McKinsey's AI ROI Numbers and What They Mean for Creative Teams

Every few months a survey lands that is mostly a mood check, and every few months one lands with numbers sharp enough to plan around. The McKinsey State of AI survey belongs to the second category, even though its headline finding is that almost nothing moved.

The numbers that stayed flat

About 37% of respondents said AI has at least some influence on their company's operating profit, essentially unchanged from the previous year. The survey covered 1,719 professionals and executives across 97 countries. The share McKinsey classifies as AI frontrunners, companies attributing at least 5% of EBIT to AI and calling the impact substantial, also held flat at 6%.

Flat profit attribution would normally suggest stalled adoption. The rest of the data says the opposite. Nearly nine in ten respondents use AI regularly in at least one business function, and the share using it in three or more functions rose from 51% to 56%. Spending keeps climbing: 28% of respondents say their companies put more than 10% of the total IT budget into AI, 60% expect AI investment to rise over the next year, and chatbots are the most common tool, used company-wide by 47%.

So the honest summary is: more companies, spending more, in more functions, with profit attribution stuck at the same number for two years running.

Where the money actually shows up

The profit that does register is unevenly distributed. Revenue increases, where they exist at all, cluster in marketing and sales, followed by product and service development and software development. Cost savings show up earlier and in less glamorous places: supply chain management and service operations were the functions respondents named most often when profit had not yet moved.

That pattern should feel familiar to anyone doing commercial image or video work. Marketing is where AI content output plugs directly into a measurable funnel, so the money is easiest to trace there. Internal efficiency is real but shows up as capacity, not as a line item.

One number in the German coverage deserves more attention than it has gotten: 32% of respondents said their company decided against buying one or more software products or features because agentic coding tools let them build it internally instead. That is a third of surveyed organizations choosing to build rather than buy at least once. It is not proof that a third of any product's customers can rebuild it in an afternoon, but it does mean "we could build this ourselves" is now a live objection in the buying conversation, and the era of selling thin wrappers at seat-based prices is closing as that objection spreads. The pressure hits small creative SaaS hardest, both as vendors and as buyers.

The gap that should actually change your behavior

Two more numbers, side by side: 80% of respondents said AI improved their individual productivity, and 50% said it helps them make better decisions. Meanwhile, only 37% can point to any operating-profit effect at the company level.

Individuals feel the gain. Companies cannot find it on the income statement. Both can be true, and the resolution is mostly that the gains are real but small per person, and they evaporate into work that would not otherwise have been done at all. A designer who generates twelve concept directions instead of three has not become four times cheaper. They have produced more directions, and the client still buys one.

There is a darker reading of the same gap. 47% of mid-level managers and staff reported experiencing at least one negative effect of AI, against only 31% of senior leadership. The people closest to the work are also the people absorbing its failure modes, rework, review burden, prompt maintenance, and they are reporting it upward less and less. Meanwhile 39% of respondents expect AI-driven headcount reductions.

If you run a creative team, this is the operating condition, not a bug to fix: every member of your team feels faster, your costs have not gone down, and some of your people are quietly carrying the tax that makes the speed possible.

What to actually do with this

Three practical moves fall straight out of the numbers.

First, put AI where attribution already works. Marketing and sales collateral, product visualization, campaign variants, the categories the survey says show revenue first. This is also where AI image generation is genuinely production-ready today: same-day concept exploration, localized ad variants, storefront and product imagery at volume.

Second, treat internal tooling as a build decision, not a buy decision. With a third of companies building instead of buying, and agentic coding tools making small internal tools cheap, the default answer to "should we get a tool for this" is now "we can probably make it." The skill worth developing in-house is specifying the tool tightly enough that building it takes hours, not weeks.

Third, measure capacity instead of productivity. The survey's productivity numbers are self-reported feelings, and they will lie to you in both directions. The number that survives contact with accounting is throughput: assets per week per person at a held quality bar. If output volume is not up, the productivity gain went somewhere invisible, usually into wider exploration at the same final volume. That is a legitimate use, but it should be a choice.

The boring conclusion

McKinsey's own framing is that companies are "on the road to ROI," which is consultant-speak for not there yet. The flatter interpretation is that AI adoption is turning out like electricity or cloud: obvious in hindsight, slow on the income statement, and much easier to see in the capabilities a company has than in its quarterly margin.

For creative teams specifically, the practical takeaway is not to wait for a productivity windfall, because the survey says your competitors are not getting one either. The takeaway is that the distance between feeling faster and being cheaper is a management problem, and the teams that close it will be the ones measuring what comes out rather than how fast everyone feels.