Contents · 9 min read
- Why do your numbers barely move when people say they are faster?
- What can AI actually change in your company?
- What do you get from individual productivity, and what does it take?
- What do you get when a step stops, and what does it take?
- When does your customer get something new, and what does that take?
- How do you use the three levels with your leadership team?
- Sort the work
- Questions
01 / 06 · The puzzle
Why do your numbers barely move when people say they are faster?
In short
Most of the gains people report are real, but a faster task is only part of the week and the work around it stays the same. That is why big personal gains rarely show up in your numbers.
Someone on your leadership team tells you their people are three times faster with AI, and you believe it because you have seen the work. Then the quarterly numbers arrive, and the department looks much as it did a year ago.
Part of that gap is plain arithmetic. Say drafting takes one day of a five-day week, and AI makes it five times faster. The drafting shrinks to a fifth of a day, but the other four days do not move, so the week takes 4.2 days. That is about 19% faster, which is a good result but far from five times.
The rest of the gap comes from what happens around the task. The faster draft lands on the desk of someone who still works at the old speed, so it waits there. Someone may also have to check it, which moves the work instead of removing it. And the person who saved the time often has nowhere useful to put it, so it leaks away into small things.
This is also why a working AI pilot can leave the business unchanged. So before you push for more speed, it helps to ask what actually changed.
02 / 06 · The sort
What can AI actually change in your company?
In short
AI can change three things: a person's work, the way work flows, or what customers get and pay for. Sort every AI move by which of the three changed, because each one pays off differently and asks for different things.
Each of the three has a name and one question that tests it, and you can put these questions to any AI idea your team brings you.
A smarter agent does not change which of the three you are in. A very clever agent can still help only one person, while a plain one can stop a step for a whole department. So the test is always what changed in the business.
These are not stairs, either. You do not have to finish one before you start another. The customer level is also not the right aim for every company, so start with the one you almost certainly have already.
03 / 06 · Individual productivity
What do you get from individual productivity, and what does it take?
In short
People do the same work faster or better, so you get time and better work but rarely money in the P&L. It takes access to the tools and people who can tell when the tool is wrong.
Your finance lead prepares the board pack with an agent, and your sales lead uses one to get ready for a customer meeting. The process around them stays as it was, so nothing stops. The test is simply who got faster.
What you get is worth having. The work comes back sooner and often better, and leaders who have used AI on their own work ask better questions about the other two levels. But the gain stays with the person, so it rarely reaches the P&L.
There is one clear exception. In work like customer support, countable cases are already waiting in line, so a person who clears them faster simply clears more of them. There the gain can reach the numbers with nothing redesigned.
What it takes is the smallest ask of the three. You need licenses and each person's own documents in the tool, and it needs no IT project. You also need people who can tell when the tool is wrong, because it sounds just as sure on the work it does badly.
That small ask is why many companies stop here, since it asks nobody to give anything up. The next level does.
04 / 06 · Team and process
What do you get when a step stops, and what does it take?
In short
Team and process redesign means a step, handoff or approval is gone, merged or moved. You get work that finishes sooner, and freed time that can show up in a budget line. It takes authority, from the line manager inside one department and from the management team when the work crosses departments.
The test is one question: what stopped? If you cannot name the step that is gone, the change stayed with individuals, however clever the agent is.
What you get shows up in numbers the department already keeps, such as how long the work takes from start to finish and how much of it is waiting. Because the capacity is freed in one place, you can also decide where it goes. We expect most AI bets in a mid-sized company to sit here, though that is our judgment and not a measured result.
What it takes is harder, because someone has to give up a step they own. Five things need to be in place:
One process
Pick repeating work with a clear start and finish that lands in a budget line, such as a customer contract or a supplier invoice.
Someone who can stop a step
Inside one department that is the line manager. When the work crosses departments only your management team can redraw it, because a department head has no say next door. Incentives have to change with it.
Connected systems
The agents need to reach company data and the next system in line, while your people move to checking the unusual cases. Your own IT opens that access one step at a time.
A place for the freed time
Name it before you start, such as a shorter queue or a vacancy you leave unfilled, because otherwise the time leaks away.
A way to tell if it failed
Agree one quality number that must not get worse, a review date and a condition for stopping.
Keep the steps that exist because mistakes are expensive. And fit the change to how the work is really done, because a redesign that ignores that can leave you worse off.
Everything so far still sells your customers what they bought last year. The last level changes that.
05 / 06 · Business model
When does your customer get something new, and what does that take?
In short
Business model or new customer value means a customer pays differently or gets something they could not get before. You get a new buyer, product or way to earn, at low odds. It takes a bet from the CEO and owners, tested with a real customer at a real price.
The test is strict, because otherwise everything counts: who pays differently, or gets something they could not get before? Internal speed does not count, however impressive it is. Neither does a new feature inside the same deal, since the customer is still buying what they bought before.
What you get is a new way to earn, and perhaps something a competitor cannot quickly copy. It is also the least proven of the three, because real cases are rare and the ones we found are only partial. Walmart began selling its route planning system to other businesses in March 2024, after using it in its own supply chain. Its announcement names no customers, price or revenue, so nobody outside can yet say how much it matters.
What it takes is a bet. Too few companies have done this to prove any list, so this one is our judgment:
A decision from you and your owners
It changes what the company sells, so it cannot sit lower down.
Something worth selling
Walmart's product is a system it built and ran for itself first, and yours may start the same way.
A real customer and a real price
Try it early with one customer who knows you, because only a customer can tell you what it is worth.
Its own budget and its own review
Your normal budget will judge it too soon.
AI may also force this on you by cutting the price of what your customers pay for, and this test shows whether it is doing that. Most mid-sized companies will not have a bet here yet, and that is fine. What every company can do is sort what it already has.
06 / 06 · Your list
How do you use the three levels with your leadership team?
In short
All three run at the same time. Take your AI list to your leadership team, put each item under one of the three, and give it what that level takes.
| Level | The test | What you get | What it takes |
|---|---|---|---|
| Individual productivity | Who got faster? | Time and better work | Access |
| Team and process redesign | What stopped? | Faster work and freed time in a budget line | Authority |
| Business model / new customer value | Who pays differently, or gets something new? | A new buyer, product or way to earn | A bet |
Run it in your next leadership meeting:
Write down everything you are doing with AI
Put every project and proposal on one page, however small it is.
Put the three tests to each item
Expect most to pass only "who got faster?", because that is where most AI work starts.
Ask what would have to stop
For each item that could matter, ask what would have to stop for it to reach the budget. If nobody can name a step, keep it as individual productivity and stop expecting it in the numbers.
Give each item what its level takes
A change to a process gets a named owner and a review date, and a new offer gets a customer to try it on.
Put the result in your plan
The process changes and any customer test go into your AI roadmap, each with an owner.
Sort the work
The task now is to sort your AI work by what it actually changes, then decide which process bets deserve authority and which customer bets deserve a test. See The Opportunity gives your management team three workshops to use AI in its own work and examine where it matters across the company. GRAIL then turns that work into a ranked roadmap, so each opportunity can be matched with the access, authority or bet it requires.
Three to four weeks, three workshops with your management team: hands-on use of AI, custom AI agents for each leader, and a ranked roadmap of opportunities. Fixed fee, and no IT project to get started.
Questions CEOs ask about this
What are the three levels of AI change in a company, and what does each one take?
AI can change a person's work, the way work flows, or what customers get and pay for. The first takes access, the second takes authority and the third takes a bet. If no step stops and no customer pays differently, even a powerful tool is still individual productivity. Your decision is where each AI move on your list should land.
Why do people report big AI gains when our numbers barely move?
Because a faster task is only part of the week, and the work around it stays the same. If one day in five gets five times faster, the week gets about 19% faster, and even that can leak away in queues and checks.
What is the difference between AI productivity and redesigning a process?
With productivity, a person does the same work faster and every step stays. With a redesign a step, handoff or approval has gone, and you can name which one.
Do we have to go through the levels in order?
No. All three can run at the same time, and we have found no study showing that companies must go in order. Many start with individual productivity because it asks the least, but a team can start on one process right away.
When has AI really changed the business model?
When a customer pays differently or gets something they could not get before. Working faster inside the company does not count, and neither does a new feature inside the same deal.
Which level should a mid-sized company spend most of its effort on?
For most, team and process redesign, because a gain can reach a budget line there without a bet on the whole company. That is our judgment, and it depends on where your own work is held up.
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
- Why does AI adoption stall, even when the pilot works?: why a working tool can leave the business unchanged. (accessed November 7, 2026)
- Is your leadership team actually AI-fluent?: what leaders gain from using AI on their own work. (accessed November 7, 2026)
- Who should own AI in a mid-market company?: who chooses what AI changes and who owns the result. (accessed November 7, 2026)
- The Access Ladder: how your own IT opens access one step at a time. (accessed November 7, 2026)
- Walmart Commerce Technologies launches AI-powered logistics product: Walmart's announcement, 14 March 2024. (accessed November 7, 2026)
Next scheduled review: February 7, 2027