“We rolled out AI and nothing changed”

Written By Rafaela Heming

If your company rolled out AI and productivity didn't move, the most likely cause is not the tool. It's that people got access without a work-specific habit, and access alone produces nothing. The research is unusually clear on this: almost every organisation that bought AI saw no measurable return, and the small group that did had one thing in common. They picked a narrow, real task and built around it.

I'm Rafaela Buena. I run AI workshops for teams in Amsterdam, and this is the sentence I hear most often from the person who signed the invoice. Usually said quietly, usually with a bit of embarrassment. There's nothing to be embarrassed about. You're in the majority.

The number that should reassure you

MIT's Project NANDA published "The GenAI Divide: State of AI in Business 2025." Based on interviews with executives, employee surveys and a review of 300 public AI deployments, it found that about 95% of enterprise generative AI pilots delivered no measurable impact on profit and loss. Only about 5% created significant value. (The report is preliminary and not peer-reviewed, and the authors say so. But the pattern has been repeated everywhere since.)

A 2026 survey by WRITER and Workplace Intelligence of 1,200 executives and 1,200 non-technical employees found that three-quarters of executives admit their company's AI strategy is more for show than substance.

So no. It isn't just you.

Why AI isn't improving productivity: the learning gap

MIT's lead author, Aditya Challapally, told Fortune the core issue was not the quality of the models but what the report calls a "learning gap." Generic tools like ChatGPT work brilliantly for individuals because they're flexible. They stall inside organisations because nobody connected them to the actual workflows, and nobody taught the workflows to the tool.

Put plainly: the company bought a licence and sent a Slack message. That's a rollout in name only.

Here's what that looks like on the ground. Someone opens the tool, types a vague question, gets a bland answer, closes it, and decides it's overhyped. Multiply by forty people. Now you have forty licences and a story that "AI doesn't really work for us."

What the 5% did differently

The MIT report and the interviews around it point to a few consistent moves.

  • They picked one pain point. Not "transform the business." One report that takes four hours every Monday. One inbox category. One recurring document.

  • They built around a real workflow, not around the tool. The question was "what do we actually do here?" before "what can the tool do?"

  • They didn't build from scratch. In MIT's interview sample, externally sourced tools reached deployment roughly twice as often as internally built ones. Self-reported, and correlation only, but a consistent finding.

  • They treated it as a learning problem, not a software problem. That's the bit most companies skip.

This is also, not coincidentally, how I run team workshops. Before the session I collect the tasks your people actually do, in their words, and we build the whole day around those. Nobody learns "AI." They learn how to get Tuesday's report done in forty minutes.

The half-hour test

Before you spend another euro, try this.

Pick three people who were given AI access. Ask each to name one task they now do with it, every week, that they didn't before. Not "I use it sometimes." A named, recurring task.

If all three can answer, your rollout is working and you're just not measuring it. If none can, you don't have an adoption problem. You have a "we never showed anyone what to do with it" problem. That's cheaper to fix than you think, and it isn't fixed by buying a different tool.

What "changed" should look like

Not a dashboard. Small, boring, visible things.

The weekly report goes out on Tuesday instead of Thursday. The person who used to dread client follow-ups does them in a batch on Friday morning. Meeting notes appear without someone staying late. The ops lead stops asking "who has the latest version?"

Those add up to hours a week. Hours a week add up to headcount you don't have to hire. That's the return. It just doesn't arrive as a line item until someone goes looking.

FAQ

Why is AI not improving productivity at my company?

Most often because people received access without a task-specific habit. Research from MIT found the failure is a learning gap, not a technology gap. People need to be shown what to do with AI on their own recurring work.

How long should it take to see results from AI at work?

For a single well-chosen task, days. For a team habit, a few weeks of consistent use. If nothing has changed after three months, the problem is almost certainly in how it was introduced, not in the tool.

Should we switch AI tools if we're not seeing results?

Rarely. Switching tools without changing how people are taught to use them reproduces the same result with a new logo. Fix the learning problem first.

If you're the person who signed the invoice and you'd like the second attempt to work, that's exactly what a team workshop is for. Built on your team's real tasks, so by the end of the day there's a named change, not a licence.

Sources

Rafaela Heming
Next

AI hallucinations: why your chatbot lies and how to catch it