstrained state for the business second. Tool selection third. Business case fourth. Redeployment plan fifth. Run those in order, and the ROI conversation has a real answer. You've built a measurement framework before you've made the investment. You know exactly what you're moving, exactly why, and exactly what the numbers should look like.

Jackson: Here is how the CHRO can move on this starting this week.

Play number one: run the constraint audit. Sit down with your CEO and your CFO — maybe some of your operators — and ask the question: if we could eliminate one structural constraint in the business in the next six to twelve months, what would it be? Don't bring a technology solution to that conversation. Bring the question, write down the answer, and use that answer as the starting point for every AI investment decision going forward.

Scott: Play number two: have the unconstrained state conversation. Take the constraint you just identified and go back to the CEO with a follow-up: what would the business look like if that constraint was gone? How would it operate? How would it advance? Get really specific. Revenue impact, speed to market, talent density in pivotal roles. Make it concrete enough to measure against. That picture is the architecture your business case is going to get built on.

Jackson: Play number three: reframe the next vendor conversation. Before your next vendor meeting, prepare three questions. What constraint does this tool eliminate — or at least relax? What does the business look like after that constraint is gone? And what's the measurement framework you would use to prove the constraint has been eliminated? A vendor who can't answer those questions is selling adoption — a solution to a problem that may not exist. A vendor who can answer them is selling outcomes. You want to be in a room with the second kind.

Scott: And I think you know, we've got to run those vendor interviews differently. Know your questions up front, connect them to those constraints, then run it like an interview. Would you let a candidate get away with a non-answer? You should not let the vendor get away with a non-answer either.

Play number four: design the redeployment plan before the tool goes live. Especially if you are investing in artificial intelligence in your current pipeline, you need a specific answer to one question before deployment: where are the freed human cycles going to go? Because they are, in fact, a part of the return on your investment. Which roles, which capabilities, which business outcomes will absorb the capacity this tool frees up? What more will the business be able to do, and where? If you can't really answer that, the investment isn't ready to move forward — because you don't know how to measure the return. The redeployment plan is what separates business outcomes from an efficiency number.

Jackson: I want to double-click on that because it's important. We're not presuming that the right answer is always redeployment. That is going to be a business-by-business, context-by-context situation. There may be times where the right answer is you take all of that savings to the bottom line. There may be times where you overinvest in areas you haven't been able to address. But the key is you do all that work up front and build it as part of the business case.

Play five: install a constraint review cadence. Add one question into every business planning cycle, every quarterly review: what's constraining the business right now, and has the answer changed? That question keeps tool selection honest and surfaces new AI applications as the business evolves. And that includes asking whether you've redeployed the freed capacity and whether you've gotten the benefit from that redeployment that you were expecting. If the CHRO owns that cadence, they own the AI accountability conversation for the entire enterprise. That's the position you want to be in.

Scott: And that fifth play has the longest tail. The constraint is changing as the business evolves. What's constraining the business this year is not what will be constraining it in two years. Most strategy frameworks get referenced at planning time and ignored the rest of the year — I laugh because I think I've been guilty of that. But a quarterly constraint review prevents that. The question stays live because it's a calendared conversation.

Jackson: Okay, senior leaders — this is your favorite part of the pod. It's your Talent Sherpa summary. Or as Scott always says, the best AI strategy is to buy everything that impressed you in the demo and then wait for the business to figure out what problem it solves.

Scott: No — please don't do that. I never said that, and I'm never going to say that, especially not around my own solution. All right, here we go, Jackson.

One: global AI investment is at historic highs, and ROI isn't following it. 95% of AI pilots are failing to deliver measurable P&L impact, and that gap isn't a technology problem — it's a sequencing problem. Organizations are selecting tools before they've named constraints.

Two: the assumptions that keep organizations stuck are predictable and self-reinforcing. Buying AI is doing AI. Adoption equals value. The business case follows the investment. Headcount reduction is the win. Each one makes sense in the room — and produces the same outcome: significant spend, no business result attached.

Three: AI is fundamentally different from prior technologies. It allows organizations to reimagine the distribution of work between humans and tools. Organizations applying the old mental model will cash a one-time efficiency check. The ones applying the new mental model are compounding a structural advantage by redeploying their human capital.

Four: the constraint-first framework. Identify the constraint. Vision the unconstrained state. Find a tool that helps you move from one to the next. Design the business case. Plan the redeployment of human capital. Every CHRO who runs those five moves in that order is having a fundamentally different AI conversation than most organizations are currently having. And that's the one that's going to produce the ROI.

Jackson: What I keep coming back to from this conversation, Scott, is the redeployment question. Most AI conversations end at the efficiency number. The right ones start with: what do you do with the capacity you free up? Redeployment is a strategic expansion question. Where does the free capacity go? What can the business do with it that it could not do before? The CHROs who are asking that question — they're shaping what their organizations are capable of becoming. That is the altitude this moment demands.

Thank you so much for tuning into the Talent Sherpa Podcast. This is where senior leaders come to rethink how human capital really works. This is so much fun to do with you all.

Scott: If you liked today's episode, please do Jackson and me a favor and hit the like button right now. Or better yet, subscribe to the podcast so that new episodes are delivered to you. Leave us a review on your favorite platform — Apple Podcasts, Spotify, YouTube — we're everywhere. It helps us reach more senior leaders.

Jackson: And I want to talk a little bit about Scott's company for a second. Propulsion AI is workforce intelligence for private equity. Their AI teammates are surfacing workforce risk before close and helping leadership teams drive execution after. They translate strategy into individual accountability, coach managers to define roles by outcomes, and give every employee a clear line of sight to what actually matters — all through artificial intelligence. You can learn more at the very cleverly named getpropulsion.ai.

Scott: And if you are a CHRO who is new to your role or preparing to step into a CHRO role, Jackson has built a suite of tools to help you operate at the altitude that role demands. Personal coaching, the Ascent Academy, his best-selling Substack — everything you need is at mytalentsherpa.com.

Jackson: Thanks, Scott. Thanks to everyone listening today. Until next time, keep raising the bar, please start with the constraints before you start with the tool, and keep on climbing.