Why Enterprise AI Budgets Are Flowing to the Wrong Place — Lessons from EY's Global AI Leader

TLDR: Dan Diasio, EY's Global AI Consulting Leader, argues that most enterprise AI budgets are misallocated — too much goes to technology, not enough to mindset and skills. The organizations actually generating enterprise-level AI value are inverting that ratio, redesigning operating models rather than automating existing ones, and treating workforce reinvention as the primary ROI lever.

Best For: COOs, CIOs, and enterprise operations leaders in traditional industries who are frustrated that AI pilots aren't translating to measurable business value — and want a clearer diagnosis of why.

In March 2026, Dan Diasio — EY's Global AI Consulting Leader and Americas Consulting CTO — sat down with Emerj's Matthew DeMello for an episode of the AI in Business Podcast that cut against much of what passes for AI strategy advice today. The message was pointed: most organizations are spending their AI budgets in exactly the wrong places, and the gap between AI experimentation and AI value isn't a technology problem. It's a leadership and organizational design problem.

This article unpacks Diasio's framework — mindset, skill set, tool set — and what it actually means for enterprise operations leaders trying to move from proof-of-concept to production-grade AI.

The Numbers Behind the Stall

Before getting to Diasio's diagnosis, it helps to understand how widespread the problem actually is.

The U.S. Census Bureau's Business Trends and Outlook Survey tracked overall AI usage among American employer businesses between December 2025 and May 2026. The number hovered between 17% and 20% — a figure that has barely budged despite years of AI hype, accelerating model capability, and record levels of enterprise AI investment. Another 20-23% of businesses reported expecting to use AI in the next six months, a pace of stated intent that has consistently outrun actual deployment for three years running.

The Stanford HAI 2026 AI Index puts a finer point on it: while 70% of organizations now report using generative AI in at least one business function, AI agent deployment — the kind that actually automates work rather than assisting with it — remained in the single digits across nearly all business functions. And despite commanding the largest share of global AI investment, the United States ranks just 24th globally in actual AI adoption, at 28.3%.

The Census Bureau's own research found that 57% of adopting firms integrate AI into three or fewer business functions, concentrated in sales, marketing, and strategy. The operational core — the processes where most enterprise value is actually generated — remains largely untouched.

What these numbers collectively describe isn't a technology gap. They describe an execution gap: the distance between what enterprises can do with AI and what they're actually doing.

Diasio's argument is that this gap exists because organizations have been solving the wrong problem.

The Toolset Trap

Here's the diagnostic that runs through Diasio's analysis: enterprises are massively overinvesting in tools relative to the organizational capabilities needed to use those tools well.

"Most of the money today is going towards the toolset," Diasio said in the Emerj episode. "That actually needs to flip — the money needs to be going more towards the mindset and the skill set."

This is worth sitting with for a moment, because it runs counter to what most enterprise AI spending looks like in practice. The typical AI investment portfolio is heavy on platform licenses, infrastructure, model fine-tuning, and vendor integrations. The assumption is that the right technology stack will unlock value on its own — that you buy the capability and the organization will adapt around it.

Diasio's position is that this gets causality backwards. Tools are only as valuable as the people using them and the workflows they're applied to. When mindset and skills lag the toolset, you end up with AI-enabled versions of the same broken processes — faster, cheaper, and still producing the wrong outcomes.

The classic symptom is what Diasio calls the "visibility trap": executives see impressive demos, pilot programs generate clean metrics, and the organization appears to be making progress. But the pilots don't scale, the ROI doesn't compound, and the underlying business model never actually changes. The tools are doing things; they're just not doing the things that matter.

Diagnosing whether your organization is in the visibility trap is straightforward: ask whether your AI initiatives are producing measurable changes in the business outcomes that drive revenue or competitive position. If the answer is "we're tracking adoption rates and usage metrics," you're probably in the trap.

What Mindset Actually Means in This Context

The word "mindset" gets thrown around loosely in AI transformation discussions, often as a placeholder for "we need people to be more open to change." That's not what Diasio means.

In the framework he describes, mindset refers to something specific: whether leadership is asking AI to automate existing processes or redesign the business around new capabilities. These are fundamentally different mandates, and they produce fundamentally different outcomes.

The automation mindset says: here is a workflow we have today; find the parts that AI can make faster or cheaper. The redesign mindset asks: given what AI can now do, what operating model would we build if we were starting from scratch?

The distinction matters because most enterprise processes were designed for a pre-AI world — for human decision speeds, human error rates, and human bandwidth constraints. Automating those processes with AI is better than not using AI, but it doesn't unlock the step-change improvements that AI-native operating models can achieve. You end up with a faster horse rather than a car.

Diasio's observation, reflected in EY's work with large enterprises across sectors, is that the organizations generating real AI ROI have made this mental shift at the leadership level. They're not asking "how do we use AI to do what we do better?" They're asking "what would a company purpose-built for this AI moment look like, and how do we become that?"

This is a harder question to sit with, because the answer often involves dismantling things that work reasonably well today in order to build something that performs at a different level tomorrow. That's a leadership challenge before it's a technology challenge.

Skill Set: The Gap That Compounds

Even when leadership gets the mindset right, the skill set problem can still kill the transformation.

Diasio's framework makes a distinction that's easy to underestimate: skills aren't just about knowing how to use AI tools. They're about being able to make good decisions in AI-augmented workflows.

Enterprise AI transforms the nature of work in ways that require new judgment capabilities, not just new technical ones. When an AI system is generating recommendations, drafting content, flagging anomalies, or routing decisions, the human in the loop needs to know when to trust the output, when to override it, how to catch the failure modes that matter, and how to apply domain expertise to contexts where the AI is operating at the edge of its competence.

None of those skills come automatically with AI adoption. They have to be built deliberately — through training, through workflow design that creates practice loops, and through organizational structures that reward the right behaviors rather than just raw AI utilization.

The Emerj research underlying the HTEC-sponsored article "The Conditions That Turn AI Pilots Into Enterprise Value" (published July 2026) puts a related point sharply: pilot programs often appear to work because they're run by experts who already have the domain knowledge to catch AI errors and compensate for system limitations. When those workflows get rolled out to the broader organization, the error-correction capacity disappears, and the failure modes that the experts were quietly managing become visible.

This is why an AI readiness assessment that evaluates workforce capability — not just technology infrastructure — is a prerequisite for any serious AI transformation effort. The toolset question is straightforward; the skill set question is where most enterprise AI programs actually succeed or fail.

For traditional industries specifically — manufacturing, logistics, financial services, healthcare — the skill set challenge has an additional dimension. The domain knowledge that makes AI outputs useful in those contexts is often tacit, distributed across experienced workers who've spent years developing judgment that doesn't live in any documentation. Capturing that knowledge, encoding it into AI systems, and then training the next generation of workers to operate in an AI-augmented environment requires a deliberate program, not a tool deployment.

Workforce Reinvention vs. Headcount Reduction

One of the sharper points in Diasio's Emerj conversation is his argument that leading organizations are reallocating AI gains toward workforce reinvention rather than headcount reduction. This is worth examining carefully, because it represents a different theory of AI value than the one that dominates most enterprise AI business cases.

The headcount reduction argument is intuitive: if AI can do what five people do, you need fewer people, and that's your ROI. It's a simple calculation, it shows up cleanly in financial models, and it gives executives a number they can defend to boards and investors.

The problem is that it's also a ceiling. If you use AI to eliminate five roles, you've captured the efficiency value of AI, but you've stopped there. You haven't asked what those five people could be doing if they weren't doing the work that AI is now handling — whether there's a higher-value activity that the business couldn't previously afford because the workforce was fully allocated to operations.

Diasio's observation is that the organizations with the best AI ROI trajectories are answering a different question: what does this organization become capable of doing at scale that it couldn't do before? The answer often involves expanding into new markets, improving service levels in ways that drive customer retention, accelerating product development cycles, or building intelligence capabilities that create durable competitive advantages.

This framing changes where you invest the productivity gains. Instead of reducing headcount, you redeploy capacity toward growth activities. Instead of treating AI as a cost reduction tool, you treat it as a capability expansion tool. And the financial case, while harder to model upfront, tends to be substantially larger over a three-to-five year horizon than pure efficiency plays.

For measuring that kind of AI ROI, you need frameworks that go beyond tool adoption rates and headcount metrics — which is exactly why traditional AI ROI measurement approaches tend to understate the value of transformative deployments.

Operating Model Redesign: What It Looks Like in Practice

The phrase "operating model redesign" risks sounding abstract. What does it actually mean to redesign an enterprise operating model around AI capabilities?

Diasio's framework points to a few concrete dimensions:

Decision architecture: In most traditional enterprises, decisions flow through hierarchies designed to manage information scarcity. Humans aggregate information upward, analysis happens at middle layers, and decisions come back down. AI changes the information economics fundamentally — the constraint is no longer information availability but decision quality and speed. Operating models designed for information scarcity need to be rebuilt around information abundance and fast-cycle decision-making.

Workflow sequencing: When AI can handle certain tasks autonomously, the sequencing of human and machine activities changes. Work that used to happen linearly — gather data, analyze, decide, act — can be restructured to run in parallel or to telescope time horizons. Organizations that redesign workflows to exploit these properties see dramatically different throughput and quality outcomes than those that simply insert AI into existing sequences.

Accountability structures: When AI is embedded in decision workflows, the question of who is responsible for outcomes becomes more complex. Operating model redesign requires making those accountability structures explicit — which decisions are AI-led with human review, which are human-led with AI input, and how the organization escalates when AI outputs are uncertain or contested.

Talent and role definition: This connects to the skill set dimension. An AI-native operating model has different roles than a traditional one — fewer people doing routine analytical work, more people doing AI supervision, workflow design, exception handling, and strategy. Getting to that operating model requires knowing what roles you're designing toward and actively building or acquiring those capabilities.

Understanding where your enterprise sits on the AI maturity curve is a prerequisite for operating model redesign — you need to know whether you're at the stage where workflow redesign is the right next move, or whether foundational infrastructure work still needs to happen first.

Why Traditional Industries Have a Timing Advantage

Here's a counterintuitive point that often gets missed in AI transformation discussions: enterprises in traditional industries — the ones typically described as "behind" on AI — may actually have a timing advantage right now.

The organizations that deployed AI earliest, in 2022 and 2023, often built on technology that has since been substantially superseded. They have legacy AI infrastructure, technical debt from early proof-of-concept work that got scaled prematurely, and change management challenges created by repeated failed rollouts. Their employees have already developed skepticism about AI initiatives, which makes cultural transformation harder, not easier.

Traditional industries that are approaching AI adoption seriously now, in 2026, have access to dramatically more capable models, more mature deployment infrastructure, better organizational playbooks, and a richer body of failure case studies to learn from. They're not starting from scratch — they're starting with much better raw materials.

The catch is that this advantage is time-limited. AI-native competitors and early enterprise adopters who have worked through their early deployment failures are now building compound advantages. The window for traditional industries to close the gap without starting at a structural disadvantage is real, but it's not open indefinitely.

Diasio's framework suggests the fastest path through that window is exactly what most traditional enterprise AI programs get wrong: invest in mindset and skills first, redesign the operating model before scaling the toolset, and target workforce reinvention rather than headcount reduction as the primary value capture mechanism.

Deploying AI agents in enterprise operations is one of the highest-leverage moves available to traditional industry leaders right now — but only if the organizational foundation is in place to make it work.

What This Means for Enterprise Operations Leaders

If you're a COO or head of enterprise operations in a traditional industry, Diasio's framework suggests a fairly specific set of priorities:

Audit your investment allocation. What percentage of your AI budget is going to technology versus to organizational capability building — training, change management, workflow redesign, process documentation? If the answer is heavily skewed toward technology, that's the first thing to rebalance.

Define the business outcome before the tool. Every AI initiative should have a clear answer to "what business metric are we trying to move, by how much, and by when?" without which the ROI calculation can't be validated and the workflow design can't be done correctly.

Identify your highest-value processes for redesign. Not every process benefits equally from AI-native redesign. The highest-value targets are typically the ones where decision quality and speed are the binding constraints, where information asymmetries create significant costs, or where human bandwidth limitations are preventing the business from capturing market opportunities.

Build the skill set before you scale the toolset. Pilot programs work because they're staffed with capable people. Before scaling, make sure the organizational skills are in place to sustain quality at scale — particularly the judgment skills needed to supervise AI outputs and catch failure modes.

Target capability expansion, not just efficiency. Use the productivity gains from AI to enable things the business couldn't previously do at scale, rather than just reducing the cost of doing things you already do. This shifts the ROI trajectory from linear cost savings to compound capability growth.

Frequently Asked Questions

What is the mindset-skillset-toolset framework from EY?

EY's Dan Diasio describes three components that must be aligned for enterprise AI to deliver value: mindset (whether leadership is redesigning the business or just automating existing processes), skill set (whether the workforce can make good decisions in AI-augmented workflows), and tool set (the AI technology itself). Diasio's key insight is that most enterprise AI budgets over-invest in toolset and under-invest in mindset and skills — an allocation that needs to flip for AI to generate enterprise-level ROI.

Why aren't enterprise AI pilots converting to full-scale deployments?

Pilots typically succeed because they're run by domain experts who have the judgment to compensate for AI limitations. When scaled to the broader workforce, those error-correction capabilities disappear. The fix isn't better technology — it's building the skills and workflow design that allow non-experts to operate AI-augmented workflows safely and effectively. This is what EY's Dan Diasio means by "organizational readiness" as a prerequisite for scale.

What does "workforce reinvention" mean in the context of AI transformation?

Workforce reinvention refers to reallocating the productivity gains from AI toward higher-value activities rather than reducing headcount. Instead of using AI to do the same work with fewer people, leading enterprises use AI to enable their workforce to do work that wasn't previously possible at scale — entering new markets, improving service levels, accelerating development cycles, or building intelligence capabilities that create durable competitive advantage.

What is the "visibility trap" in enterprise AI?

The visibility trap is a pattern where executives see impressive AI pilot metrics and conclude that their AI transformation is on track, while the underlying business model and competitive position haven't actually changed. Pilots generate usage data and efficiency metrics, but those metrics don't translate to the business outcomes that drive growth. The trap is particularly dangerous because it can persist for years — organizations continue investing in AI while producing increasingly sophisticated tools that don't change the trajectory of the business.

How should enterprises measure AI ROI beyond tool adoption?

The clearest signal of real AI ROI is change in the business outcomes that drive competitive position — revenue per employee, market share, customer retention, speed-to-market, or decision quality in high-stakes workflows. Tool adoption rates, usage metrics, and efficiency gains from automating existing processes are leading indicators at best. The organizations with the strongest AI ROI track records define the business outcome first and work backwards to the tool and workflow design.

Why do traditional industry enterprises have a timing advantage in AI adoption right now?

Organizations deploying AI now have access to substantially more capable models, more mature deployment infrastructure, and a richer body of failure cases to learn from than early adopters did in 2022-2023. Early movers are dealing with legacy AI debt, skeptical workforces, and technical infrastructure built on superseded technology. Traditional enterprises entering seriously now can build on what works without inheriting those constraints — but that advantage is time-limited as compounding returns accrue to earlier adopters.

What is the first step for a COO trying to move AI from experimentation to enterprise value?

Audit your investment allocation before changing anything else. Most enterprise AI programs are heavily skewed toward technology spending and under-invested in the organizational capabilities — training, change management, workflow redesign — that determine whether tools actually generate value. Rebalancing that allocation, and establishing a clear business outcome for every AI initiative before technology selection, is the move that unlocks the transition from experimentation to enterprise value.