Why AI isn't saving you money - and what to do differently

At the end of August, a McKinsey survey of 1,719 executives was published. The figure that grabbed the headlines: 94% of companies fail to attribute a change in operating profit to their investment in AI. Only 6% do.

But there is a second, less-quoted statistic that explains the first: 80% of employees report that they work faster thanks to AI.

Notice the gap. The employees feel the savings. The company doesn't see them.

This isn't a contradiction, nor is it proof that AI doesn't work. This is exactly what happens when you save time without deciding what to do with it.

The freed-up time doesn't disappear - it fills itself

Suppose you have an employee who spends an hour every morning typing orders from email into the system. He starts using a tool that does it in ten minutes.

You've saved fifty minutes a day. Where are they?

In most cases, they fill up on their own. He answers more emails, organizes things he didn't get to, takes a bit more time on each task. No one did anything wrong - there just wasn't a decision about what goes in instead.

For the savings to reach the profit line, someone needs to say: This time goes to calls with potential customers. Or: We aren't hiring for the next position. Or: Customer response time drops from two hours to half an hour, and that’s what we’ll measure.

This is a management decision, not a technological one. And that is exactly what the research found - among the 6% who do see a change, the difference isn't the tool, but that they changed the process.

Why this gap is bigger in a small business

Two reasons.

No measurement. A large company knows how much it costs to handle a request. In a business of eight people, no one has ever measured it, so even after improvement, there is nothing to compare it to. The savings exist, but they are invisible.

Processes aren't written anywhere. They are in the head of the person doing them. When you introduce a tool into a process that exists only in one person's head, one of two things happens: either the tool doesn't fit what actually happens, or everyone uses it differently. In both cases, there are no savings you can point to.

Three things that actually work, in order

Measure before. Even a rough estimate. How many hours a week go into the process you want to improve? How long does it take to answer a customer? If there is no "before" number, there won't be an "after" number, and any discussion about ROI will turn into a matter of feelings.

Choose one process that happens the same way every time. Not the most complex one, and not the one that annoys you the most. The boring, repetitive one: lead ingestion, invoice generation, meeting scheduling. The more consistent the process, the higher the chance the tool will actually stick long-term.

Decide in advance what goes in instead. Before the tool goes live, know what happens with the hour that gets freed up. If there is no answer, it will fill itself - and you're back to the 94%.

The other side of the argument

There is a counter-argument that deserves honest consideration: not every improvement must appear in the profit to be worth it.

If your response time to a customer dropped from two hours to twenty minutes, it likely affects closing deals, even if you don't know how to measure it. If your employee is less burned out because they aren't typing data four hours a week, that's worth something even without a line in the report.

This is true. But notice that even in these two examples, someone did decide something - to use the freed-up time to answer faster, or to reduce the load on a specific employee. The difference between the 6% and the 94% isn't whether you measured. It's whether you decided.

What to do this week

Pick one process. Write down how long it takes today - a rough estimate is enough. Then write one sentence: "If this is cut in half, the time will go to ____".

If you can't complete that sentence, it's not the time to bring in a tool. It's time to sit down and work on the process.

---

Source: McKinsey State of AI, survey of 1,719 executives, published August 26, 2024.