Practice

Leading AI transformation

For the leaders and managers who have to turn an AI ambition into how the work is actually done.

AI changes what people do, how teams work together, and what a manager is there for. Those changes are much harder to lead than the technology is to buy. Leading them takes a view of where the value is, a working understanding of what the technology can and cannot do, and a grasp of how people and organizations actually change. Most organizations hold those three capabilities in three different departments.

The two questions

An AI program has to answer two questions. Where is the value, and who has to learn something new. The first question gets a budget, a use-case list, a platform decision and a steering group. The second one is usually treated as a communication task.

A communication task is the right answer when the knowledge already exists and only has to reach people. It is the wrong answer when nobody has the knowledge yet, because then the work is people changing how they judge their own output, what they call finished, and what they were proud of doing by hand.

That work has no obvious owner. The executive team owns the ambition, the technology function owns the platform, and the manager a few levels down is left with the part where people have to work differently.

Why this is genuinely hard

An AI transformation asks for several different kinds of expertise at the same time, on the same problem. Each of these is a discipline in its own right.

• A strategic view: Which parts of the business this should touch first, what it is worth if it works, and what to deliberately leave alone this year.
• A working grasp of the technology: What these tools can do this quarter, where they fail, and which of those failures matter for your work. The technology moves fast enough that an understanding formed last year is now a liability.
• Risk and compliance, brought in early: Most of the use cases worth doing touch data that somebody is accountable for. Involving that person at the end is a common reason a working pilot goes no further.
• People, and how organizations change: Roles, incentives, career paths, and what people believe good work looks like. This is human resources and learning territory, and it is usually the largest part of the work.
• Mechanisms rather than intentions: An agreement that relies on people remembering it tends not to hold. This work needs review points, criteria, named owners, and something that inspects whether any of it happened.
• Experiments, and the ones that fail: The answer usually cannot be specified in advance, so progress comes from running things and reading the result. That needs an agreed definition of an acceptable failure, settled before the first failure happens.
• Empathy, and staying in the conversation: People are being asked to change work they are good at, sometimes work that defines their professional and personal identity. A leader that shies away from that conversation will get a polite version of it and act on the wrong information.
• Culture, capability and resilience: This runs over years. The organization has to be able to keep going after the first two or three attempts do not produce much, which is often the case.

Most organizations already hold every one of these somewhere. They sit in different functions, with different budgets and different reporting lines, and they have rarely been asked to work on one problem together. This kind of transformation needs all the relevant departments and disciplines to come together.

What is different for a manager

Some managers have led a change before. It might have been a reorganization, a system migration, or a new operating model. AI transformation asks things of them that previous changes did not.

• You may not be the most expert person in the room: For most of a management career, the person running the work had done the work. Some of your people are now further ahead on these tools than you are, and that will keep happening. Position has stopped being a reliable proxy for knowing.
• Your output changes from answers to questions: Where somebody already knows how, a manager assigns, resources and unblocks. Where nobody knows yet, what a manager produces is a question sharp enough for a team to work on, the protection that lets them work on it, and a clear statement of what is not changing while they do.
• You have to name what people are losing: Most rollouts announce the upside. Far fewer name what a person is being asked to give up, which is often a craft they were good at and part of why they were valued. That objection then arrives in the form of a process complaint, which nobody can resolve, because the complaint is about something else.
• You have to make it safe to be slower before faster: Real use of an unfamiliar tool is slower for a while. If the quarter punishes that, people will use the tool where it is safe and keep doing the work that counts the way they always did.
• You have to change what gets measured: Existing measures keep the existing behaviour in place. Changing what is asked for in a review, and what gets recognized, does more than any amount of encouragement.

Two kinds of work

The first practical move is telling which kind of problem is in front of you. Every change in an AI program is a mixture of these two, and the mixture decides who has to do the work.

• Technical: Somebody already knows how. The knowledge exists, inside the organization or outside it, and it can be hired, bought or taught. The problem can be stated clearly and success measured against a known target. It feels like pressure: hard, sometimes exhausting, and orienting, because you know what good looks like and who is accountable.
• Adaptive: No existing expertise contains the answer. Progress requires the people involved to change what they believe, what they value or how they work, so it cannot be handed to a specialist. It feels like distress: disorienting and personal, conflicts surface, people avoid it, and to some of them it feels like loss.

Technical does not mean easy. A heart bypass is technical work: years of training, thousands of repetitions, and a life on the table. What makes it technical is that the surgeon already knows how, and the patient does not have to change. Getting that same patient to eat, drink, sleep and handle stress differently once they are home is the adaptive half, and it is the half that decides the outcome. A lot of stalled AI rollouts bought a technical answer for an adaptive problem.

The distinction between technical and adaptive work comes from Ronald Heifetz and his colleagues at the Harvard Kennedy School. What follows is our own framework for applying it to AI, built over several years of doing this work with leadership teams.

The higher the ambition, the more of the work is adaptive

We sort every engagement onto three value paths. The further up an organization reaches, the less of the work is the kind somebody already knows how to do.

• Boosting productivity: Mostly technical. The workflows are understood and the tools are proven, which is why most organizations start here. The adaptive part is small and sharp: somebody is being asked to give up a craft they were proud of, and a rollout plan has nothing to say about that. The leader’s move: Name the loss as well as the upside, and change what gets measured, or the existing scorecard will keep the existing behaviour in place.
• Creating new value: An even split. Building the thing is technical. Deciding what deserves to exist, who the change is for, and which of your current commitments it breaks, is not. The leader’s move: Say what a good experiment and an acceptable failure look like before anyone runs one, protect the work from this quarter, and stop the bets that are not working.
• Driving disruption: Mostly adaptive. There is no reference implementation for a business model that does not exist yet, and the people who would have to let go of the current one are the same people holding the decision. The leader’s move: Discuss the disruption openly, say clearly what does not change, and keep the question open longer than the room would like.

The AI value framework in full

Three moves a leader makes

These are not three stages. A leader runs all three inside a single conversation, more than once, and the skill is knowing which one the room needs at that moment.

• Diagnose the work: Split the challenge into the part somebody already knows how to do and the part nobody does yet. Then ask two questions about the second part: who has to learn something, and what do they give up to do it. Write it as one sentence with a verb in it. Run it again every quarter, because the line moves as the tools improve.
• Regulate the heat: Enough pressure to move, not enough to freeze. Raise it by naming what is at stake and keeping the question open past the point where the room wants it closed. Lower it by breaking the work into smaller pieces, setting a pace, and sequencing what comes first. Both directions are moves, and many managers have practiced only one of them.
• Give the work back: Frame the challenge without handing over the answer. The tell is in your own sentence: if it still contains the solution, you have delegated the task and kept the problem. Giving the work back does not mean stepping away from it. You stay responsible for the standard, the deadline, the pace and the protection, and you back the experiment when it does not go to plan.

Why this is never finished

Adaptive work turns technical over time. People experiment, something works, it gets written down, and once it is written down it can be specified, verified and handed to a tool. At the same time AI takes over more of what was already technical. Capacity keeps coming free on the technical side.

The decision that matters is where that capacity goes. Spending it on more of the same work is the common outcome and the smallest return. Putting it back into the adaptive side, on the questions that were not worth anyone’s time when answers were expensive, is where the difference accumulates.

So a manager runs the diagnosis again rather than once. A challenge that was adaptive last year may have a known answer now, and a workflow that was settled may have become a question again.

The workshop

Leading Transformation in the Age of AI

The framework above, run as a working session rather than a talk. Your own challenges go on the table, your own people do the sorting, and everyone leaves with one sentence they have to act on.

• For a leadership team: The group that set the ambition sorts the three or four changes they are actually driving. The output is an agreement about which parts have a known answer and which do not, who owns each, and where the heat needs to go up or down.
• For a management layer: Run as a short series for the managers who carry the work. Each session takes live challenges from their own teams, so they practice the moves on real problems rather than on cases, and they leave with language they can use with each other.
• Inside a program you already run: Where there is an existing management or talent program, this goes in as a module and uses the measures, the vocabulary and the cases you already have, rather than arriving as a separate initiative to be absorbed.

We run it with your own people and your own challenges. There is no standard deck, and we would rather spend the time on three of your real problems than on ten of somebody else’s.

Where this connects

• Where the value is: Before the leadership question comes the portfolio question. Use cases scored onto the three paths, with an owner and a first result named for each. The AI value framework
• Enablement and adoption: Training, champions across functions, and the governance that lets people move without asking permission every time. Implementation and adoption
• Mechanisms that hold: The review points, criteria and routines that carry a change once attention has moved on. Innovation mechanisms
• Sessions that start it: Where a leadership team needs to see what is possible before they will commit to changing how they work. Inspiration and activation

Who this is for

Executive teams who set an AI ambition and cannot yet see it in the work. Heads of function whose rollout went quiet after the launch week. Management populations being asked to lead a change nobody has handed them a method for. And human resources and learning teams who own a leadership program and can see that the existing modules do not cover this. We work in regulated businesses as often as unregulated ones.

Common questions

What is the difference between AI training and AI adoption?

Training moves knowledge that already exists to people who do not have it yet. That is useful, and it is often exactly what is needed. Adoption is what happens afterwards: people decide whether to change how they work, what to stop doing by hand, and whether they trust the output enough to put their name on it. A program that funds only the training produces certified people whose work looks much the same as it did before.

Why is leading an AI transformation harder than other change programs?

It needs several kinds of expertise at once. A view of where the value is, a current understanding of what the technology does and where it fails, risk and compliance judgement, knowledge of people and organizational change, and the mechanisms to make any of it stick. Those capabilities sit in different functions in most organizations and have rarely had to work on one problem together. On top of that, the destination is unclear rather than unpopular, and some of the people being led know more about the tools than the person leading them.

Our AI rollout stalled. Where do we start?

By sorting the specific thing that stalled. Write the challenge in one sentence with a verb in it, then ask whether somebody somewhere already knows how, who has to learn something, and what they give up. A rollout that stalled on the adaptive part will not restart with a better tool or a louder mandate, and knowing which part failed is what separates a second attempt from a repeat of the first.

Who should be in the room?

The people who hold the problem, which usually means the leadership team for the framing and the management layer for the work. Sending a delegate is a common way this goes wrong. The adaptive part cannot be handed to a specialist group without turning back into a technical project, which then stalls in the same place for the same reason.

How long before we see anything?

A first sort takes a session, and it usually changes what a team argues about that same week. Managers running the moves without us takes a few months of practice on real decisions. The change in how an organization handles the next unfamiliar thing shows up over quarters.

We shape this around the value you are reaching for, the layer of the organization that has to carry it, and how much of the work turns out to be adaptive once you look properly.

Talk about your leadership team30 minor send a message