Has AI improved your bottom line? Only 37% say yes

Using artificial intelligence has become routine at the companies surveyed, but pointing to where it shows up in results is still rare. In an article published on November 9, 2026, McKinsey's Strategy & Corporate Finance Practice cites the same survey, in which only 37% of respondents attribute some EBIT impact to AI, essentially unchanged from the year before. The authors' explanation is less technological than it sounds. Projects start with the tool and only later look for a number to justify them.

What happened

Cristina Catania, Davide Grande, Holger Harreis and Sébastien Lacroix, with Andrea Tricoli and Pierfrancesco Nemola, write for CFOs, but the argument applies to anyone who signs off on a budget. Their starting point is a mismatch. In McKinsey's The State of AI 2026 survey, with 1,719 respondents in 97 countries polled in May and June 2026, 89% say their organizations regularly use AI in at least one function. Returns have not kept pace: the share seeing any EBIT impact barely moved from 2025 to 2026.

The article lays out three fronts. The first is setting ambitious, quantified targets tied to a business outcome, without waiting for perfect data. The second is redesigning entire processes instead of scattering isolated pilots. The third is investing in changing how teams work, with the CFO acting as sponsor and gatekeeper of value. The examples describe targets still in progress. A financial institution wants to cut its budgeting cycle from more than 120 days to fewer than 60, and a utility that recovered about 10% of what it lost to contract-invoice discrepancies is aiming for the full amount, more than €20 million.

Why it matters

The most useful detail in the article is the order of the questions. According to the authors, many companies begin by deciding which tool to use. In that order, the project tends to be judged by adoption, by how many people opened the system. When it starts with "which number needs to go down or up, and by when?", it gets an owner, a deadline and a way to be canceled if it doesn't deliver.

There is also a cost that usually stays off the books. The authors point out that AI consumes computing power, billed by usage, and hours of people reviewing what the machine produces. A cheap per-user subscription can hide a much larger total bill, and without a target no one notices that it isn't paying for itself.

Another warning targets the temptation to wait. One company cited in the article put AI on hold until its management-system replacement was finished and ended up two years behind. It is a single case, but the mechanism is familiar.

What changes for you

Owner. Treat every AI line in the budget as an investment, with an expected return and a date to check it. Also ask what will be switched off if the target isn't met. A project with no exit criterion tends to become a fixed expense.

Sales leader. AI sales tools arrive promising generic productivity. Swap the promise for one of your own metrics, such as proposal response time or conversion at a single stage of the funnel, and measure before and after.

Operations lead. The call to redesign processes end to end lands on your desk. Pick a workflow with a known loss, such as reconciliation, collections or invoice checks, and redesign who decides, who reviews and where the machine comes in.

What changes in Brazil

For a midsize Brazilian company, the most useful lesson is budget discipline, which does not depend on scale. Here it is common for AI to come in through a subscription, with each department signing up for its own, and it rarely goes through the committee that would approve a new machine. The article has no figures for the country, but a similar gap, between adopting technology and getting value from it, shows up in local data, such as the numbers showing Brazilian companies in the cloud and almost absent from data analytics.

Keep in mind, too, that at a 50-person company the finance department is usually two or three people. The "gatekeeper of value" role McKinsey assigns to the CFO ends up with the owner or the partner who handles the numbers. Adding that oversight to leadership's already full calendar takes a choice, the subject of another analysis on leadership time as a scarce resource.

The move

1. Take inventory of AI spending. List subscriptions, usage-based charges and projects underway, with the monthly cost and the department responsible. Many companies may find spending there that no one formally approved.

2. Require a target, a deadline and a process for each item. Every line needs to answer which number changes, by how much, by when and which routine will be redesigned. Anything without an answer in 30 days goes on the list to be cut.

3. Focus on one front at a time. The authors recommend investing in one promising domain at a time. Choose the process with the most visible loss, track the results for a quarter and only then expand.

The other side

The figures come from McKinsey's own survey, and McKinsey sells AI consulting; they measure what respondents report, not audited company results. They work as a snapshot of the mismatch and call for caution as a measurement. The company examples are anonymous and mostly describe targets, without showing whether they were met.

One could argue that demanding a quick return limits learning. Part of AI's value may lie in future capabilities that are hard to put in a spreadsheet today, and an overly rigid financial yardstick can steer a company away from cheap, useful experiments. Finally, the article comes from a consulting firm that sells exactly this kind of transformation, which leaves the reasoning standing and calls for a careful reading before turning the recipe into a project.