Contents · 7 min read
01 / 05 · Current measure
Why do good pilots leave the current measure in place?
In short
Your pilot may prove that a tool can do the task, but managers still answer to the current performance measure. They will protect that measure because their review depends on it.
At your next monthly review, a manager may show that AI cut the time needed for a job, then still have to explain why the team recorded fewer filled hours. The pilot worked, but the measure used in that meeting still rewards filled hours, so keeping the old workflow protects the manager's review.
But the demonstration does not change what they must report each month. A manager judged on filled hours still needs filled hours, even when AI cuts the work.
This is why a pilot can look good in the room and still change little after it. As The Margin That Blinds You explains, the tool may work while the workflow and permission to change stay fixed.
Your managers are responding to the measures your company uses. Those measures make their choice rational because their review depends on them.
02 / 05 · Rational response
Why is keeping the old workflow rational?
In short
Faster work can leave a manager short of a utilization target because the same work now fills fewer hours. It can also turn saved time into a headcount cut. Goals based on usage or output can reward visible activity while the real work stays the same.
| What leadership asks for | What the current measure rewards | The rational manager response | What leadership must change |
|---|---|---|---|
| Deliver faster with AI | High utilization | Keep hours filled | Reward faster and better delivery |
| Release time for better work | Headcount savings alone | Avoid showing spare time | Count the better work made possible |
| Get teams to use AI | Tool usage | Drive logins without changing work | Judge the work that changed |
| Produce more with AI | Generated volume | Create more drafts | Judge the result and its review |
| Learn through use | No visible mistakes | Avoid tests that may fail | Give teams room to test and check |
Utilization is useful when your business still earns from filled time. It becomes a brake when the same manager is told to reduce the time each job takes.
Headcount goals can show a real saving, so they have a place in a wider plan. But if every saved hour points to fewer roles, managers learn that proof of speed may weaken their team.
Staff also watch what happens after a gain. The Only Asset AI Can't Build contrasts headcount cuts with using the same people to do work that was not possible before.
Usage goals reward the act of using the tool. Generated volume has the same flaw because more drafts can hide more review work.
Once you see the conflict between faster delivery and the current utilization measure, ask why a working tool still leaves daily work untouched.
03 / 05 · Daily work
How can the tool work while daily work stays the same?
In short
A pilot proves that the tool can perform one part of the job. Daily work changes only when the team changes the steps around it and has time to check the result.
A manager can add the tool while keeping every old check and meeting. The team then does the new work on top of the old work, so the gain never reaches the working week.
Review can also grow when output grows. If the team makes more drafts but no one has more time to check them, the tool creates a queue.
The manager responsible for the workflow and the leader who reviews the result must agree on which checks can go. If released time changes staffing, finance must also decide how that time is used. A manager cannot remove an old check if another leader still needs it for the current performance measure.
Why AI Pilots Don't Scale shows why a working tool is not enough when ownership and the work around it stay unchanged.
To set fair new performance measures, your leadership team must first learn what the new work asks of people.
04 / 05 · Lead first
Why must your leadership team use AI before setting new measures?
In short
Your leadership team cannot judge the new work from a demonstration alone. It must use AI in real work so it sees the checks and extra review that appear later.
Your leaders may support AI and still lack the facts needed to set fair goals. That is understandable because a short demo cannot show the full work.
Using AI in real work changes the discussion because leaders see which systems the tool must access. They also see where a person must check the result. They can then tell which old steps should go.
A people team offers a useful contrast. When it tests AI in its own work first, it can see weak access and the review load that the demo hid. What an AI-augmented HR department looks like explains why the people team must change itself before asking staff to change.
Leaders cannot set a fair goal for work they have not tried, which is why delegating AI too early fails.
With that experience, your team can change the measures before it asks others to change.
05 / 05 · New goals
What should you reward while the workflow changes?
In short
Reward managers for faster and better delivery while they change the work around the tool. Give them room to test and review, then judge whether daily work changed instead of counting raw use.
Have the leadership team test one live workflow
Each leader should use AI on real work and note where review adds time.
Have the workflow owner show what will change
Remove old steps only when the right leader agrees they no longer serve the work.
Change the manager's performance measure
Reward faster and better delivery without letting utilization punish the gain.
Give the team time to learn
Make room for tests and checks, so one weak result does not send everyone back.
Have the workflow owner track delivery
Check whether delivery improves, rather than counting logins or drafts.
The shift needs more than a new note from the CEO because goals shape choices each week. Managers need to know that good use will help their review instead of putting their team at risk.
If leaders change their own work, they can judge the same change in others.
Change the measures
Your next task is to have the leadership team test one live workflow, then change the measures that punish faster and better delivery. The paper on an AI-augmented management team shows how leaders can use AI in their own work while keeping judgment accountable, so your team can see where review is needed and decide what should change before setting goals for others.
The paper on what changes in the management team, with the first changes to make. Free, with a PDF to share.
Questions CEOs ask about this
Which management incentives cause AI adoption to stall?
AI adoption stalls after a successful pilot when managers are asked to use AI but judged by measures that reward the old workflow. At the next monthly review, faster delivery can leave a manager short of a utilization target, while saved time can put the team at risk. The decision is whether to change those measures before demanding wider adoption, then have your leadership team test the work itself.
Why would managers resist tools that make their teams more productive?
Before you call it resistance, compare the requested behavior with the manager's review criteria. If faster delivery would hurt the review, change the measure before you ask for wider use.
How do utilization and headcount targets distort behavior?
Decide whether released time should improve delivery or reduce staffing before you expand the pilot. Then make that choice explicit in the manager's goals so proving a gain does not create a hidden penalty.
Which incentives punish testing and learning?
Drop login and draft-count goals. Give the team time to test and check results, and judge whether the work improved after the workflow changed.
What should leaders model personally?
Choose one live management workflow and use AI in it before setting goals for others. Record where access fails and where human review is still needed.
How should goals change without becoming vanity metrics?
For each workflow, name the intended result and the required review. Then track faster and better delivery instead of logins or draft counts.
Founder, GRAIL
Johan has spent fifteen years in management consulting and executive leadership, including as Chief Product Officer of a SaaS company. He founded GRAIL to help European leadership teams work out what AI changes at the core of their business. He is the author of The Execution Revolution and the host of the ThinkRoom podcast.
Related reading
Sources
- The Margin That Blinds You: why a good pilot can fail to change the work around it. (accessed October 29, 2026)
- Why AI Pilots Don't Scale: why the workflow and its owner must change with the tool. (accessed October 29, 2026)
- What does an AI-augmented HR department look like?: why a people team must test new work before asking others to change. (accessed October 29, 2026)
- The Delegation Problem: why leaders need direct use before they can set the aim. (accessed October 29, 2026)
- The Only Asset AI Can't Build: how headcount cuts shape what staff believe AI means for them. (accessed October 29, 2026)
- What does an AI-augmented management team look like?: how leaders use AI while keeping judgment accountable. (accessed October 29, 2026)
Next scheduled review: February 7, 2027