Nearly nine in ten organisations now use AI regularly in at least one function. Far fewer can find it in their earnings. McKinsey's State of AI 2026 finds that 37% of respondents report any positive EBIT impact from AI, and only about 6% qualify as high performers: respondents who attribute 5% or more of EBIT to AI and report significant value from it.
That is the AI value gap in three readings from the same survey series. Regular use is close to universal among respondents. Measurable earnings impact is much rarer, and meaningful earnings impact (the high-performer bar) is rare.
If you lead a business, you have probably felt this gap before you saw it in a chart. The licences are bought, the pilots are running, people say they like the tools, and yet the quarterly numbers look much as they did before. The board asks what the AI budget is returning, and the honest answer is a list of activities rather than a line on the P&L.
This piece is a map of that gap: what the numbers measure, where the value leaks out between using AI and earning from it, and what the small group on the far side does differently. It is not a vendor list, and it will not end with a recommendation to buy another tool. The argument is simpler and less comfortable than that. Most organisations are adding AI to work that was designed for a world without it, and then wondering why the work does not change.
What the numbers actually measure
Before drawing conclusions, it is worth being precise about what each figure counts, because headline AI statistics are easy to stretch.
All three headline figures above come from McKinsey's State of AI survey series. The 2026 report puts regular organisational AI use at nearly nine in ten (use in at least one business function). That is a low bar on purpose. One team using a model to draft marketing copy counts the same as a company that has rebuilt its operations around AI. It tells you AI is present, not that it matters. (An earlier McKinsey 2025 figure of 88%, republished in Stanford HAI's AI Index 2026, sits in the same ballpark; this piece sticks to the 2026 survey so the comparison is within one year.)
The 37% and the ~6% are also from McKinsey's State of AI 2026. The 37% are respondents who report any positive impact on EBIT from AI, however small. The ~6% are what McKinsey calls high performers: respondents who attribute 5% or more of EBIT to AI and report significant value from their AI use. That is attribution by survey respondents, not an audited organisational fact — and it is the closest thing in the data to AI showing up in earnings in a way a CFO would recognise.
One caveat belongs here in plain sight. These are survey answers, not financial restatements. But the direction is consistent across the McKinsey series, and it matches what most leaders see inside their own companies: a wide base of usage that narrows sharply by the time it reaches the income statement.
Productivity is not P&L
The most common response to the value gap is that the benefits are real but hard to measure. There is some truth in that, and the McKinsey data shows it. Around 80% of respondents say AI has improved individual productivity, and around half say it has improved the quality of decisions. People are getting real help from these tools.
But individual productivity is not the same thing as enterprise financial impact. McKinsey's data shows the share of organisations reporting enterprise-level financial impact roughly flat compared with the prior year, even as more organisations moved beyond pilots and scaled their AI use. More scale, broadly the same earnings picture.
The reason is not mysterious. When an analyst saves an hour a day writing reports, that hour does not automatically turn into revenue or lower cost. It goes somewhere: more reports, longer meetings, a slightly earlier finish, a little more polish on work nobody asked to be polished. Unless someone decides what the freed-up capacity is for, and changes the process, the headcount, the targets or the service model to capture it, the gain stays personal and dissolves into the day.
Where does the saved time go?
Captured: changes cost, capacity or revenueDissolves into the day
This is why personal productivity gains can hide organisational inertia. Every survey of users looks encouraging. Every review of the P&L looks unchanged. Both are true at the same time, because the unit that improved is the individual task, while the unit that earns money is the workflow, the team and the business model.
What high performers do differently
One of the biggest differences between the ~6% and everyone else is whether they change the work. High performers also tend to spend more on AI and use a wider range of technologies — so this is not a claim that budget and tools do not matter. According to McKinsey, nearly three in four high performers say they have fundamentally redesigned workflows because of their AI use, compared with about one in four other respondents.
Redesign here means more than adding an assistant to an existing step. It means asking what the process should look like now that some of its steps can be done by software: which handoffs disappear, which approvals no longer need a person, where a human adds judgement rather than effort, and what the team does with the time that comes back. It often means fewer steps, not smarter steps.
High performers also aim wider. They pursue growth and innovation alongside efficiency, rather than treating AI purely as a cost-cutting exercise. A process that is only made cheaper has a ceiling; a process that lets you serve customers you could not serve before, or launch something you could not have staffed, does not.
McKinsey also finds high performers are about twice as likely to report senior commitment to AI and to have defined processes for measuring its impact. That rigour is unglamorous, but it is what turns a promising pilot into something that can show up on the income statement. Separately, our recommendation for buyers is to treat AI like any other operating investment: named owners, outcome targets, and a clear view of what success looks like before the next licence is bought.
What high performers mostly do not do is bolt more tools onto old processes and call that transformation. The organisations still waiting for AI to show up in earnings often already have plenty of tools. What they lack is a decision about which work should change.
What this means if you are buying
For leaders deciding where the next AI budget goes, the implication is a change in the order of questions. The usual sequence starts with tools and partners: which platform, which model, which implementation firm. The data suggests starting one step earlier.
- Pick the workflows that should change first. Name two or three processes where a redesign would move revenue, cost or speed in a way the P&L would show.
- Define what changes, not what gets added. Which steps, handoffs and approvals go away, and what the people involved do instead.
- Decide how the gain is captured. Capacity, cycle time, cost per task or new revenue, with an owner and a target for each.
- Then choose tools and partners. Judge them on whether they can deliver that redesign, not on the length of their feature list.
This order changes the conversation with vendors and implementation partners. Instead of asking what a platform can do, you ask whether it can help you deliver a specific change to a specific workflow, and how you will both know it worked. Partners that can answer that well are rarer than partners with an impressive demo, and the difference matters more than any feature comparison.
It also keeps the budget honest. A redesign with a named owner and an earnings target is much easier to evaluate, stop or scale than a portfolio of pilots justified by user satisfaction.
This is the gap Scayl exists to make clearer. Scayl is a hub for AI implementation: sourced data on where AI spend is and is not paying off, and practical context on how implementation actually works, so that leaders can make these decisions with evidence rather than vendor claims.
Redesign the work, then buy
The AI value gap is not a technology shortage. Regular use is already close to universal among surveyed organisations, and individual productivity gains are real. The gap sits between those gains and the income statement, and the organisations that have crossed it mostly did so by redesigning how work gets done — not by treating another tool as the whole answer.
If your AI spend is not yet showing up in earnings, the most useful next step is probably not another licence. It is a short list of workflows you are prepared to change, and a clear view of what success looks like for each.
The sourced charts behind this piece are on scayl.ai. When you get to choosing an implementation partner, our upcoming guide on how to choose one picks up where this leaves off.
Sources
- McKinsey, The state of AI (2026)
- Stanford HAI, AI Index Report 2026
- Scayl, AI implementation hub
