Writer’s 2026 enterprise AI survey landed this month with numbers that should end a few arguments. They surveyed 2,400 workers across global enterprises. Ninety-seven percent of executives said their organization has deployed AI agents. Only 29 percent reported meaningful return on investment. That is a 68-point gap between having AI and getting anything from it.

The spending side is just as lopsided. Separate research from the same period shows 59 percent of companies now invest more than a million dollars a year in AI. The money is flowing. The returns are not.

This is not a technology failure. The models work. The tools are mature enough. What is failing is the layer between the tool and the outcome: the actual work.

The practice gap has a number now

For the past two years, the conversation has been about adoption. Are you using AI? How many tools? What percentage of your workforce has access? Those were the right questions in 2024. They are the wrong questions in 2026.

The right question is whether your organization changed anything about how it operates. Not whether you bought a license. Not whether you rolled out a chatbot. Whether you looked at a workflow, identified what an agent could own, restructured the handoffs, and measured the result.

Ninety-seven percent cleared the first bar. They bought the thing. They deployed it. They checked the box. Twenty-nine percent cleared the second bar. They actually changed the work.

That second bar is where the value lives. And it is where most companies stopped climbing.

What the 29 percent did differently

The companies reporting real ROI did not buy better AI. They did something less exciting and more effective: they redesigned the process before they plugged in the tool.

Look at what Microsoft reported in its FY26 year-in-review this week. One of their highlighted customers, Chow Tai Fook, deployed over 400 customized AI agents supporting more than 24,000 employees. The result was a 70 percent efficiency gain across core business processes. That number did not come from a smarter model. It came from mapping agents to specific decision points across customer, product, and operational workflows before a single agent went live.

That is the pattern. The 29 percent who see returns did the unglamorous work of understanding their own operations first. They identified which decisions repeat, which handoffs create friction, and which steps could be handled by an agent with clear boundaries. Then they deployed.

The other 68 percent deployed first and figured they would find the value later. They are still figuring.

Why the gap compounds

Here is what makes the 68-point spread dangerous rather than just disappointing. The organizations getting value from AI are now reinvesting that value into more refined deployments. They have data on what works. They have teams that understand how to scope an agent’s role. They have muscle memory around workflow redesign.

The organizations still searching for ROI have none of that. They have cost. They have complexity. And they have a growing pressure from leadership to show results, which pushes them toward more deployment rather than better deployment. More tools, more licenses, more agents. The same mistake at higher volume.

This is how a practice gap becomes structural. The 29 percent are learning faster because they have something to learn from. The 68 percent are spending faster because they have nothing else to try.

Six months from now, the 29 percent will have iterated three or four times on workflows that already produce returns. The 68 percent will still be running pilots.

The question leaders should be asking

The instinct is to ask “which AI should we be using?” or “are we spending enough?” Both questions are already answered for most companies. You are using AI. You are spending on AI. That phase is over.

The question that matters is simpler and harder: what changed about how your team works since you deployed?

If the answer is nothing, you are in the 68 percent. And that is not a waiting position. It is a compounding disadvantage. Every month you run AI without redesigning the underlying work, you accumulate cost, technical debt, and organizational confusion while your competitors accumulate operational intelligence.

Writer’s survey gave the gap a number. It is 68 points. And it measures the distance between deploying AI and actually using it to change how an organization operates.

The organizations that close that gap will not do it by buying better tools. They will do it by doing what the 29 percent already did: start with the work, not the technology.