The question leadership asks that nobody in the company can answer
The short version · 4 key points
- 56% of chief executives report no revenue gain and no cost reduction they can attribute to artificial intelligence (PwC, 2026).
- Eight in ten users say they are more productive, yet only 37% of organizations report a positive EBIT impact (McKinsey, 2026).
- 93% of data and AI leaders name culture and change management as the main obstacle. Only 7% blame the technology (Data & AI Leadership Exchange, 2026).
- What the board is missing is not technology. It is management: knowing which tools are running, what they are measured against, and who answers for each one.
The meeting was going well. Budget reviewed, projections in order, everyone ready to wrap up. Then someone at the table looks up and asks a short question.
"Last year's AI investment. What did it give us?"
And the silence runs a beat too long.
It isn't that there's no answer. There are too many, all of them partial. Marketing produces three times the content. Engineering ships faster. Customer service has an assistant handling a good share of tickets. Every function brings its own story and none of them add up. Nobody can put a number on the table that survives a follow-up question.
If the scene sounds familiar, it's worth knowing it's the norm.
What the measurements say
PwC's 29th Global CEO Survey, presented at Davos in January 2026 with 4,454 chief executives across 95 countries and territories, found that more than half, 56%, have seen neither revenue gains nor cost reductions they can attribute to artificial intelligence. Three in ten report higher revenue and a quarter report lower costs.
McKinsey's The state of AI in 2026: On the road to ROI, published on 25 August 2026 with 1,719 participants surveyed between 4 May and 8 June, puts its finger on that silence. Eight in ten users report that artificial intelligence improved their personal productivity. Only 37% report any positive contribution to their organization's EBIT, essentially the same figure as the year before, even as more companies move from pilots to scale.
And just 6% of organizations reach what McKinsey calls high performance: attributing 5% or more of EBIT to artificial intelligence.
People are working better. The company isn't seeing it. That gap is what produced the silence.
Why the answer doesn't exist
The 15th annual AI & Data Leadership Executive Benchmark Survey, run by Data & AI Leadership Exchange, collects responses from roughly 110 Fortune 1000 companies and global brands, with senior data and AI leaders participating by invitation. 93% pointed to human factors, culture and change management, as the main obstacle to adoption. Only 7% blamed the technology. It is the highest that figure has been in the survey's fifteen years.
Translated into what happened at your company: someone approved a purchase, someone else implemented it, teams started using it, and at no point was there a conversation about what it would be measured against. When the question comes twelve months later, there is no baseline to compare against. The answer wasn't lost along the way. It was never built.
Spend isn't under control either. In that same McKinsey study, close to 20% of respondents report that the cost of running artificial intelligence, token consumption included, is already limiting how much they use it. There are companies throttling usage because of an invoice nobody budgeted for and nobody is reviewing.
One clarification about the most repeated figure of the year, the one that says 95% of AI projects fail. It comes from the 2025 MIT NANDA study and says something more specific: 95% of generative AI pilots showed no measurable impact on the income statement. It does not say the technology doesn't work. It says nobody measured. That distinction changes entirely what needs fixing.
The three questions that can be answered
A board doesn't expect a perfect return figure. It expects evidence that someone is in charge. There are three questions an organization with artificial intelligence under control answers in minutes, and that most organizations can't answer in weeks.
Which tools are running, and who approved them. Not the corporate license list, the real one, including what came in on a department's card and what someone uses from a personal account. If that list doesn't exist, any return calculation is incomplete from the first line.
What it was compared against. How long that process used to take, what it cost, at what error rate. If the prior baseline was never captured, there is no way to demonstrate improvement today, even when the improvement is real and the whole operation can feel it.
Who answers for each one. Not who uses it. Who decides whether it stays, expands, or gets shut off. When the answer is an entire department rather than a person, in practice there is nobody.
All three are management questions, not technical ones. That's where it gets uncomfortable.
What's actually missing
What follows doesn't come from any one audit. It comes from watching the same conversation repeat across companies that look nothing alike, in different sectors and at different levels of maturity. The pattern is remarkably consistent.
When a company buys artificial intelligence, it buys a tool and assumes management comes with it. It doesn't. The vendor delivers access, basic training and support. What the vendor does not deliver is the decision about who authorizes, who reviews consumption, who defines what success is measured against, and who has the authority to shut down something that isn't working.
That is what managing artificial intelligence means, and it's the part nobody took on when the contract was signed. It isn't an exotic discipline. It's the same thing your company has done for years with fleet vehicles, with access to critical systems, or with corporate card spend. Nobody argues those things need an owner, rules and periodic review. Artificial intelligence arrived fast, arrived through several doors at once, and skipped that step.
The silence in that meeting isn't a failure of the technology or of the teams using it. It's the result of buying a capability without assigning anyone to manage it.
The next board meeting is already on the calendar. Before that one, there's a simpler question that requires no budget, no project and nobody's permission.
If you asked right now for the list of every artificial intelligence tool your company is running, with the name of the person who answers for each one, how long would it take them to put it together?
The speed of that answer says more than any return figure.
Sources
- PwC. 29th Annual Global CEO Survey. January 2026. Survey of 4,454 chief executives across 95 countries and territories. View study
- McKinsey & Company. The state of AI in 2026: On the road to ROI. 25 August 2026. Global survey of 1,719 participants, fielded 4 May to 8 June 2026. View study
- Data & AI Leadership Exchange (Randy Bean). 15th Annual AI & Data Leadership Executive Benchmark Survey. January 2026. Invitation-only participation across roughly 110 Fortune 1000 companies and global brands. View study
- MIT Project NANDA. Study on generative artificial intelligence in enterprises, 2025.