Ask a room of executives what is blocking their AI return, and most will land on the same answer: our people don't have the skills yet. Deloitte's State of AI in the Enterprise 2026 — a survey of 3,235 senior leaders across 24 countries — confirms the instinct. Insufficient worker skills is named the single biggest barrier to integrating AI into existing workflows (Deloitte, State of AI in the Enterprise 2026). So far, so unremarkable.
The remarkable part is what companies do next. Faced with a skills barrier, 53% are educating the broader workforce to raise AI fluency — the most common response by a wide margin. Far fewer are touching the thing the skills are being poured into: only 33% are redesigning career paths, and a similar minority are re-architecting roles and workflows at all. The dominant move is to train people harder for a job that has not changed shape. That is the quiet reason so much AI training spend never converts into AI value — and for a mid-market Head of Operations, it is the most fixable mistake in this quarter's budget.
The Reflex: Train Harder Into the Same Role
There is nothing wrong with raising AI fluency. A workforce that cannot operate the tools is a non-starter. The problem is treating fluency as the whole answer when it is only the entry ticket.
Picture an accounts-payable clerk whose job is defined as "process 300 invoices a week." You send them to AI training. They come back able to prompt a model, summarize a contract, draft an email faster. Then they sit down in front of the same role — same 300 invoices, same workflow, same definition of the job — and use the new skill to shave minutes off tasks that the AI could have removed entirely. You have funded a more skilled person to do an unchanged job slightly faster. The skill went up; the work stayed the same. The return on that training is real but marginal, and it is nowhere near what the AI investment was supposed to deliver.
This is the trap Deloitte's data exposes. The 53%-versus-33% gap between funding fluency and funding redesign is not a rounding difference. It is the difference between equipping people for the work and rethinking the work itself — and the market is overwhelmingly choosing the former.
Why Upskilling Into an Unchanged Role Doesn't Convert
The reason training-without-redesign underperforms is structural, not motivational. The shape of a role determines the ceiling on what any skill inside it can produce. If the role is still defined as a sequence of manual steps, then the most AI-fluent employee in the world can only optimize those steps. They cannot delete them, recombine them, or hand them to an agent — because the job description, the handoffs, and the success metrics all still assume a human does each one.
AI role redesign is the act of lifting that ceiling. It asks a different question than training does. Training asks, "How do we help this person do their current job with AI?" Redesign asks, "Given AI, what should this job even be?" Those produce wildly different answers. The first keeps the 300-invoice clerk and makes them faster. The second notices that an agent can clear 270 of the invoices and reshapes the role into an exceptions-and-controls function — fewer transactions, higher judgment, a different success metric. Only the second captures the step-change the AI was bought for.
This is why Deloitte finds only 34% of organizations using AI to deeply transform their business, while roughly two-thirds remain on incremental or surface-level gains. The incremental majority is not under-trained. It is under-redesigned. Skills are being added to role-shapes that were never rebuilt to use them.
What "Redesign the Work" Actually Looks Like
If this still sounds abstract, Gartner's 2026 research makes it concrete and quantifies the prize. In its Future of Work Trends for CHROs, Gartner names "process pros, not tech prodigies" as the talent that unlocks AI value, and reports that teams who redesign their workflows with AI are roughly twice as likely to exceed revenue goals as teams who simply layer AI onto existing tasks (Gartner, Future of Work Trends for CHROs, 2026). The doubling does not come from better tools or more training hours. It comes from redrawing the process so the tool changes the outcome, not just the speed.
The organizational version of the same point shows up in MIT Sloan Management Review and BCG's study of the agentic enterprise: among companies with extensive agentic AI adoption, 45% expect reductions in middle-management layers (MIT Sloan Management Review & BCG, The Emerging Agentic Enterprise, 2025). Layers collapse only when the work underneath them is genuinely restructured — not when the same hierarchy simply gets more fluent. Redesign is what produces structural change; training alone produces a more capable status quo.
The Counter-Argument: "Training Is Faster and Safer Than Restructuring"
The honest objection from an operations leader is that redesigning roles is disruptive, and training is not. Reshaping a job touches headcount, comp bands, reporting lines, and people's sense of security. Sending the team to a workshop touches none of that. Given a choice between a safe lever and a disruptive one, why not pull the safe one first?
Because the safe lever quietly costs more. Training into an unchanged role is not free — it is a recurring spend that buys a marginal return, repeated every budget cycle while the structural gain stays out of reach. The "disruption" of redesign is real but bounded and one-time; the underperformance of training-only is modest but permanent. And the disruption is smaller than it looks when you do it at the grain of a single role rather than a whole function. You are not restructuring the company this quarter. You are rebuilding one job around what the AI can now actually do, learning from it, and deciding whether to extend the pattern. The risk of that is a contained experiment. The risk of training-only is a compounding gap you never named.
The Payoff Is Large — and It Only Shows Up After Redesign
The reason this is worth the discomfort is the size of the lift on the other side. The Harvard Business School and BCG field experiment on knowledge workers found that those who used AI well completed tasks roughly 25% faster and produced output rated about 40% higher in quality than the control group (Harvard Business School & BCG, 2023). That is the prize every AI budget is implicitly underwriting.
But notice the condition: used AI well. The lift materialized where the work was genuinely reorganized around the model's strengths and away from its weaknesses — not where people simply had access to the tool inside their old workflow. The 40% quality gain is a redesign outcome wearing a training-budget label. Pour money into fluency without reshaping the role, and you fund the access but forfeit the lift. That is the precise mechanism by which two-thirds of organizations end up on incremental gains while believing they invested in transformation.
The Q3 Move: Redesign One Role Before Funding More Fluency
You do not need to restructure the organization to act on this. You need to invert the order of operations on one role. Before approving the next tranche of AI fluency training, take a single high-volume role on your team and redesign it around the AI first.
Three steps make it concrete. First, map the role as it exists — every recurring task — and mark which an agent could own outright, which a human must keep, and which should simply stop. Second, rewrite the role around what remains: a new task mix, a new success metric (judgment and exceptions, not transaction volume), and the handoffs between human and agent made explicit. Third — and only third — fund the training, now aimed at the redesigned role rather than the old one. Done in that sequence, the fluency spend lands on a job built to use it, and you can measure the difference against the roles you have not yet touched.
The Decision for This Quarter
The Deloitte data is a mirror, and most operations teams will not like the reflection: the AI skills gap is real, and the standard response to it is making the return smaller, not larger. Training a workforce into role-shapes that predate AI is not a path to transformation; it is the most expensive way to stay incremental.
So before you sign off on another round of AI fluency training, ask one question of your most AI-heavy role: have we redesigned this job around the AI, or are we just teaching people to run faster inside the old one? If it is the latter, hold the training budget for two weeks and spend them redesigning the role instead. Everyone on your team can be trained. The companies that win the next two years will be the few who redesigned the work first — and then trained people for the job that actually exists.
