Eighty-eight percent of enterprises use AI in at least one function, but only thirty-nine percent report measurable impact on earnings. Only six percent qualify as high performers, defined as five percent or more of EBIT attributable to AI. That gap has not closed, and the share reporting EBIT impact has not grown year over year despite far more spending.
Distance to first proof measures how many days until a real person uses a real version of something and has an opinion. Enterprise AI has shortened the distance to adoption. It has not shortened the distance to proof.
The six percent who reached measurable returns share a pattern. Fifty-five percent of high performers redesigned workflows around AI, compared to twenty percent of everyone else. Nearly three-quarters have redesigned now, up from about half a year ago. They did not buy a tool and graft it onto an existing process.
Most organizations did the opposite. Deloitte's 2026 State of AI survey found thirty-seven percent using AI at surface level with minimal process change, thirty percent redesigning key processes, and only thirty-four percent deeply transforming. The largest single group is the one doing the least.
Michael Hammer wrote in 1990 that "heavy investments in information technology have delivered disappointing results, largely because companies tend to use technology to mechanize old ways of doing business." He described an application that took twenty-two days of elapsed time but only seventeen minutes of actual work. Even doubling the speed of each step recovers roughly eight minutes while leaving the queue bottlenecks untouched.
Thirty-six years on, the pattern has not changed. A UK Department for Business and Trade trial of generative AI found no robust evidence that time savings translated into improved productivity, despite seventy-two percent user satisfaction. Users liked the tool. The work did not change.
The agentic wave should fix this. It is making it worse. Deloitte's 2026 agentic readiness survey of 501 senior manager to C-suite respondents found that only twenty-one percent say their processes are ready for agentic AI adoption, the lowest of any area tested. Only fifteen percent have scaled orchestrated, cross-functional multi-agent adoption.
Fewer than one-third expect majority process redesign within two years. When you layer a copilot onto step seven of a ten-step process, the proof you collect is about step seven. The user is faster at step seven. They report satisfaction with step seven. But the bottleneck was never step seven: it was the approval queue after step nine, or the three steps that exist because of a policy nobody revisited.
The winning teams asked whether the process should have ten steps. The rest made step seven slightly faster and filed a success report.
Distance to first proof, applied to enterprise AI, shows a design failure the spending reports miss. The buyer reported adoption. The dashboard showed usage. Neither is proof. Proof is a changed outcome for the person doing the work.
Until the process changes, that person's work does not change. The enterprise that redesigns its workflows will keep outperforming the one that spends more on models. The rest will report higher adoption next year and the same EBIT. The spending looks like investment. It is decoration.
.webp)
.webp)
.webp)
