Using AI is now common; seeing it in your company's profit is not

Eight in ten respondents to McKinsey's 2026 survey say AI has improved their individual productivity. Only 37% credit AI with some effect on their organization's operating profit, almost the same share as a year earlier. The owner of a mid-sized company needs no study to recognize the gap between those two numbers: the team works faster, and the result somehow fails to show up.

Individual gains do not always reach profit

The McKinsey survey, published on August 25, 2026, gathered 1,719 participants from 97 countries between May 4 and June 8. Eighty percent say AI improved their individual productivity and 50% say it helps them decide better. Even so, 37% attribute some effect on EBIT (operating profit) to AI, and the group reporting a significant effect holds at around 6%. More organizations are scaling AI across the whole company, 44% versus 38% a year earlier, yet the share with an effect on profit barely moved.

Two caveats are in order. The sample includes a high proportion of large companies: 36% of participants work at organizations with revenues above $1 billion. And these are self-reported perceptions, not financial statements.

The same shape appears elsewhere. A working paper from the Federal Reserve Bank of Atlanta, published in March 2026 and based on a survey of nearly 750 corporate executives, describes a productivity paradox: perceived gains are larger than measured ones. According to the authors, that gap probably reflects a delay in revenue showing up.

What small U.S. businesses show

On October 8, economists at the Federal Reserve Bank of New York published an analysis of the 2025 Small Business Credit Survey, fielded between September and November of that year. Among firms with at least one employee, 46% said their business or employees were using AI tools, and another 15% planned to adopt them within twelve months.

The reported effects were modest. Seventy-seven percent saw no change in labor costs from using AI, and only 31% reported higher sales. AI users did expect more revenue: among them, the share expecting an increase exceeded the share expecting a decrease by 48 percentage points, against 21 among non-users. The authors note that whether those expectations come true remains to be tested.

These are U.S. data and do not describe Bolivian companies. They are useful for the shape they show, and the shape is already familiar: heavy use, high expectations and a result that takes time to appear.

Where the saved time goes

Picture a hypothetical distributor with three branches in Cochabamba. The administrative assistant uses AI to draft quotes and goes from one hour to twenty minutes per document. Each quote, however, still waits for the owner's approval on WhatsApp, and dispatch leaves when a truck is free. How many more orders were delivered that week?

Probably the same number. AI did its part and the process as a whole never changed speed, because when the bottleneck sits in approval or delivery, speeding up the quote only moves the wait elsewhere. The minutes gained scatter across five small tasks, nobody adds them up and the payroll stays the same. Each person feels more productive, and is right, but the company neither sells nor collects any differently.

The same happens in other trades. Imagine a restaurant with several locations where the manager uses AI to build the weekly supply order in ten minutes instead of forty. If the order still arrives late because the supplier bills at 30 days and cash is too thin to pay on time, those thirty minutes change neither the cost nor the service.

Saved time turns into money only when someone decides what it is for. That is a management decision, and it rarely happens on its own.

A cost worth adding up

There is another side to the ledger. In the same McKinsey survey, one in five respondents says their organization has limited its use of AI because of operating costs, which include per-use charges for the tools. When a company pays per-user subscriptions and also spends hours reviewing what the tool produces, the net saving may be smaller than imagined. It is worth running that calculation before taking the saving for granted. For a mid-sized company it fits on one sheet: what is paid per tool, plus the hours spent reviewing, set against the hours that were genuinely no longer spent.

The group that does see it in the books does something different

McKinsey identifies a small group, about 6% of participants, that attributes at least 5% of EBIT to AI and describes its impact as significant. Nearly three quarters of that group redesigned workflows because of AI, against one quarter of the rest. It is also twice as likely to have defined processes for measuring the impact of its initiatives.

This is an association, and the survey does not show that redesigning workflows causes the result. But the visible difference is not the tool, which is the same for everyone. It lies in what gets built around it: redefined steps, an owner and a number that is measured.

What to do on Monday

A spreadsheet and a conversation with your team are enough to start.

1. Pick one process that ends in money: quoting, collecting payment, restocking inventory or handling a complaint.
2. Write down today's starting point: how long it takes, how many times a week it runs, how many times it is redone and what money result it produces, such as quotes closed or payments collected on time.
3. Draw the steps and mark where the work waits: approvals, missing data, handoffs between people. Decide which step the AI will handle, which stays with a person and which disappears.
4. Name one person accountable for that number, even if it is the owner, and decide before starting what you will do with the time freed up. Put it in writing.
5. After thirty days, measure the same number from step 2. If the time dropped and the business number did not, the bottleneck is in another step, and that is where the next change goes.

These steps are done with what the company already has: half an hour of conversation with the team and the habit of looking at the same number twice.

One note on step 1: choose the process where you feel the delay, not the easiest one to automate. If today you do not know how long it takes, that is already a useful first answer, because without a starting point any improvement is only an impression. For the distributor in the example, the number could be quotes closed per week. For the restaurant, the cost of supplies as a percentage of sales.

The destination of the time can be the visit to five customers you have postponed for months, or collecting on Monday instead of Friday. What matters is that it exists before you measure, because saved time with no destination dissolves into the workday.

This year's surveys say that using AI is now commonplace and that seeing its effect in the accounts is not. In a mid-sized company, the owner closes that gap with one decision: which process to measure and what to do with the hours that get freed.