Latest / Master Your AI Future | AI Executive / Why Is the AI Operating Model Becoming the New Competitive Advantage?
Transcript
- Welcome to the debate. Imagine spending, I don't know, fifty million dollars to roll out a revolutionary new technology across your entire enterprise. Right. A massive, massive investment. Exactly. And your employees, they love it. They are working twice as fast, and yet at the end of the year, your profit margins haven't moved a single inch. Not a single inch. That is the exact paradox facing the enterprise AI landscape today. We're seeing these blistering adoption rates, but when we look at the actual enterprise value, like the positive impact on earnings before interest and taxes, or EBIT, it is seriously lagging. It's just a massive disconnect. It really is the defining strategic gap of this current technological cycle, isn't it? I mean, we have AI adoption accelerating exponentially, but the organizational systems required to actually convert that speed into sustained measurable performance, they're breaking down. Yeah, they're completely failing in a lot of places. The underlying data from McKinsey tells us that eighty percent of respondents are seeing these huge individual productivity gains, right? But only thirty-seven percent report a positive EBIT impact, and a mere six percent actually qualify as AI high performers. The capability's obviously there, but the value is, well, it's leaking out of the system. Right, which is why the consensus across the source material makes it unmistakably clear that AI is no longer just an IT deployment issue. It is a fundamental operating model decision. So the core question we're tackling today is how that operating model needs to change to actually capture this value. Exactly. So I'm gonna argue for radical AI first business reinvention, that organizations must rapidly rebuild their end-to-end workflows around AI to seize competitive advantage, you know, before the window closes. And I will argue for a controlled evolution, that competitive advantage actually lies in institutionalizing governance, strict financial ROI thresholds, and human accountability long before scaling AI. Okay. Let me lay out why aggressive reinvention is really the only viable path here. Because if you just drop AI into your existing legacy workflows, you are setting yourself up for failure. Well, you're automating a silo. Exactly. That is exactly what creates that fifty million dollar paradox I just mentioned. Value only compounds when your workflows, your decision-making structures, and, uh, your incentives all change simultaneously. You cannot automate a silo. But changing all of that simultaneously is exactly what terrifies most leadership teams. Sure, but think about how this actually works mechanically, right? Based on the BCG and McKinsey findings. Instead of a human analyst taking like three weeks to gather data from five different departments, normalize it in a spreadsheet, and build a forecast- Right, the classic painful spreadsheet nightmare. Yeah. Instead of that, an AI orchestrator pulls the live data, runs ten thousand simulations, and presents optimized scenarios in an afternoon. That is how early orchestration compresses planning cycles by up to ten times. You can see first value in twelve weeks. In theory, yes. In practice too. But the fundamental reality is this: if your competitors can buy access to the exact same foundational AI models as you, generic access provides zero durable moat Your only real moat is a completely redesigned AI-first operating model. I hear that, but I look at this landscape, and I see aggressive reinvention as a fast track to, um, basically automating chaos. Automating chaos? How so? I mean, prioritizing speed over sustainable economics and risk management is incredibly dangerous. The strategic frameworks from Gartner and KPMG that we reviewed, they strongly emphasize a different sequence entirely. Technology deployment has to align with talent, process, and crucially, governance long before you attempt to scale. But they also emphasize moving quickly. Let me just finish this point because the survey data is pretty alarming. Look at the EY data. Only sixteen percent of surveyed CEOs have strong confidence in their ability to actually measure AI return on investment. Sixteen percent. Well, yeah, because they're using old metrics. But if you radically rebuild your business without formal financial measurement and embedded risks controls, things like traceability, decision logs, escalation pathways, you aren't creating a competitive moat. You are just masking hidden costs and amplifying operational fragility. Okay, but if you're listening to this as a department head, you might be thinking, "I don't have the time to build a perfect enterprise-wide financial tracking system before I start modernizing my workflows." And that's the reality on the ground. Trying to fit AI into legacy processes is like putting a jet engine on a horse-drawn carriage. A horse-drawn carriage? Yes. If we don't rebuild the chassis, our operating model, the speed will just tear the organization apart. Look at the BCGK study of the global bank. They projected thirty to fifty percent workflow automation and a hundred and fifty percent ROI over five years. Right. They aren't achieving that by just giving their analysts a chatbot to play with. They are completely redesigning workflows around agent-led execution. Maintaining all those traditional functional handoffs between departments in the name of, quote, unquote, "governance" simply destroys the value of AI orchestration. See, I look at the math you're pointing to, but you're totally ignoring the systemic friction that those handoffs represent. That global bank case study, it's a target projection. It is a business case. It's a highly vetted business case. But business cases in enterprise AI have a notorious habit of assuming a frictionless environment. Handoffs exist for a reason. They represent checks, balances, and domain expertise. Look at the KPMG data. Fifty-eight percent of leaders view enterprise-wide capabilities as critical, but only twelve percent deliver them effectively, and only twenty-four percent proactively integrate risk management. Right, which is an execution problem, not a strategy problem. It's both. Trying to drop AI into legacy workflows and instantly stripping away the governance is like upgrading your car's engine to go two hundred miles an hour, but keeping the brakes and steering from a nineteen ninety Honda Civic. The speed doesn't make you win the race. It just means you're gonna crash into the wall much faster. Okay, but the brake and steering analogy assumes that governance has to look like a nineteen ninety Honda Civic. I'm saying we build an entirely new chassis. But you have to actually build it. Right. And if we don't rebuild the entire operating model to handle that speed, yes, it will tear the organization apart. But you don't slow down the engine to match the old brakes. You upgrade the brakes. You rebuild the vehicle. Upgrading the brakes requires immense discipline, not just the mandate to move fast. Let's ground this in how it actually works in regulated industries, like the examples from the text. Think about credit decisioning in financial services or patient safety in healthcare. Right, high stakes environments. Exactly. In those environments, governance, decision logs, human escalation triggers, auditability isn't just an administrative hurdle you jump over, it is the operating model. If an AI agent hallucinates a reason for a credit rejection, or worse, misreads a patient's medical history, you cannot retroactively apply a control. You can't un-ring that bell once the damage is done. I completely agree you can't un-ring the bell, but the answer to hallucination or processing errors isn't to throttle the workflow redesign, it's to change who or what is executing the workflow. We have to redesign the human AI workforce alongside the process. But, uh, what does that actually look like in practice? Because it's one thing to say redesign the workforce on a podcast and another to manage thousands of employees whose daily tasks are suddenly vanishing. It looks like Accenture's view of a unified human AI workforce, treating AI as a digital employee rather than just a software tool. I mean, fifty-eight percent of C-suite leaders are already planning to fundamentally change how work is distributed between people and machines this year. Planning being the operative word. But they have to execute it. We have to redesign roles around outcomes, treating AI and humans as a single integrated productive system. Every AI agent must have a named human steward. A human steward. Okay. Yes. That is how you handle the governance you're so worried about in clinical support or credit decisioning. You don't slow down the AI. You explicitly map where the AI advises, where it executes independently, and crucially, where it hits a confidence threshold and escalates to its human owner. Right. The human isn't the bottleneck anymore. They are the reviewer and the final authority. Their day shifts from typing code or writing copy to managing a portfolio of AI agents. In theory, that sounds like a perfectly balanced ecosystem, but we have to confront the sheer organizational readiness required for that radical shift. Drawing a unified workforce chart on a whiteboard is easy. Operationalizing it across a global enterprise is incredibly difficult. It's difficult but necessary. But look at the EY survey of twelve hundred CEOs. It's flashing red on this issue. Forty-seven percent say AI skills aren't developing fast enough, and seventy-two percent expect skills shortages to be a major growth constraint. McKinsey found a massive readiness gap. Seventy percent personal readiness versus only twenty-seven percent organizational readiness. Yeah. You are asking for a system of human stewards when the workforce doesn't yet have the skills to actually steward these specific models. I have to admit, the data about the massive skill shortage is... Well, it's the one thing that gives me pause. The readiness gap is very real. But wait Isn't that exactly why we have to force the transition now? Force it? Yeah. How? Yeah. Okay. No, no, no, no. Like, if we, if we wait for a hundred percent organizational readiness, won't we lose the competitive window? So the friction of apprenticeship and learning is precisely why we must force the structural change today. And if we, if we don't force the organization to adapt its training models right now, right now, okay, by the time we feel, quote-unquote, ready, a competitor will have already scaled an AI-first model that operates at a fraction of our cost. But forcing structural change when the workforce isn't ready introduces massive operational fragility. Let's break down how expertise actually develops in a company. Sure. Historically, junior employees learn the business by doing the manual, repetitive execution tasks. They process the forms, they draft the initial reports, they do the basic coding. If you hand all of those execution tasks to an AI agent, how do your junior employees learn? How do they develop the necessary judgment, the gut instinct about what a good output looks like to eventually become those human stewards you're relying on? That's actually a fascinating mechanical problem, but the solution is that we change the apprenticeship model itself. We organize roles around outcomes rather than inherited task bundles. Okay, but how? Instead of a junior employee learning by spending five hours aggregating data, they learn by reviewing the AI's aggregation and verifying its accuracy against the source documents. They learn by editing, by auditing, and by managing the AI. They elevate the strategic thinking earlier in their career because they're forced to evaluate the output rather than just generate it. I have to push back there. Reviewing an output is fundamentally different cognitively than generating it. Is it, though? Absolutely. If a human steward doesn't deeply understand the underlying logic of the AI they are managing because they never wrestle with the raw data themselves, they cannot effectively govern it. They will just rubber-stamp the machine's output. That is why protecting human judgment and maintaining rigorous apprenticeship mechanisms is vital. You cannot just delegate early career tasks to an agent and assume the human overseer will catch the subtle errors. But they will catch the errors if their performance metrics are tied to the final outcome of that process. And honestly, this really bridges us to the ultimate stress test of this whole debate. All of this workforce restructuring, all of these workflow redesigns, it's just academic theory if it doesn't survive the CFO's desk. Exactly. The CFO conversation is where the rubber meets the road. It is where the hype cycle ends. Right. We have to look at the economic reality, and the reality is that the productivity gains we are seeing, those individual employees working faster and producing more, will inevitably lead to financial returns if the organization tracks them correctly. If. That's a big if. Well, the issue isn't that AI fails to create value. The issue is that companies lack the mechanisms to track that value all the way through to the P&L. The data confirms this. BCG found that companies tracking AI directly through the P&L realized three times the value, three point six percent versus one point two percent. If we redesign the operating model to release human capacity, that capacity doesn't just evaporate. It can be captured as hard cost savings or it can be deliberately reinvested for growth. I fundamentally reject the premise that productivity naturally equates to financial value. You completely reject it? Yes. It doesn't inevitably lead anywhere unless you capture it. And historically, large, complex organizations are terrible at capturing released capacity. You save an employee two hours a day, they don't hand you back a quarter of their salary. Well, no, obviously. They just fill that time with more emails, more meetings, or more administrative bloat. This is why we need a very strict definition of net AI value, like PwC outlines. You cannot just look at hours saved or system adoption rates. Those are purely activity metrics But they are early indicators of scale. You have to measure activity before you can measure impact. Activity metrics are like measuring a runner's heart rate and assuming they are winning the race. The heart is beating fast, they're sweating, the dashboard looks great, but they might be running in the entirely wrong direction. Okay, that's a clever analogy, but- You have to measure the net economic benefit. You must deduct the implementation costs, you must deduct the recurring operating costs, because inference isn't free. Actually, let's pause and clarify that for the listener, because inference is a huge part of the hidden cost structure. Yes, it is. Every single time an AI model generates a response or infers an answer based on a prompt, you are paying for computing power. You are paying API costs, cloud hosting fees, and energy costs. Add to that the cost of continuous model maintenance, data pipelines, and the transition friction of retraining your staff. All of that must be deducted before you can claim any economic victory. The warnings from EY are explicit. Productivity gains are often entirely absorbed by organizational complexity before they ever reach the bottom line. I hear the concern about hidden costs, I really do, but the cost per outcome model directly solves this. Think about it mechanistically. If we establish a financial baseline for priority value stream, let's say the total blended cost to process an insurance claim from the moment it's filed to the moment it's paid, we bypass the activity metric trap entirely. How so? Because we aren't measuring hours saved by the claims adjuster anymore. We are measuring the definitive reduction in the cost to achieve that specific outcome, factoring in the API costs, the computing power, and the human oversight. If, and it is a massive if, you have the operating discipline to measure it that way. The data showed an overwhelming majority of CEOs are essentially flying blind on AI ROI. They don't have real-time visibility into costs and returns at that granular level. Pushing them to radically reinvent their core workflows when they can't even accurately measure the cost of their current workflows is incredibly reckless. But you're- This is why you must establish the financial measurement and the governance structure first. Prove the business case on a small number of high-value workflows, establish the baseline economics, and demand strict P&L accountability before you unleash this technology across the entire enterprise. You are demanding perfect visibility into a new paradigm using the accounting tools of the old paradigm. If you concentrate your capital on areas capable of creating truly differentiated value, you just have to accept a period of transition friction. Transition friction is a nice way of saying burning cash. No. The most advanced future-built organizations aren't waiting for perfect ROI visibility on every single micro process. They are committing to a common enterprise AI platform, a central control plane that manages security, access, and model deployment, and they are building the architecture for governed execution at scale. They are moving aggressively because the opportunity cost of waiting is just too high. But building that architecture is exactly what I am advocating for. Architecture includes the brakes, the steering, and the dashboard. When the consensus points to governed execution, the operative word is governed. Right. The strongest performers embed governance, trust, and accountability directly into the workflow from day one. They don't treat risk management as a separate compliance activity that happens after the fact. My argument is that this embedded governance is the actual competitive advantage. As these models commoditize, your ability to safely and reliably operate AI at scale without catastrophic regulatory breaches or runaway cloud costs is what will separate the winners from the losers. I see the logic there, I really do. But I still maintain that governance is the necessary foundation, not the moat itself. The moat is the completely reimagined business model. We'll have to agree to disagree on which is the cart and which is the horse. Fair enough. If we pull all these threads together, my position remains that while governance and strict measurement are undoubtedly important, the most critical imperative for leadership right now is boldness. The transition to an AI first operating model requires fundamental business redesign. We have to concentrate capital on differentiated value pools and build hard to replicate advantages. If we just optimize legacy processes, we're wasting the technology's potential. We must be willing to tear down the old workflows and construct a unified human AI workforce focused on entirely new ways of delivering value. And to summarize my stance, I conclude that lasting competitive advantage is built on discipline, not just capability. Boldness without structural controls is simply speed toward failure. By prioritizing governance as a core operating capability and by demanding strict P&L accountability, calculating true net AI value rather than getting distracted by how fast the runner's heart is beating, organizations protect their core. You don't have to sacrifice speed entirely, but you must ensure that your institutional knowledge, your risk frameworks, and your human judgment scale proportionately with your automation Well, I think where we definitely converge today is on the realization that generic access to AI models offers absolutely no durable differentiation. Totally agree. Every enterprise on earth will have the same foundational models. Exactly. The true battleground is how the work gets done inside the company. The operating model is the strategy now. There's just no separating the two, and we also have to acknowledge that the tension we've debated today, you know, between aggressive speed and strict control, is gonna vary wildly depending on the industry. Oh, absolutely. A retail company optimizing supply chain logistics can tolerate a much higher risk appetite for rapid reinvention than, say, a hospital network automating clinical support. The complexity of enterprise AI transformation is immense, and there is so much more nuance in the source material for leaders to explore as they assess their own organization's readiness. It really forces us to rethink what an organization actually is and how humans and machines collaborate to create value. It's a massive shift. It is. We'll leave it to you to determine the best path forward for your own operating model. Until next time, keep questioning the consensus.