Is AI actually making us more money?

Think we Israelis are on edge? Hundreds of millions of investors are nervously watching the US stock markets right now.

Everyone is trying to work out one thing: when does it burst? Now? Tomorrow? Next year? Is the AI frenzy that has inflated share prices, stock indices and retirement savings a bubble of overblown expectations, about to collapse?

In the meantime the hype and the share prices have soared, although a certain cooling has set in over recent months. Even so, Anthropic is preparing to go public, with some investors valuing it at around $1.5 trillion.

This entire industry needs a solid and simple business foundation:

Business + AI = bigger business.

If that doesn't happen, all those valuations are hard to justify.

I know, it's an odd question for me to ask, here in the fall of 2026. Surely it's obvious?

I've been through a pile of studies. Some measured feelings, others numbers. Allow me the short version: it is far from obvious.

Plenty of managers and employees feel it absolutely is. 80% of the 1,007 managers who participated in a Wharton survey published in October 2026 reported a positive return on their investment so far, and 28% reported a strongly positive one. These are self-reported returns, rather than independently verified financial results.

In departments that substantially reworked their processes around AI - AI fully embedded, agents in use, and significantly changed workflows - 67% reported strong returns.

I feel the same. It handles countless tasks for me. But I can't honestly say it is directly making me more money.

Another study looked at 7,000 workers, 80% of whom used AI and reportedly saved two hours on email a week. They also work less outside office hours and from home, which is wonderful. But they don't produce more. Interesting. That's what you get from handing AI tools to individual employees rather than deeply changing how a company works.

In an experiment with 776 employees at Procter & Gamble, the world's largest consumer goods company (you have at least five of their products at home right now), people were assigned to work either alone or in pairs. The headline finding: individuals working alone with AI performed about as well as teams without it. AI helped people step outside the boundaries of their own specialism.

Without AI, everyone proposed solutions from their own field: researchers talked research, marketers talked marketing. With AI, the proposals suddenly became broader and reached into other areas, whatever the employee's background. In other words, the benefit wasn't only speed or nicer wording, but a wider range of professional thinking. That doesn't mean workers can be replaced, and it doesn't mean AI has turned into more successful products or more revenue.

In 2025, 16 experienced software developers were given real tasks, some with AI and some without. By the end of the experiment they were convinced AI had sped them up by 20 per cent. The clock showed that with AI they had in fact worked more slowly, by almost 20 per cent.

Another study, covering half a million developers, found that AI writes an awful lot of code for them: 240% more coding activity was recorded. The amount of real work that reached the market as software releases, however, rose by only 30%. Nice, but also a great deal of unnecessary noise.

A study of customer service followed the staggered rollout of an AI assistant among 5,172 support agents. It measured one clear operational metric: issues resolved per hour worked. That number rose by 15%. Lovely! But note this: the biggest gains went to less experienced, lower-skilled agents, who improved in both speed and quality. Among the most experienced and skilled, there were small gains in speed alongside small dips in quality. The researchers also found evidence of learning and of improved English, especially among agents outside the US. The benefit was particularly large for moderately rare problems: the kind an agent knows less well, but on which the system has enough data.

Bottom line: more queries resolved in every working hour. The customer experience improved too, with customers more polite and less likely to ask for a manager. Still, 15% more resolutions per hour doesn't necessarily mean 15% lower costs or 15% more profit. Nor should anyone expect the same benefit from every employee.

In short, the people and organisations that manage to genuinely rebuild the way they work around AI, and are willing to get it wrong, try again and again, and invest in it, may well profit. Those who simply bought subscriptions and still work the same way? Mostly they get a warm feeling and slightly better-worded emails.