McKinsey's 2026 State of AI survey, as reported in secondary coverage, found that nearly nine in ten organisations now use AI regularly. Only 37% say it adds anything to their profit.
That gap tells you most of what you need to know about AI in business right now. Using it is easy. Making money from it is a different skill.
Wide, but shallow
The same coverage reports that 44% of organisations are scaling AI across the business, up from 38% a year earlier. That is real progress, but it also means more than half are still experimenting: a chatbot here, a writing tool there, a pilot that never quite becomes "how we work".
There is a size gap too. Large companies are pulling ahead with AI agents, with around 40% scaling them, while smaller firms have stayed flat at about 22%.
If you run a small business, that can sound discouraging. It shouldn't. Large companies have more processes, more data and more budget for pilots. Small companies have something more useful: fewer layers between deciding and doing.
Why most AI projects don't pay
From the outside, the businesses that profit from AI and the ones that don't look similar. They often use the same tools. The difference is in what happens around the tool.
They add AI on top of a messy process. If a quote takes four days because it passes through three inboxes, a faster writing tool saves twenty minutes, not three days.
They never measured the "before". Without a baseline, nobody can say whether it worked, so the project quietly fades.
They spread thin. Ten small experiments with no owner produce ten nice demos and no change to the numbers.
They automate the wrong thing. The task that is fun to automate is rarely the task that costs the most.
What the profitable group does differently
The organisations seeing returns tend to treat AI as a change to how work flows, not as a new app. In practice that looks like:
- Starting from a costly problem, not from a tool. "We lose leads because we reply too slowly" beats "we should use ChatGPT".
- Redesigning the process first. Remove the steps that shouldn't exist, then automate what remains.
- Putting a number on it. Hours per week, response time, conversion rate, error rate. Pick one.
- Giving it an owner, someone whose job includes making it work.
- Scaling only what works. One proven workflow becomes the template for the next.
A practical plan for a small business
You can run this in a month without hiring anyone.
Week 1: find the leak. List the tasks your team repeats every week. Next to each, estimate the hours spent and what slow or wrong handling costs you. The top two or three are your candidates.
Week 2: map one process. Write down every step of the top candidate, including who does it, which tools are involved and where it gets stuck.
Week 3: build the smallest version. Automate the boring middle and keep a person approving the output. Use the tools you already pay for where you can.
Week 4: measure and decide. Compare against your baseline. If it saves meaningful time or wins more work, keep it and pick the next process. If not, adjust or drop it. Both are fine outcomes.
The point
AI on its own doesn't make a business more profitable, just as a new website doesn't on its own bring in customers. What changes the numbers is a better process, measured honestly, with AI doing the repetitive part.
The good news for smaller businesses is that this is exactly where being small helps. You can decide on Monday and change how you work by Friday.
