Your AI leader was hired to change the company. Their charter only lets them run AI.

Two-thirds of companies say AI has made them more productive. Only a fifth say it has grown their revenue (Deloitte, January 2026). If that’s your company too, the gains are stopping before they reach revenue.

AI’s payoff seems to need changes to how the whole company works (who owns what, who decides what, what people are rewarded for), and the people hired to lead it are mostly given authority over AI and nothing else.

What the people leading AI were given

In May, IBM reported that 76% of the 2,000 organisations in its CEO study now have a Chief AI Officer, up from 26% a year earlier. That only counted titles. I wanted to know what responsibility and authority the titles came with, so I wrote a rubric and had an AI model apply it to the charters of 50 senior AI roles, from job postings and appointment announcements, quoting the sentence behind every code.

Most of them (32 of the 50) could build AI capability and nothing else: tools and platforms, training, and a pipeline of use cases. Another 16 could also change how work gets done inside other teams. Two held a P&L, both for an AI business. None could change how the company is organised. Thirteen reported to the CEO or the board, but none was given the CEO’s authority to do it.

These are job postings and appointment announcements, and a company can give authority it never writes down. So don’t take my sample’s word for it: the check below is for your own charter.

Why that matters

The same IBM study found that organisations which had redesigned how they work across five areas, including how functions work together, were four times more likely to have delivered on their business objectives. BCG’s September survey of 1,330 executives found that the companies getting the most from AI redesign the organisation around it, rather than giving existing staff new tools to run the same processes. And in Microsoft’s survey of 20,000 people who use AI at work, 45% said it felt safer to stick to current goals than to redesign their work, while only 13% said they’re rewarded for reinventing it, even when the results aren’t met.

None of that proves cause, and not every study agrees. The Bank for International Settlements, looking at more than 12,000 European firms, found that spending on organisational and process change didn’t measurably add to AI’s productivity gain, while training and data did. It’s the only one of these that measures productivity rather than asking about it. But it measured money spent on change, not who has the authority to decide, which is the gap I’m pointing at.

So if your AI leader’s job is to make the spend show up in delivery, and the charter only lets them build AI capability, I’d expect the work to wait on the decisions nobody gave them. In June, KPMG found that only 24% of companies make the CEO or executive committee accountable for AI-informed decisions, and that where the CEO is accountable, companies report established ROI more than three times as often.

Check your own charter

Take the delivery improvement you expect from AI and follow the work through to the customer. Where does it still wait? Pick a decision that would shorten the wait, then find the sentence in your AI leader’s charter that lets them make it.

If work waits on another department’s approval, can your AI leader change where that approval sits? If delivery needs teams to share ownership, can they change who owns what? Or can they only recommend a change to the people who decide now?

One CEO-reporting Chief AI Officer charter in my sample said the role would “help inform” and “guide how to embed AI”. If yours reads like that, the role can recommend changes to how your teams are set up, and someone else has to decide them. Usually that someone is you.

We’ve been here before

I spent ten years as an Agile coach, and I think we’re repeating what happened then. Agile asked companies to change the whole system around their teams. Most rollouts changed local processes and terminology, and left the rest alone: work still split across narrow specialists, everyone kept as busy as possible, and plans still treated as promises. I think those habits did more to stall Agile than anything the teams did. Most companies now seem to be treating AI the same way, as another tool to roll out (the way they’d roll out Office 365), with the same habits left in place.

So I’m testing it. Until the end of January, I’m coding AI postings beside postings for other change roles at the same levels, on what each is asked to change and the authority it’s given. My two predictions, fixed in advance: AI roles will less often be asked to change how the company is organised, and when they are, they’ll more often lack the authority to do it. If I’m wrong, I’ll publish that companies set up every kind of change role this way. The results and the full plan come out in February.

I help leadership teams find where the work waits once AI makes it faster, and then work with them on changing who can decide what. At my last client, I helped the leadership reorganise to take better advantage of AI.

If you’d like a second reader on your own AI leader’s charter, or the job ad you hired from, send it to me here. I’ll tell you, in writing, which decisions the work needs that it doesn’t give them. No charge.

You can find me on LinkedIn.

I’ve written more about the parallel with Agile.

Appendix: the evidence

The research behind “Why that matters”

All of these are surveys or observational studies; none shows cause. Figures are as each publisher states them.

• Deloitte, State of AI in the Enterprise (January 2026; 3,235 leaders surveyed August–September 2025). Benefits achieved today: efficiency and productivity 66%, lower costs 40%, revenue growth 20% (74% hope for it). Deloitte doesn’t say whether the productivity gains are individual or company-wide. 84% haven’t redesigned jobs or the nature of work around AI.
• IBM Institute for Business Value, CEO Study (May 2026; 2,000 CEOs and senior leaders, 33 countries, surveyed February–April 2026). 76% of organisations have a Chief AI Officer, up from 26% in 2025. 79% of executives say they’re decentralising decision-making. Organisations that redesigned five business areas (technology, finance, HR, operations and cross-functional collaboration) were four times more likely to have delivered on business objectives.
• BCG, The Applied AI Index 2026: The Formula for Agentic AI Value (September 2026; 1,330 executives, 60+ countries; self-reported). Calls AI transformation fundamentally an organisational shift. The leading companies redesign the organisation around AI rather than giving staff tools to run current processes, expect most of the value from reshaping processes, and put nearly three-quarters of their top priorities into people, organisation and processes, where BCG’s 10-20-70 rule puts 70% of the effort. BCG finds almost half of companies now generate measurable value.
• Microsoft, 2026 Work Trend Index (May 2026; 20,000 people who use AI at work, 10 markets). 45% say it feels safer to focus on current goals than to redesign work with AI; 13% say they’re rewarded for reinventing work with AI even when results aren’t met. Organisational factors explained more than twice as much of the reported AI impact as individual ones (67% against 32%).
• KPMG, Global AI Pulse (Q2 2026; 2,145 respondents). 24% make the CEO or executive committee ultimately accountable for AI-informed decisions. Established ROI: 14% where the CEO is accountable, 4% where the CEO is less accountable or not. (Q3 2026: 18% CEO or executive committee.)
• Against: BIS Working Paper 1325, Aldasoro and others (January 2026; 12,000+ EU firms). AI adoption raised labour productivity about 4%. Investment in training and in software and data increased the gain; investment in organisation and business-process improvement didn’t, to a significant degree.

The 50 charters

The sample. 50 roles at 49 organisations (Chief AI Officers, heads and VPs of AI, directors of AI transformation, AI centre-of-excellence leads): 39 job postings and 11 appointment announcements, dated 13 August 2025 to 31 July 2026. It was built from title searches, not drawn at random, so it supports no claim about all AI roles. No logged-in LinkedIn session was used. I wrote the rubric and its predictions; an AI model (Claude) did the coding, and every code had to be defensible from a quoted sentence.

The codes. “Operating model” here means how the company is organised, including who decides what.

The two 2s are a P&L for an AI practice at Atos and pricing and revenue for an AI-infrastructure business at Deeter Analytics. A stricter coder would put both at 1 and report zero; they met the P&L test set before coding, so they stayed.

Words that kept a charter at 0: build AI capability, deploy, platform, use cases, AI literacy, centre of excellence, roadmap of AI initiatives, govern AI. Words that would have lifted it to 2: redesign the operating model, reshape teams, own the P&L, change incentives, authority to reorganise, set reporting lines. “AI operating model” didn’t count by itself: at Cint and Intradiem it meant the AI office’s own intake and delivery (1). Relativity named process redesign, then “influence across teams without formal authority” (1). WTW’s charter, the advice-verb example above, coded 0. Lundbeck’s announcement quoted its CEO on how AI will change the industry, but the charter named AI strategy, governance, capability and deployment (0).

The predictions. Two were registered before coding. At least 70% would be function-only: that failed, at 64% (exact 95% interval about 49–77%). Fewer than 20% would have authority over the operating model or a P&L: that held, at 4% (about 0.5–13.7%).

A first comparison. In October, using job postings only, live or posted August to mid-October 2026, 14 more AI-transformation roles were coded beside 32 COO, general-manager, president and transformation-head roles. None of the 14 AI charters gave authority over the operating model or a P&L. Of the 32 others, 11 did on grounds other than P&L, and 21 counting P&L, so some companies do write that authority down. Fourteen was too few for the test I’d set in advance, so it wasn’t run. A model from another company (OpenAI’s), coding blind, read the rubric more generously and found the same direction with a much smaller gap. COOs and general managers often own a business outright, so they were always likely to score higher on P&L.

Checks. A fresh pass of the model re-coded 10 mixed rows: authority agreed on nine, reporting line on all 10. That’s one model checking itself, much weaker than two independent coders. By 19 August, 22 of the 50 sources could still be opened, live or in a checked archive capture. For the other 28, the charter sentence captured at coding is all that’s left, which is why each row carries it.

Limits. The coding reads what charters say, not what the people in the roles can do, and a company can mean authority it never writes down. The sample is chosen by the AI label: someone who owns how the company works, with AI inside it, is often a COO or general manager with none of the titles searched, which probably pushes the result the way it went. The comparison is small and its groups weren’t matched.