AI at Work: Why More Tools Usually Means Less Done

You’ve probably heard the pitch: AI will free your team from busywork, sharpen your decisions, and make everyone more productive. The reality, for most small businesses, is a bit more complicated.

Recent research found that nearly nine out of ten companies report no measurable improvement in productivity or employment from AI over the past three years. Some are seeing things get worse.

Too many tools, not enough thinking

A Boston Consulting Group study tracked over 1,400 workers and found something worth sitting with: people were more productive when using three or fewer AI tools. Once they hit four or more, productivity dropped. Workers reported more mistakes, more confusion, and the feeling that they were running faster but covering less ground.

There’s also the problem of what’s being called “workslop” — content that looks finished but isn’t. A finance worker receives an AI-drafted report that’s technically coherent but wrong in the details. Now someone has to rewrite it. A retail manager follows up on incomplete information that a colleague had AI-generated in a hurry. According to research cited by Entrepreneur, around 40% of the time AI saves gets spent fixing its mistakes.

And there’s another catch: workers using AI often don’t work less. They work faster, take on more, and quietly burn out.

What’s actually working

The businesses seeing genuine gains aren’t throwing AI at everything. They’re being deliberate. The pattern that keeps coming up:

• Start with one specific problem. Not “use more AI” as a goal. Something concrete: slow email turnaround, reports that eat your admin’s Friday afternoon, meeting notes nobody reads.
• Set clear boundaries before you start. What information can go into the tool? Who checks the output before it goes anywhere? When should AI draft, and when should it stay out of the way?
• Train your team on what good looks like. Workslop happens when people don’t know how to spot it or push back on it.
• Measure what actually matters. Not hours “saved” in theory — whether your team is doing higher-value work, and whether anyone’s quietly drowning. Microsoft 365 Copilot is a reasonable place to start for many businesses, because it sits inside tools your team already uses: Word, Excel, Teams, Outlook. It handles the scaffolding — a first draft of a proposal, a summary of a long meeting, an anomaly in a spreadsheet — and your people take it from there. That’s the model that tends to work: AI does the rough work, humans do the thinking.

A 10-person law firm in Bangor can use it to summarise lengthy documents faster. A construction business can use it to draft site reports. An accountancy practice in Wrexham can use it to surface patterns in client data. But only if someone’s still responsible for the output.

The honest answer

If AI feels chaotic or unproductive right now, the answer is almost never to add more tools. Pick fewer, use them more deliberately, and keep the judgment firmly with your team.

Worth knowing about. Whether it changes anything for your business depends on what you’re actually trying to solve.