Here is the strange thing about AI at work right now. Almost everyone is using it, and almost no one is getting much out of it. In its 2025 State of AI survey, McKinsey found that 88% of organisations use AI in at least one function. In the same report, only a sliver could point to a real, company-wide effect on profit. Adoption is nearly universal. Impact is rare.
So what separates the few who get results from the many who do not? It is almost never the model. It is whether they changed how the work flows, or just bolted AI onto the process they already had.
Everyone adopted. Almost no one redesigned.
The same McKinsey survey puts a number on it: only about 21% of organisations using generative AI have redesigned even some of their workflows. Nearly four in five just layered AI on top of the old steps. And it shows up in the results. The teams McKinsey calls high performers are about three times more likely to have fundamentally redesigned how work happens, 55% of them versus 20% of everyone else. Meanwhile fewer than 10% of organisations are actually scaling AI agents in any function. The tools are everywhere; the rebuild is not.
Toggle between the two stories below. The same year looks like a triumph or a disappointment depending on which number you read.
The adoption–impact gap
Same year, two very different stories.
AI is everywhere now.
- Organisations using AI in at least one function0%
- Say they use generative AI0%
- Experimenting with AI agents somewhere0%
Nearly everyone has switched AI on. Roughly one in five has changed how the work actually flows. That space between the two is where the results are hiding.
Bolting on vs. redesigning, in one picture
A bolt-on keeps every manual hand-off and squeezes AI into one step in the middle. You get a slightly faster draft and then re-key it, reformat it, and check it by hand, exactly as before. A redesign asks a sharper question: if AI can do this step reliably, what does the whole process look like now? Usually it looks shorter. Flip the toggle and watch the steps collapse.
Bolt-on vs. redesign
Same task, two ways to add AI. Watch the steps.
AI is a stop in the middle. The manual hand-offs around it all survive, so the day barely gets shorter.
How to close the gap without a moonshot
Redesign sounds like a grand transformation project. It is not. The teams that pull ahead do something much more modest, one process at a time:
- Pick one painful, repeatable process. Not the whole company. One workflow that eats hours every week and follows clear rules.
- Map how it really works today, including the re-keying and the checking, then ask which steps exist only because a human was doing the middle.
- Rebuild the flow around what AI can now do reliably, and give a person the job of owning judgement and the edge cases, not the busywork.
- Measure the hours reclaimed, prove it on one process, then copy the pattern to the next. Momentum beats a big-bang rollout.
This is the heart of what we do in our Flow Automation work at NeuralYug: not sprinkling AI over a broken process, but redrawing the process so the automation has somewhere solid to stand. It is also why our automation projects tend to save real hours rather than produce a demo that impresses once and helps no one. If your AI pilot stalled, the model is probably not the problem. The workflow around it is.
Frequently asked
- Why do most AI projects fail to show results?
- Because the process around the AI never changed. If you add a smart tool to a workflow full of manual hand-offs, you get a slightly faster version of the same slow process. The gains come from redesigning the flow, not from a better model.
- What does redesigning a workflow actually mean?
- It means asking what the process should look like now that AI can do certain steps reliably, then removing the steps that only existed for a human. A person keeps judgement and edge cases; the routine flow gets rebuilt around the automation.
- Do we have to redesign everything at once?
- No, and you should not. Start with one painful, repeatable process, prove the hours saved, then copy the pattern. One redesigned workflow that works beats a company-wide programme that stalls.
