Redesign the Work, Not the Tool

Two companies buy the same AI model, on the same budget, in the same quarter. A year later one is beating its return targets and the other is quietly writing the investment down. The model did not decide the outcome. The work around it did.

Deloitte’s 2026 Global Human Capital Trends, a survey of more than 9,000 leaders across 89 countries run with Oxford Economics, puts a number on the gap. Sixty-six percent of leaders say intentionally designing how humans and AI work together is important to their success. Only 6% say they are actually leading on it. Almost everyone can see the work that matters, and almost no one has started it.

The reason is not access to technology. Nearly 60% of workers already use AI intentionally at work, according to a Melbourne Business School study cited in the same Deloitte report, so the tool is already in the building. What is missing is design. What follows is the method the 6% use, broken into four decisions any CEO can make about a single workflow before the end of the week.

Why Bolting AI On Underperforms

Most organizations, 59% of them, take what Deloitte calls a tech-focused approach. They buy the model and leave the workflow untouched, which means a new tool runs the old process a little faster. It feels like progress because something changed, and the invoice proves it. The work itself did not change at all.

Those are the companies least likely to see the returns they expected. The ones that redesigned the work first, deciding where a human still adds judgment and where the machine adds scale, are twice as likely to exceed their AI ROI targets. The organizations leading on intentional human-AI design go further still, and are nearly 2.5 times more likely to report better financial results. Same tools, same budget, different results, and the difference is design.

So the question worth asking about your last AI investment is not how good the model is. It is whether anyone redesigned the work, or simply bolted AI onto the process that was already there. The four decisions below are how the redesign actually happens.

Decision One: Draw the Line Between Judgment and Scale

Take one workflow where you have already deployed AI and have the person who runs it lay out its steps in plain language. Then force one question onto each step. Either this step needs a human’s judgment, the kind that weighs context, reads a room, or owns a consequence, or it needs the machine’s scale, the kind that handles volume, speed, and repetition without tiring.

That line is the whole design. It is not obvious, it moves from one workflow to the next, and it is never permanent, but it has to be drawn on purpose. When no one draws it, AI does not wait politely at the boundary. It drifts into the judgment steps too, because those are often the slow ones, and the slow steps are exactly what a speed tool is drawn to. That is how a team quietly loses the muscle it most needs to keep, which is the case for deliberately keeping some work manual even when the machine could do it. Deciding where the machine stops is the same call as deciding when a human still has to weigh in, and getting it right is where the redesign begins.

Decision Two: Rebuild the Workflow Around the Line

Automation asks a small question: where in this process can AI save us some time. Redesign asks a harder one: if we built this work from scratch today, knowing exactly what the machine can now do, what would it look like. The first question keeps the old shape and trims it. The second is willing to throw the shape away.

This is also where intent shows. Deloitte found that 56% of leaders design their AI primarily for business outcomes such as cost and speed, while a growing 40% design for both the business outcome and the human’s experience of the work. The gap between those two numbers is not a soft preference. Work that is redesigned only to cut cost tends to hollow out the roles around it, and hollowed-out roles are where good people leave and quiet errors accumulate. Naming which outcome you are designing for, before the redesign starts, is the part most teams skip.

Decision Three: Measure the Work, Not the Tool

Most AI dashboards measure usage. Seats activated, prompts sent, hours logged inside the tool. All of it counts activity, and none of it tells you whether the work got better. A team can push its usage numbers up every quarter while the outcome the work exists to produce stays flat, and the dashboard will call that a win.

The 6% measure the other thing. For each AI initiative they name one metric that describes the work outcome, the resolution that actually held, the decision that landed faster, the error that stopped happening, and they track that instead of the tool counter. This is the same trap that shows up across the market, where almost everyone reports adoption and almost no one reports impact. Redesign gives you something real to measure, because you changed the work, not just the toolset.

Decision Four: Own It as an Operating Decision, Not a Culture Program

Only 14% of leaders say they are adept at shaping human-AI interactions, and that thin number explains why the redesign so often gets handed to the wrong owner. It looks like a people problem, so it goes to HR as a training push or a culture initiative, and it stalls there.

Deciding where humans and machines each earn their place in the work is not a culture program. It is a decision about how the company operates, which department carries which call, where a person signs off and where the system runs on its own. That is an operating decision, and operating decisions belong to the person who owns the operating model. This is the CEO’s redesign to lead, not a program to delegate downward and check on later.

The Nerve to Redesign

There is a reason so few leaders do this, and it is not laziness. Redesign is slow, and 7 in 10 leaders say their competitive edge over the next three years is being fast and nimble. So they move fast, skip the redesign, and land squarely in the 59% that underperform. Speed without redesign is not an advantage. It is a faster way to run the old process with a more expensive tool.

The companies pulling ahead did the opposite. They slowed down long enough to draw the line, rebuild the work around it, measure the outcome, and own the call at the top, and then they moved. The edge in 2026 is not access to AI, because nearly everyone has that already. It is the nerve to redesign the work around it, and the willingness to treat that redesign as the real investment, with the model as the smaller line item it always was.