Think about the person on your team who has gotten the most out of AI this year. The one who collapsed a workflow everyone else still does by hand. Who built the thing the team now quietly depends on. Who is doing the work of two or three of the old seats and making it look easy. You have been meaning to recognize it properly, and you will, once the quarter calms down.
Here is the part that was not in the AI briefing. That person is, statistically, the most likely person in the building to leave.
A 2025 survey of over two thousand employees and leaders found that seventy-eight percent of AI power users were actively looking for a new job. Among the employees most resistant to AI, sixty-five percent planned to stay.1 Read those two numbers next to each other. The people creating the leverage you have been bragging about to the board are heading for the door, and the people who never opened the tool are the ones settling in.
The traits that produce leverage are the traits that produce mobility
This is not a coincidence, and it is not disloyalty. The disposition that makes someone an AI power user is the same disposition that makes them hard to keep.
The champions in Recognizing Leverage share a temperament. They delegate work they used to own. They treat a wrong answer as a prompt to refine, not evidence the tool is broken. They go looking for the work where being wrong is cheap and automate it before anyone asks. That temperament does not switch off at the org chart. A person who reflexively removes friction from a workflow feels the friction in their own career with the same sharpness, and they have spent the year building exactly the portfolio that makes another company want them.
They also know their price. A market that can cheaply measure output, which is increasingly every market AI touches, can bid for the people who produce it. Your champion has a number in their head, and it is not the one in your compensation band. What sits underneath the survey is a simple asymmetry: the leverage is real, it is legible from the outside, and the person carrying it can take it with them.
The raise is not the lever, and you probably can’t win with it anyway
The instinct when a strong performer looks restless is to reach for comp. For this person, comp is the weakest tool you have.
The evidence is almost perverse. EY found that employees with extensive AI training were fifty-five percent more likely to leave their organization, and that only twelve percent of employees received enough training to unlock the productivity in the first place.2 So the standard playbook, invest in your best people, produces a portable skill and no reason to stay. You paid to make them more employable and did nothing to make them more attached.
What the same research points to as the actual retention levers are not money. Flexibility. Access to the technology they want. Room to grow.2 Translate that out of survey language and it is autonomy, air cover, and a next problem worth solving. Your champion is not primarily leaving for a bigger number. They are leaving because the leverage they discovered has nowhere to go inside your walls. They automated their own job, and instead of being handed a harder one, they were handed the same job with a thank-you.
The model is not going to advocate for this person. You have to.
The number for your boss
If comp will not hold them, the case for acting still has to survive a budget conversation.
Replacing an employee costs somewhere between fifty and two hundred percent of their salary once you count recruiting, the vacant seat, and the months a replacement takes to ramp.3 For an ordinary hire, that is a painful line item. For the person operating at real AI leverage, the visible replacement cost understates the damage badly, because what walks out is not one seat’s worth of output. It is the workflow they redesigned, the automation the team leans on, and the institutional knowledge of which corners were safe to cut. Up to forty percent of a company’s AI productivity gains get lost to weak talent strategy.2 The gap opens when the people who generate the gains leave faster than the org can absorb what they built, and no amount of training closes it.
That is the figure to bring to your boss. Not the salary. The leverage that leaves with the salary, and the quarters it takes to rebuild.
Spotting them is the easy part; keeping them is a design choice
You already know who this is. It is the same person the talent guide tells you to find: quiet, usually shipping, rarely given air cover, doing work that has not made it onto the dashboard yet. The reason they are a flight risk and the reason they are hard to spot are the same reason. Their wins do not show up in the standard review template, so the org undervalues them right up until a competitor does not.
Retention here is an org-design decision, and it looks like the opposite of a retention program. Give them a function to redesign, not a spot bonus. Put them on the workflow work that Driving Adoption says has to happen anyway, the redesign of the broad middle’s stuck processes, and let being good at it be the visible, promotable path it currently is not. Make their leverage measurable, so it survives contact with a promotion committee. Measuring Returns is partly a retention instrument for exactly this reason: a champion whose before-and-after is on a slide is a champion the organization has a reason to keep and a language to reward. One whose wins are invisible is one you will read about in a two-week notice.
Something to carry
Write down the name of the person on your team getting the most out of AI right now. Next to it, write what they would say if you asked them, honestly, what would make them want to still be here in two years.
If the honest answer is a number, you have a problem, because it is a number a better-funded competitor can beat and you probably cannot. If the honest answer is a harder problem, more room, and the standing to redesign how the work gets done, then you are holding a retention plan you can act on this quarter, and the quiet person who built half your leverage has a reason to build the other half here.
Footnotes
-
Betterworks, “2025 State of Performance Enablement: AI and the Employee Experience.” Survey of more than 2,100 employees and leaders in the US and UK. 78% of AI power users were actively seeking new roles; 65% of AI-resistant employees planned to stay. ↩
-
EY, “Companies are missing out on up to 40% of AI productivity gains due to gaps in talent strategy,” November 2025. Employees with extensive AI training were 55% more likely to leave; only 12% received sufficient AI training; cited retention levers were total rewards, flexibility, technology access, and career opportunity. ↩ ↩2 ↩3
-
SHRM guidance on turnover cost: replacing an employee runs roughly 50 to 200% of annual salary once recruiting, vacancy, and ramp-up are counted, commonly estimated at six to nine months of salary. ↩