Leading the AI Transformation: What the Job Actually Becomes

Leading an AI transformation is not a technology programme with an executive sponsor attached. It is a change in what the leadership job consists of: deciding faster on less certainty, choosing deliberately what will not be automated, keeping the organisation honest about what the tools are actually doing, and being the person who answers when one of them is wrong.

Most of the difficulty is not in the decisions themselves. It is that the people who have to make them are the ones with the least slack, and the information reaching them is poor.

That second part is measurable. PwC’s 2026 Annual Corporate Directors Survey found that only 18% of directors rate the quality of the AI information they receive as good or excellent when it comes to metrics linking AI outcomes, risks and business performance. For AI investments and ROI it is 25%. For workforce readiness, 31%. Boards are being asked to oversee something they are not being shown clearly.

The literacy question, answered properly

There is a tedious argument about whether executives need to understand AI technically. It resolves cleanly once you separate two different things.

Leaders do not need to know how a model is trained. They do need to know what a question costs to answer badly, which is a different kind of knowledge and mostly unavailable from a vendor. Enough to tell a real trade-off from a pitch. Enough to ask what happens to the data on exit and recognise an evasive answer. Enough to spot when a proposal automates judgment rather than handling, which is the single most common way these projects go wrong.

The market has already priced this in. BCG’s 2026 survey of 625 CEOs and board members found 79% believe AI literacy should be a baseline requirement for board leadership. Not technical skill. Literacy: the ability to participate in the decision rather than defer on it.

Deciding what will not be automated

The instinct in a transformation is to ask what can be automated. The more useful question, and the one only a leader can answer, is what should not be.

Every organisation has work where the inefficiency is the point. The call where a client says the thing they would never write down. The review where a senior person catches what the process missed. The judgment that is slow because it is weighing things that do not reduce to a number. Automate those and the metrics improve while the business gets quietly worse, which is a failure mode with a long lag and no alarm.

This is genuinely a leadership decision, and it does not delegate. It is a statement about what the company is for, and whoever makes it is setting the ceiling on what the transformation can cost you. Getting it wrong is not recoverable by buying a better tool.

The adoption problem is the whole problem

Here is the number that reframes most transformation plans. PwC’s Global Workforce Hopes and Fears survey found that only 14% of workers use generative AI daily at work, against near-universal organisational adoption.

That gap is not resistance. It is usually the absence of a reason: no redesigned process to use it in, no clarity about whether using it is encouraged or risky, and no visible signal from above. An Adecco Group study of 2,000 C-suite executives across 13 countries found the same pattern from the top, describing a widening gap between AI ambition and workforce readiness, with organisations lacking the leadership clarity and trust-building to convert adoption into results.

The practical implication is unglamorous. The binding constraint on most AI transformations is not model capability or budget. It is that the work was never redesigned, so the tools sit alongside the old process rather than replacing it, and people default to the way that still works.

Three things move this, and all three are leadership acts rather than technical ones: redesign one process visibly so there is something real to use, say plainly what is encouraged and what is not so people are not guessing about risk, and use the tools yourself in view of the organisation, because nothing signals permission like a leader doing it badly in public and continuing anyway.

Deciding faster on less certainty

The uncomfortable part of the job is the tempo. The honest answer to most AI questions in 2026 changes within a few quarters, which breaks the normal executive instinct to wait for clarity before committing.

The resolution is not to decide recklessly but to change what you commit to. Make decisions that are cheap to reverse and expensive only if never made: short vendor commitments rather than annual ones while a category is young, one process proven before the second is bought, an architecture where the model underneath can be swapped without a rebuild. That last point is why keeping switching costs near zero is a leadership decision rather than a procurement detail. It converts an irreversible bet into a revisable one.

The corollary is that some decisions genuinely should be slow. Anything touching client data, anything that would be expensive to unwind, anything where being wrong is invisible for a quarter. Fast on the reversible, slow on the rest, and the discipline is knowing which is which.

What stays yours

Underneath all of it, the leader keeps three things that do not transfer.

• The trade-off. What gets automated and what gets protected, which is a statement of values before it is a decision about efficiency.
• Accountability. When an automated process harms a client, the answer is never the vendor and never the model. Governance is how that accountability is made operable rather than theoretical.
• The standard. What good looks like here, which the organisation reads from what you accept rather than from what you announce.

Those three are why this cannot be reduced to hiring the right person. An accountable owner makes the stream of decisions manageable, which is the whole argument of the wider pillar. It does not remove the part of the job that was always yours.

Where to start

Pick the one process where the gap between what the tools could do and what your organisation actually does is widest, and redesign that process rather than adding a tool to it. Do it visibly. Then say out loud what you will not automate, and why.

Those two acts do more for a transformation than any procurement decision, and neither requires a budget.

If you already know which process you would redesign first, tell Daniel what it is, what you have built around it so far, and what it needs to do for the business. Book 30 minutes with Daniel.

Frequently Asked Questions

What does leading an AI transformation actually involve?

Four things that do not delegate: deciding faster on less certainty, choosing deliberately what will not be automated, keeping the organisation honest about what the tools are doing, and answering when one of them is wrong. The technical implementation belongs to an accountable owner. The trade-offs, the accountability and the standard stay with leadership.

Do executives need technical AI skills?

Literacy rather than technical skill. Leaders need enough grounding to tell a real trade-off from a vendor pitch, to ask what happens to the data on exit, and to spot a proposal that automates judgment rather than handling. BCG’s 2026 survey of 625 CEOs and board members found 79% believe AI literacy should be a baseline requirement for board leadership.

Why do AI transformations stall after the tools are bought?

Because the work was never redesigned, so the tools sit beside the old process and people default to the way that still works. PwC’s Global Workforce Hopes and Fears survey found only 14% of workers use generative AI daily despite near-universal organisational adoption. That gap is usually absence of a reason rather than resistance: no redesigned process, no clarity on whether use is encouraged, and no visible signal from leadership.

How should leaders decide what not to automate?

Look for work where the inefficiency is the point: the call where a client says what they would never write down, the review where a senior person catches what the process missed, the judgment that is slow because it weighs things that do not reduce to a number. Automating those improves the metrics while the business gets quietly worse, and the lag means nothing alarms.

How do you make AI decisions when the answers keep changing?

Change what you commit to rather than deciding recklessly. Favour short vendor commitments over annual ones while a category is young, prove one process before buying the second, and keep an architecture where the model underneath can be swapped without a rebuild. Move fast on reversible decisions and slowly on anything touching client data or expensive to unwind.

What should a leadership team do in the first month?

Two things, neither requiring budget. Pick the single process where the gap between what the tools could do and what the organisation actually does is widest, and redesign that process visibly rather than adding a tool to it. Then state plainly what will not be automated and why. Those two acts move a transformation further than any procurement decision.