The AI Productivity Boom Is Hiding an Operations Problem

AI adoption has reshaped how employees work. However, the productivity gains have not reshaped how businesses perform. According to the World Economic Forum, about two-thirds of firms actively use AI, yet roughly 89% reported no measurable improvement in labor productivity over the past three years. The tools are present. The gains are not.

The adoption curve is steep. According to the Census Bureau, 18% of firms used AI in a business function between November 2025 and January 2026, rising to 32% on an employment-weighted basis. Use is concentrated in large firms and knowledge-intensive sectors.

The gap between adoption and outcome has a name. According to MIT, roughly 95% of generative AI pilots at large companies produce no measurable P&L impact. The pilots are not failing because the models are inadequate. They are failing because the workflows around them do not change.

AI can make an individual employee faster. It cannot make a business better if the handoffs, escalations, and communication paths connecting that employee to the rest of the organization remain broken. Value leaks at the seams. The productivity gain stops at the desk.

Sunstone Systems is a company that works with mid-market businesses on operations and technology. The firm was founded by Scott Dorn, a former automation engineer who spent years building operational systems for large enterprises. He now applies that work to agencies and service businesses.

Dorn says, "AI is not the problem; the problem is that most companies are using AI to work around broken workflows instead of fixing the workflows themselves. That hides the real issue. It does not solve it."

The problem shows up in communication between departments. Companies adopt AI to speed up those exchanges without first defining what the exchanges should contain. Business-critical information ends up in emails, chat threads, and reports. Traceability drops. Tasks hang.

"Most companies have not defined what the signals between teams should look like," Dorn says. "They add AI on top of that. Then there is communication everywhere and no clarity anywhere."

Most companies do not find out that the process was never working until they try to scale it. The workarounds that held the system together at one size stop working at another. An approval that once took a phone call now requires three. A handoff that once worked because two people sat near each other now spans time zones and tools that do not speak to each other. The AI layer does not expose those fractures. It hides them under a faster surface.

Hours saved is the default metric for AI success. Dorn argues it is the wrong one. He points to output and throughput instead. "A sales role that once handled a certain volume of calls and now handles roughly double that volume at the same close rate is a measurable gain," he says. A time savings that does not translate into more or better output is not.

"Time is not the goal," Dorn says. "Output is. If you cannot point to more throughput or better quality from the same team, you have not improved anything. You have just moved the work around."

His advice to companies evaluating AI is to look at their data first. If the data is not clean and reliable enough to build a report on, the company is not ready. If the data looks good but one person is manually cleaning it before it reaches leadership, the company is still not ready.

"If you are truly ready for AI, the answer to where it should go is obvious," Dorn says. "If it is not obvious, the foundation is not there yet. You have someone plugging holes."

Dorn describes the work as operations first, technology second. Sunstone Systems maps how work and tasks flow through a business, identifies where handoffs and approvals break down, and builds a central system to replace manual coordination. AI is added only after that foundation is in place.

The argument is not that AI is overhyped. It is that AI is being deployed in the wrong order. Automation and workflow design come first. AI comes last on top of a foundation that can support it.

The question for mid-market operators is not whether to adopt AI. It is whether the workflows underneath are ready to receive it.