
What the 5% do differently with AI

The paradox
Eighty percent of organizations report using AI, and only one in ten says it has real impact. The numbers shared at TransformHR’s August 2026 panel on AI showed how wide that gap runs: 95% of pilots deliver no measurable return, and 42% of companies have abandoned AI implementations entirely, up from 17% the year before.
A small group is pulling ahead all the same, and joint research from Revelio Labs and Ramp found that the companies investing most heavily in AI kept employment about 10.2% higher than companies that had not adopted it, and raised the entry-level share of their workforces by 1.15 percentage points. Adopters were already growing faster, so those figures show a correlation, and they suggest that the firms using AI most intensively are still hiring as they adopt it.
Why do two groups with access to the same technology get such different results? The panelists from Toast and Revelio Labs called it a people and organizational design problem. Burke Autrey, President of Fortium Partners, a ZRG company, sees it from the technology leadership seat and takes it a step further: “When an ‘AI initiative’ fails, what failed was a business initiative. Either the problem wasn’t understood, the options weren’t evaluated, or nobody had the authority to make the change.” In his view, much of what gets reported as AI failing was mislabeled from the start, because it should have been an initiative to achieve a business outcome, with AI informing the options and opening up things that weren’t possible before.
Why the other 95% stall
Four causes come up again and again. Most companies redesign nothing: Deloitte’s 2026 State of AI in the Enterprise survey of 3,235 business and IT leaders found that 84% have not redesigned jobs or workflows around AI, so they speed up a broken process and hand managers more output to review. Leaders are often the least AI-fluent people in the building, and many privately admit they can’t keep up, so they hold back permission and scaling freezes. Tool sprawl makes it worse, as every vendor launches overlapping AI features and enterprises spend months connecting products that were never designed to work together. And enablement doesn’t scale, because a webinar is forgotten by lunchtime and leaves a few capable people scattered across a confused organization.
The predictions many of us believed haven’t held up either. Hierarchies haven’t flattened, and Revelio Labs finds demand for experienced leaders growing, because someone has to review and make sense of all that new output. Managers who expected AI to free up their time are carrying more review work than before, and entry-level work is holding up at the firms that adopt most intensively.
Five patterns of the 5%
- Redesign the workflow before choosing the tool. The companies that succeed start with a workflow they want to improve and then look for the tool that helps. BNY, a 50,000-person, highly regulated bank, asked how it could spread expertise across the organization with AI and designed backwards from that goal, and it has started rolling out AI agents company-wide. Revelio Labs’ Dr. Lisa Simon explained the reason: AI’s effects on the labor market are showing up first in what jobs contain. The firm estimates that 87% of the recent change in the economy’s mix of work activities is happening inside existing occupations. People keep their titles while what they do day to day changes substantially, which is what redesigning a workflow does to a job.
- One executive owns the outcome. t works best to put the executive who owns the P&L in charge of the result, because companies commonly assign AI to a leader, committee or Center of Excellence that has no authority over the business outcome. “To achieve a business outcome, someone has to have the authority to make sweeping changes,” Autrey says. “That rarely happens by committee, and it works best when one executive owns the outcome and can overturn decisions from departments that aren’t ready for the change.” That lines up the incentives, because the leader’s KPI is concrete, such as improving a workflow by 40%, and the ROI has to show up in the business results that leader is measured on. That P&L owner still needs a technology leader to own the AI agenda behind the result, and Autrey is specific about who that should be: the CIO in a company that buys its technology, and the CTO in a company that builds commercial software.
- Pick two workflows and get them right. The 5% pick two workflows with large ROI potential and get them right before moving on. Toast identified interview transcription as a high-impact workflow, found the right tool, piloted it for two months, and then told everyone this was how the company would work from then on, and adoption reached 97% company-wide. The approach works because it removes ambiguity and gives people a clear story about the problem being solved and what changes for them. It also cures tool sprawl, since these companies start by asking what their strongest people do and what judgment they apply, and then build that into systems the company owns.
- Build proficiency through hands-on coaching. The 5% teach by showing people how the tools apply to their own work, one example at a time. Toast’s Jeff Moore described it this way: “I don’t tell my team to use AI. I show them. I send them examples. Are you using this for senior hires? What if you threw their LinkedIn profile into this tool? Then they come back with things cooler than I did.” He was blunt about the alternative: “We just throw them into the wild and think magic beans are gonna sprout.” The pattern we see working is small groups of 10 to 15 people meeting four times over four weeks, led by their own managers and built around their own jobs. Each session combines a one-hour demo, 30 minutes of discussion and 30 minutes of hands-on work with a real tool, so people start from where they are and gain confidence by doing the work.
- Make room in people’s workload. People told to adopt AI while still hitting their normal targets usually manage neither. The 5% take other priorities off people’s plates and accept that early work will be imperfect. Moore put it plainly: “You have to create space. We’re gonna launch this thing and it might fail. And that’s okay. Because you’ll learn a ton.”
Putting the patterns to work
If you are a CEO or business leader, here is how to sequence the first months.
- Week 1: Audit every pilot. Keep the ones producing useful learning or measurable ROI and stop the rest, so the team can concentrate on a few.
- Weeks 1 to 2: Run working sessions where your executive team does real AI work on your company’s problems, guided by someone who uses these tools daily, so the team builds a shared language and loses some of its fear.
- Week 2: Identify two or three high-impact workflows by asking frontline managers what they wish they could automate to free up 20-plus hours per person each week.
- Week 3: Assign the P&L owner for each workflow, with a clear target and the freedom to decide how to get there.
- Week 4: Give explicit permission to fail. Call the work experiments, and celebrate the failures that taught you something
From there, build proficiency through hands-on training and plan about six months for the first workflow, covering design, tooling, a soft launch with 20-30% of users and then full rollout. The pace is set by how much change the company can absorb. “Every company can endure only so much change in a given period,” Autrey says. “The job is to understand why things are the way they are, whether that’s the industry, regulation, investors or the CEO’s own ideology, then lay out how much change the company can take and in what sequence.” By month six you should see one workflow improved by 30-40% and a team that wants to take on the next two.
Leaders stalled by the fluency gap need one-to-one coaching from their own manager on the problems in front of them this week. Legacy organizations will move more slowly, so bring compliance in early and assign senior ownership.
Autrey has seen what happens when that ownership is missing. A board told its CEO it wanted an AI-focused executive in the business within weeks, and the company brought in a capable leader. The work stalled anyway, because the new executive arrived with an open checkbook and a green field while the CIO had already been told that a regulatory deadline was the top priority for the next six months. “The organization set two conflicting mandates at the same time and never decided who had the authority to change priorities,” Autrey says.
AI also makes it easy to produce more work, so build review and validation into the rollout from the start, or managers will end up with more to check than they had before.
The choice
The companies in the 5% plan on 6 to 18 months before results show and three to five years for the full payoff, and they spend that time working through the patterns above. Companies that keep buying tools and running one pilot after another usually end up with wasted budget and a workforce that has stopped believing the next pilot will be different. Dr. Simon expects the gap between the two groups to widen.
For most companies, the place to start is the week 1 audit of every pilot already running.
This article draws on the TransformHR panel discussion held August 27, 2026, featuring Jeff Moore (Toast) and Dr. Lisa Simon (Revelio Labs), moderated by Matt Corbett, as well as the Revelio Labs AI Labor Market Tracker (August 2026) and Revelio Labs/Ramp research, “The Companies Spending the Most on AI Are Also Spending the Most on Humans.”
.webp)








.webp)








































