AI Transformation Governance and Executive Sponsorship

Before reading on, make a guess. In companies that use AI, how often does the chief executive personally oversee how AI is governed? Half the time? Most of the time? Hold a number in your head.

In McKinsey’s global survey published in March 2025, 28 percent of respondents whose organizations use AI said their CEO was responsible for overseeing AI governance. Seventeen percent said their board of directors was. Yet the same survey found that CEO oversight of AI governance was one of the attributes most closely linked to higher self-reported bottom-line impact from generative AI, and at larger companies it was the attribute with the strongest link of all¹.

Read that finding carefully. It is a correlation in a survey, and the impact is self-reported. Nobody should conclude that a chief executive’s signature creates value. The more plausible reading is about what a very senior owner can do that nobody below can: settle trade-offs that cross functions, move money and people between units, and stop things. Building the Multi-Year AI Roadmap ended with a roadmap made of decision milestones. This chapter is about who makes those decisions, with what authority, and how fast.

The idea: govern the trade-offs, not the tasks

Transformation governance is the way leadership sets direction, allocates money and capacity, settles trade-offs between functions, accepts material risk and removes blockers, so that the roadmap actually moves. Executive sponsorship is the authority, resources and time that one senior leader brings to those decisions, together with the duty to answer for them.

The word governance is overloaded, and three different things often share it. Keeping them apart saves a great deal of meeting time.

The first row belongs to Module 7, where AI Governance Committee and AI Governance Office sets out the bodies that approve systems, and the last row belongs to program teams. This chapter also assumes who decides what, from AI Decision Rights and Accountability, and where each capability sits, from Centralized vs Federated AI. It asks the narrower question that sits above them: when two functions disagree about the transformation itself, who settles it?

Governance is not management

Governance decides what, why, how much and with what risk. Management decides how and when, and does the work. Both are needed, and they fail in opposite directions. Executives drift down into management when their review turns into a reading of red, amber and green status lines, with every hard question “taken offline”. They drift up and away when they approve the roadmap once a year and are not seen again until it is late.

The research on decision-making supports judging governance by its decisions. Marcia Blenko, Michael Mankins and Paul Rogers of Bain argue that executives over-rate the org chart and under-rate how well critical decisions are made and carried out². In a Bain survey of almost 800 companies, those that excelled at making and executing key decisions were strongly associated with top-tier financial results³. A transformation forum, then, should be judged by what it decided this month and how fast, not by how many reports it received. A useful question for every such meeting is whether we are on the right roadmap, not only whether we are on schedule. A project can be exactly on time while the direction is wrong.

A sponsor is not a champion

A champion promotes, advocates and encourages. Champions are valuable, especially early, and anyone can be one. A sponsor can do four further things: decide, fund, escalate and remove blockers, and then answer for the result. The two roles can sit in one person. They are not the same role, and much weak sponsorship is a champion with a sponsor’s title.

Two long-running practitioner surveys point the same way. The Project Management Institute’s 2018 Pulse of the Profession, covering more than 4,400 practitioners, 447 senior executives and 800 project office directors, found for the sixth year running that an actively engaged executive sponsor was the top driver of projects meeting their original goals. The figures measure different things, so read them separately. One organization in four named inadequate sponsor support as the primary cause of failed projects. Among PMI’s top performers, 83 percent of projects had an executive sponsor assigned, against 42 percent among underperformers. And organizations with actively engaged sponsors on more than 80 percent of their projects reported 40 percent more successful projects than those below 50 percent⁴.

Prosci’s change management benchmarks, built from more than 10,800 responses over 25 years, have ranked active and visible sponsorship as the top contributor to success in every study since 1998. Projects with extremely effective sponsors met or exceeded their objectives 79 percent of the time, against 27 percent for those with extremely ineffective sponsors. Only 48 percent of participants said their sponsor was effective or very effective⁵.

Treat both with the caution any self-reported survey deserves. They measure association, and teams that succeed may be generous to their sponsors. But the finding has held across two bodies and two decades, and both describe the effective sponsor in the same active terms: making decisions, removing roadblocks, communicating directly and building a coalition. Neither describes visibility on its own.

The four powers of a sponsor

What exactly must a sponsor be able to do? Four powers, and the time to use them.

Mandate means setting the priority and defending it, which includes saying out loud what the organization is not doing. Money means the authority to move budget and people between units, because AI capacity is often contested. Trade-offs means settling conflicts between functions, or taking them to the one level that can. Escalation means removing blockers and providing cover: when an initiative must stop, the sponsor takes the heat, so that teams are not punished for the evidence they produced. The stop rules set in Prioritizing the AI Portfolio only work if someone holds this last power.

The time matters as much as the powers. A sponsor does not run the program; that is the program lead’s job. But a sponsor who cannot find two hours a fortnight for it will turn every decision into a delay. A simple test settles most arguments about whether a program has a real sponsor: ask the named sponsor for the last three decisions they made on it. If there are none, the program has a champion.

Who should it be? Usually the executive whose results or mission the change serves most, with authority across the functions it touches. That is often not the chief information or technology officer. Technology leadership is essential, but, as AI Operating Model argued, the outcome is owned in the business. Write the mandate down on one page: the outcomes, the budget authority, what the sponsor may stop, the escalation route and the review dates. Name a deputy. Executives change roles, and a mandate that lives only in one person’s head leaves with them.

One sponsor, a coalition beneath

One sponsor cannot be present in every unit where work changes. John Kotter’s study of failed transformations named “not creating a powerful enough guiding coalition” as the second of eight common errors⁶. Prosci’s benchmarks make the same point in practical terms: effective sponsors build a coalition of the leaders and managers whose people must change⁵.

The transformation board is where the portfolio is decided. It should be small enough to decide and broad enough that the owners of the usual conflicts (business, technology, risk, finance, people) sit at the table. It passes a simple authority test: can it move money, can it stop an initiative, and can it re-sequence the roadmap? A body that can do none of the three is a discussion forum, however senior its members. A transformation office may support it, tracking dependencies, milestones and realized value, but it does not approve small decisions. In many companies the same executives also sit on the AI governance committee of Module 7. That is fine, provided the two agendas stay separate: one approves systems, the other steers the investment.

Escalation with a clock

Escalate when a decision exceeds someone’s delegated authority, when two functions conflict, when risk or investment is material, or when the trade-off is strategic. Do not escalate routine execution, and take each issue only to the next level that has the authority to settle it. A low-risk experiment should never climb the whole ladder; risk-based routing, covered in The AI Governance Operating Model, keeps light things light.

A common failure is not the wrong decision. It is no decision, while an issue circulates between steering meetings for weeks. The remedy is a clock. Each level gets a fixed number of working days. If the issue is still open when the time runs out, it moves up one level automatically, with a one-page paper that sets out the options, the evidence and a recommendation, in the format AI Governance Committee and AI Governance Office describes.

When functions disagree, the sponsor’s questions are short. What outcome matters here? What risk are we willing to accept? What are we trading away? Who owns the decision? Then the sponsor decides, or takes it to the executive committee, and records the decision with its reasons and a review date. Phased options often beat a yes or a no: launch where the risk is controlled, hold back where it is not, and gather evidence on the rest.

What the board needs

The board oversees the transformation; it does not run it. It needs to see the strategic opportunity and threat, the material risks, the major investments and their realized value, and whether the governance described in this chapter actually works. It does not need project detail. Given that only 17 percent of respondents in the McKinsey survey said their board oversaw AI governance, many boards have work to do¹. How AI matters should reach the board, and in what format, is set out in AI Governance Committee and AI Governance Office.

Story: the publisher that had a sponsor and no decisions

A national news publisher, with a daily paper, a large digital subscription business and a podcast network, had an AI roadmap with four initiatives: short summaries at the top of articles, synthetic-voice audio versions of stories, better recommendations for subscribers and an archive search tool for reporters. Its chief product officer was named sponsor. She was an excellent champion. She spoke about AI at every town hall, and the newsroom liked her.

The summaries were ready to launch in the spring. Then four functions disagreed. The standards editor wanted a human to check every summary. The commercial director wanted them live before a renewal campaign. The technology team’s engineers were committed to migrating the subscription platform. The legal team pointed out that the licenses for wire-agency stories might not allow summaries at all. The issue went to a monthly steering group of fourteen people, which had no budget and no power to move engineers. Each meeting ended with the same action: take it offline. Eleven weeks passed.

The chief executive changed three things. The managing director of the digital business, who held the subscription budget, became sponsor, with a written one-page mandate and a deputy. The steering group was replaced by a transformation board of five, the editor-in-chief, the commercial director, the chief technology officer, the general counsel and the sponsor, with the power to fund, stop and re-sequence. And every open trade-off acquired a clock: ten working days at the board, then to the sponsor with an options paper.

The summaries question reached the new board first and was settled in eight days. Summaries would launch on the publisher’s own reporting only, with wire copy excluded until the licenses were renegotiated, and editors would review a daily sample rather than every item. The subscription migration kept its engineers. The board also stopped the recommendations initiative, which had no owner in the commercial team, and freed its budget for the audio pilot. None of these decisions needed new technology. Each needed someone who was allowed to make it.

What this means for leaders

Sponsorship is a job with powers, not an honor. If you are named as a sponsor, you should be able to say what you can fund, what you can stop and where you take what you cannot settle. If you appoint one, give them the budget authority to match the title. A transformation board earns its place by deciding: it funds, stops and re-sequences, or it is a forum. And every open trade-off needs a clock, because in AI programs the costliest decision is often the one that nobody makes.

Check yourself

1. A respected senior executive named on the program charter is enough sponsorship.MythFact
2. In the 2018 PMI survey, a quarter of organizations named inadequate sponsor support as the primary cause of failed projects.MythFact
3. The transformation board should review every AI initiative, however small, to keep control.MythFact
4. A forum that cannot move money, stop an initiative or re-sequence the roadmap is a discussion forum.MythFact
5. The chief information officer should always be the sponsor of AI transformation.MythFact
6. An unresolved trade-off should move up a level automatically when its time box runs out.MythFact

Reflection: whose decision is it?

What comes next

Authority settles trade-offs, but it does not change how people work. A sponsor can decide to launch the summaries; editors still have to trust them, and the business still has to see that they matter. The next chapter, AI Change Management, KPIs and Operating Rhythm, turns to execution discipline: how people adopt the change, which measures show it is working, and the operating rhythm that keeps decisions coming.