AI won't fix a broken process

AI won't fix a broken process

88 % of companies already use artificial intelligence in some function. 94 % say they get no significant value from that investment. Both figures come from the same McKinsey report from November 2025. They don't contradict each other: they describe what's actually happening. Almost universal adoption, almost zero return.

The interesting question isn't whether AI improves processes. With that level of adoption, the question is why so few companies notice it in their numbers.

Bolting a model on top changes nothing

Most AI projects do the same thing: they take a process that already worked badly and bolt a model on top of it. The process keeps working badly, now with an extra layer of complexity and a new cost.

There's one piece of data worth facing head-on. METR, an independent lab, ran a controlled trial with experienced software developers in early 2025. The result: they took 19 % longer on their tasks when using AI assistants. The striking part isn't the figure. It's that those same developers estimated they had been 20 % faster. They believed AI was speeding them up while it was slowing them down.

What the 6 % that actually make money with AI do differently

McKinsey separates a small group from the rest. They call them high performers: the 6 % of companies, those that attribute at least 5 % of their EBIT to AI.

They don't have better models. They have a different way of working. They pick a specific process, redesign it end to end, tie it to a business metric and measure the result. The model is the last piece they put in place, not the first.

Where the improvement actually shows up in the numbers

When the process is well designed, the data is good and it's real. In customer service, a field study with real agents measured 13.8 % more queries resolved per hour. The detail that matters: the biggest gains came from the least experienced employees, who leaned on the system's responses to resolve cases that used to stump them.

It fits with what we see in automating repetitive tasks: generating quotes, sales follow-up, answering frequent queries, reconciling data between programs that don't talk to each other. High volume, clear rules, a lot of wasted time. That's where the return is direct and can be calculated.

The Spanish case: the gap is the opportunity

In Spain, 21.1 % of companies with 10 or more employees were already using AI in their processes in 2024, according to the INE: nearly nine points more than the previous year. But the figure that really matters is the distribution.

Among small and medium-sized companies, adoption has tripled since 2022, according to the Hiscox report, but four out of ten still see no advantage in it. And here we need a figure that's often misused.

The Bank of Spain and Fundación Cotec found that companies using at least one AI technology have productivity that is 27 % higher on average. It sounds devastating. But it's correlation, not causation. It's very likely that the already more productive and better-organized companies are the ones adopting AI, rather than AI having made them productive. Anyone selling you that 27 % as a promise of returns is selling you smoke.

“

The advantage doesn't come from the model. It comes from having put the process in order before automating it.

”

Where to start so you don't end up in the 94 %

Four conditions, in this order:

1. A specific process, high-volume and repetitive. Not "the company". A process.
2. A measurable metric before touching anything: time per operation, error rate, cost, queries handled.
3. Redesign of the flow. If the process makes no sense without AI, it won't make sense with it either.
4. Real measurement afterwards, compared against the starting point. Not "a feeling of improvement".

At E2D we work this way because it's the only thing that delivers results: software built on each company's real process, not a generic product you have to bend your business around. AI is one piece inside that, never the headline.