Latest / Master Your AI Future | AI Executive / Why Is AI Investment Still Failing to Reach the P&L?
Transcript
- 📍 📍 📍 📍 📍 Welcome to the debate. So I want you to imagine that you just spent, uh, let's say, five million dollars on a massive rollout of enterprise AI licenses. Right. A very real scenario for a lot of organizations right now. Oh, entirely real. And, you know, your employees are absolutely thrilled. They are clearing their inboxes faster. They're, uh, drafting reports in seconds. Honestly, they're probably leaving the office an hour early every single day. Yeah, they love it. Exactly. But then you walk into your CFO's office at the end of the quarter, and the profit margin, it hasn't moved a single inch. It really is the defining corporate paradox of our time, isn't it? I mean, we are looking at recent research from six of the leading consulting firms. So McKinsey, Accenture, PwC, Gartner, KPMG, and Forrester. The heavy hitters. All of them. And they are all pointing at the exact same flashing red light on the dashboard. Enterprise AI spending is rising exponentially, faster than verified financial returns. The executive commitment is, you know, it's clearly there, the adoption is happening, but those localized productivity gains, they are completely failing to translate into measurable PNL impact. Profit and loss is just... It's static. Which is precisely the crisis we are unpacking today, because if you are a department head or, uh, a strategic leader listening to this, you are probably feeling this exact pressure from your board. So the core disagreement we have to hash out is this: to actually solve this AI ROI crisis, should organizations prioritize end-to-end business model transformation and growth, or on the flip side, must they first enforce strict process-level efficiency, cost visibility, and governance? Right. We are looking at two very, very different philosophies for how to survive this technological transition. Exactly. So my stance today is that the primary path to realizing AI value is through structural operating model redesign, heavily focused on revenue growth and proprietary workflows because, well, because generic efficiency gains will never create a durable competitive mode. And I will counter that this crisis fundamentally stems from a lack of financial discipline. I'll be arguing that organizations absolutely must prioritize what we call capacity conversion, you know, turning localized operational efficiency and time savings into hard process-level economics long before they chase transformational growth. Okay, let's start with the fundamental flaw in how companies are deploying this technology right now. I mean, we have to clearly articulate that AI value realization simply cannot happen by squeezing a highly capable probabilistic technology into old, rigid legacy processes. You're talking about just bolting it on. Right, just bolting it on. McKinsey and Company takes a very firm stance here. They note that leading organizations don't just automate existing tasks, they completely redesign how work is performed and, uh, how decisions are made end to end. Sure, but redesigning how work is performed, I mean, that is a massive capital intensive undertaking. It is. It absolutely is. But look at the cost of not doing it. PwC just released findings showing that the top twenty percent of surveyed companies captured a staggering seventy-four percent of all AI-driven returns. Wow, seventy-four percent? Yeah. It's a massive concentration of wealth. Those organizations generated financial performance seven point two times higher than their peers. And, you know, they didn't achieve that by using AI to trim, say, ten percent off their travel and expense processing. No, they wouldn't. Right. They achieved it because they focused on growth, they built out proprietary workflows, and they created highly differentiated customer experiences. Mere cost reduction just doesn't get you those numbers. I mean, I come at it from a different way. The data on that concentration of returns is stark, yes, but your interpretation sort of skips over the operational reality of the other eighty percent of companies out there. How so? Well, when I look at the lack of PNL impact across the broader market, I see a fundamental failure of basic cost governance and workforce management, not necessarily a lack of transformational ambition. So you think they have the ambition, just not the discipline? Exactly. Look at Gartner's data. Right now, eighty-four percent of finance AI spending is directed at individual productivity. The problem isn't that the technology isn't working. Uh, the problem is that those financial gains are entirely trapped. Trapped in what way, though? Trapped in the daily routine. To understand why, you really have to look at KPMG's crucial economic control metric. They found that organizations with full visibility into their AI operating costs, meaning, you know, they meticulously track model usage, crowd consumption, human oversight, those organizations are five times more likely to report established ROI. Five times? Five times. We're talking fifteen percent versus three percent. So if you don't have that foundational financial discipline, I mean, no amount of transformational ambition is gonna save your margins. Okay, I think we need to ground this in the reality of the listener because we have these overarching philosophies, right? But what actually happens on the floor of a business when an employee saves two hours a day using a new AI tool? Yes. This is exactly where the rubber meets the road, and it brings us to this core concept of capacity conversion. I'd argue this is probably the most vital management concept in the AI era. Go ahead and define that for us. It is the formal process of taking the employee time that AI has saved and actively turning it into a defined business outcome. Right. Because if you don't actively convert it, it just... It evaporates. Precisely. It vanishes. PWC finds that seventy-seven percent of respondents say their AI investments aren't delivering measurable ROI, and the mechanics behind that failure are actually quite simple. Management observes the productivity, they see people working faster, but they completely fail to decide what happens to the released capacity. Let's break down the mechanics of that. Give me a scenario. Sure. Say you have a team of ten financial analysts, right? And they use AI to save ten percent of their weekly hours. On paper, you have just saved the equivalent of one full-time employee. Makes sense. But unless management makes a hard decision to either, uh, reduce headcount, remove that cost from the budget, or explicitly direct that team to absorb ten percent more workload without hiring anyone new, that time saved remains an entirely theoretical benefit. You just have a happier team of analysts. Exactly. You have a happier team who might, I don't know, browse the internet from four PM to five PM. Your P&L hasn't changed by a single cent. The failure isn't the AI model hallucinating. The failure is management lacking the discipline to execute capacity conversion. I see why you think that, but let me give you a different perspective. Focusing hyper-vigilantly on extracting those saved hours is completely missing the structural flaw of the business itself. You're essentially treating AI like a slightly faster typewriter. I'm treating it like an enterprise investment that needs to pay for itself. But capacity conversion, at least in the way you're describing it, is just a Band-Aid on a broken system Let's use an analogy here. Trying to capture time savings in an unchanged workflow is literally like putting a race car engine in a horse-drawn carriage. Okay, that's a bit dramatic. No, think about it. You might get somewhere a fraction of a second faster, and sure, you can measure that little burst of efficiency, but you haven't changed the nature of the vehicle. You've completely wasted the engine's capability. True value requires fundamentally redesigning the workflow around the desired outcome rather than just the existing sequence of tasks. But redesigning a workflow from scratch is incredibly disruptive to the business. It has to be disruptive. Look, if you just take a terrible, bloated ten-step approval process, and you use AI to automate steps three, four, and five so they happen instantly, you still have a terrible ten-step process. But it's cheaper. Marginally, but you haven't fundamentally improved the business. McKinsey's point is that you have to change decision rights, accountability, the very flow of data. You might realize that with a true AI integration, steps two through nine shouldn't even exist anymore. I understand the appeal of the race car analogy, I really do, but you can't build a racetrack if you can't even pay to feed the horses. End-to-end redesign requires massive institutional effort. It requires capital. Of course. So if you aren't capturing those localized efficiencies first, if you aren't doing the unglamorous work of capacity conversion to actually remove costs from your current budget, where are you getting the runway to fund this grand transformation? You are asking executives to fund a structural redesign based on pure faith, all while they are bleeding cash on cloud consumption and model licensing fees. It's not faith, though. It's a completely different framework of evaluation. And honestly, that forces us to look at how executives should actually measure and fund these initiatives because if you measure a transformation initiative with a localized efficiency metric, it will fail every single time. So how do you measure it? I champion Gartner's concept of AI portfolio economics. We need to evaluate AI investments as entirely different classes of assets. Much like a financial portfolio, you wouldn't evaluate a volatile tech stock using the exact same metrics you use for, say, a municipal bond. Right, but- So you cannot apply a single universal ROI formula to every single AI use case. What does that actually look like in practice, though? It means segmenting your initiatives. Process automation, the daily task level stuff you're talking about, sure, assess that by unit cost. But transformational investments, the ones that actually invent a new product or change the business model, those need staged milestones. They need funding gates. Forrester agrees with this explicitly. They point out that conventional measurement frameworks often fail completely to capture AI's mix of operational, customer, and strategic effects. I'm sorry, but I just don't buy that. Let me tell you why. Go ahead. The term portfolio economics can incredibly easily become a sophisticated loophole for avoiding accountability. It's not about avoiding accountability. It's about capturing long-term value. But think about how this plays out in a boardroom in real life. When a department head says, "Conventional measurement fails," what they are usually saying is, "We haven't made any money yet, so let's change the definition of money." We have to demand a focus on process-level value metrics. Even for long-term strategic plays? Especially for them, because otherwise the spending goes totally unchecked. Take financial services, for example. In lending or, uh, insurance, deploying an AI tool only matters if it improves a hard mechanical metric. Such as? Let's talk about straight-through processing rates. For the listener, that means a customer submits a claim or a loan application, and the system processes, approves, and executes it without a single human ever having to click a mouse. That is a hard metric. It's either straight through or it's not. What is the cost per claim? What is the exception rate? But those are purely efficiency metrics. Because efficiency pays the bills. KPMG's research is very telling here. They found that seventy-six percent of business leaders report seeing real business value from their AI. Which is a good thing. But real business value is just a warm, qualitative feeling. Establishing a formal finance-validated ROI based on hard metrics remains incredibly rare. If you abandon strict financial governance for this, uh, portfolio approach, you end up with an expanding roster of very expensive science experiments. They look fantastic in a slide deck, but they absolutely drag down the operating margin. That's an interesting point, though. I would definitely frame it differently. I'm not advocating for abandoning financial validation. I'm advocating for contextual validation. Contextual validation sounds a lot like moving the goalposts. Not at all. Look at wealth management. If you design an AI initiative to create an entirely new scalable advisory service, something that allows you to offer personalized financial advice to middle income clients at a massive scale, that is a growth play. Right. If you use a rigid process-level metric like cost per existing claim to measure that new service, the initiative will fail its first review. You'll kill the innovation before it even matures. But you have to prove the tool works efficiently before you trust it to generate revenue. Gartner isn't saying don't measure. They are saying segment the portfolio: productivity, growth, transformation, and strategic options. If you only focus on localized, strictly governed cost-cutting, you're ignoring the ultimate endgame of these investments. The whole point of AI isn't to just be a cheaper company. It's to create a long-term competitive advantage. That's a compelling argument, but have you considered that competitive advantage doesn't exist in a vacuum? It absolutely requires a foundation of operational discipline. See, I'm not convinced by that line of reasoning, mostly because efficiency gains will quickly diffuse across the entire market. AI tools are rapidly becoming broadly accessible commodities. They are accessible, sure. Right. So let's say your entire strategy is using an off-the-shelf AI vendor to reduce your underwriting turnaround time by ten percent. You declare a massive victory for process-level economics. But guess what? Your competitors will buy the exact same software from the exact same vendor and achieve the exact same ten percent reduction within six months. True. Your competitive moat instantly disappears. But you've still protected your margins in the near term. But you haven't grown. A durable moat requires using AI for new revenue models, faster onboarding, better customer experiences. These are capabilities that competitors cannot easily reproduce because they are deeply tied to your proprietary workflows and your unique enterprise data. That assumes you have the talent to build those proprietary workflows in the first place. Which brings up the human capital aspect, and this is critical. PWC found that workers with specialized AI skills currently earn an average wage premium of fifty-six percent. Yeah, that is a staggering premium for talent right now. Exactly. So let's unpack the mechanics of that. If you are paying a senior developer or an AI architect a fifty-six percent premium over market rate, you absolutely cannot waste that time optimizing a back office invoicing process just to save a few dollars. The math doesn't work. I see what you mean. The incredibly high cost of talent means organizations must direct AI toward higher value growth work. You need them building products that bring in new revenue, not just localized cost cutting. The economics of talent literally demand a focus on growth. I understand the logic there, I really do, but you cannot scale a broken system. You talk about the rapid diffusion of efficiency tools, but the data suggests executives are still struggling just to get that baseline right. You can't run before you walk. Why must it be sequential, though? Why can't we redesign for growth from day one? Because attributing growth to AI is incredibly noisy and financially risky. Think about it. If you launch a new AI-enabled customer proposition and your revenue goes up 5% this quarter, was that actually the AI? It could be. Or was it external macroeconomic conditions? Did a competitor raise their prices, driving customers to you? Did you just run a really successful marketing campaign? Growth attribution is incredibly muddy. But that's always been true for any business initiative, hasn't it? Yes, but with AI, the upfront capital cost is so much higher. Efficiency attribution, on the other hand, is clean. It is undeniably clean. If I implement an AI tool in my finance operations and my quarterly close cycle duration drops by two days, and I execute capacity conversion to explicitly remove three contractor headcounts from the budget, I have an isolated finance-validated return. Right. It's easy to put on a spreadsheet. Exactly. There is no ambiguity. I have proven the governance, I have proven the integration, and I have proven the human oversight. But you've wasted valuable time proving something small. I've built institutional muscle memory. Nearly 60% of CFOs plan to increase finance AI investment by at least 10% this year, and Gartner notes they are doing this specifically with efficiency as a dominant objective. Which I'd argue is a mistake. I think it's the most defensible executive approach. It's sequential. Once you have established process economics, once you know how to govern a model and control cloud costs, then you can scale into growth where the variables are much harder to control. Jumping straight to structural transformation without cost visibility is exactly how companies end up with massive AI budgets and zero bottom-line impact. The problem with the sequential approach, though, is the timeline. The competitive window is narrowing rapidly. If you are a business leader and you spend the next two years meticulously proving unit economics on mundane back office tasks, the top twenty percent of organizations, the ones PwC identified as capturing seventy-four percent of the returns, they will have already rewritten the rules of your industry. They might also have burned through millions of dollars in failed experiments, to be fair. Maybe some of them, but they are embedding AI into their core customer-facing products. They are capturing market share. The sequential approach feels very safe to a conservative CFO, but it poses a massive strategic risk to the survival of the business. We must move beyond treating AI as an incredibly advanced calculator that just trims margins. And yet, if you treat AI as a magic wand for business transformation while entirely ignoring the difficult, unglamorous work of cost governance, you end up exactly where our underlying research says the market is today: high executive commitment, rising spending, and totally trapped value. The potential of the technology depends heavily on cost transparency, integration, and process ownership, not just model capability or grand strategic ambition. Let's bring this together and summarize where we stand on this crisis, because the stakes for anyone leading an organization right now couldn't be higher Definitely. To avoid carrying an expanding portfolio of AI activity without any economic evidence, I maintain that organizations must fundamentally embrace operating model redesign. They need to stop looking at task-level efficiency and focus on proprietary data and workflows. They must prioritize growth, because squeezing new tech into unchanged processes simply will not yield the outsized returns that the top twenty percent of organizations are currently enjoying. And my summary is that the foundation of any sustainable AI value realization must rest on cost visibility, strict governance, and the disciplined execution of capacity conversion. Time saved is absolutely not money earned until leadership makes a hard decision to remove cost, absorb new volume, or strategically redirect that capacity. That fundamental discipline must happen at the process level before an organization attempts structural transformation. Now, despite our different lenses, we do have significant points of convergence here, which I think is incredibly useful for the listener. We both explicitly agree that basic AI adoption, whether we are talking about prompt volume, logged-in users, or just the sheer number of deployed licenses, is merely a leading indicator. Right. It absolutely does not equal financial value without deliberate, forceful executive intervention. Precisely. Activity is not ROI. And we also completely agree that finance must validate the returns. Whether a company uses your portfolio economics framework or my strict process-level metrics, the era of relying on qualitative feelings of real business value is officially over. The results must be mathematically defensible. It really highlights the sheer complexity of institutionalizing value realization right now. The competitive window is narrowing, and every organization is feeling the pressure from their stakeholders. The tension we've discussed today between securing near-term process efficiency and pursuing long-term transformational ROI is something that requires careful navigation in every single strategic planning session. It is a delicate balance, and there are profound risks on both sides of the equation. We leave it to the listener to look at their own balance sheets, their own talent pools, and their own market position to decide which imperative is more critical for their organization at this moment in time. We started today by talking about a CFO staring at a profit margin that hasn't moved despite a massive technology investment. The challenge now is writing the new laws of economic physics for your own operating model before your competitors do it for you. Thank you for joining us.