Most big organisations are in exactly the same place with AI right now.
I like to call it the “policies, training, and Copilot” phase.
You publish an acceptable-use policy. You run a few training sessions. You buy everyone a Copilot licence. Then you wait for the transformation.
But it doesn’t come.
Recently, MIT reviewed more than 300 enterprise AI efforts and found that 95% had produced no measurable return. None.
Gallup, meanwhile, reckons only about one in ten employees uses AI at work every day.
The tools are everywhere, but the results are nowhere to be seen.
The thing is: this isn’t a technology problem. It’s a design problem. And People Teams are better placed to fix it than almost anyone else in the building.
The Trap of the Personal Productivity Tool
The default way to “integrate AI” is to treat it as a personal productivity tool.
Give everyone a clever assistant, teach them to prompt it, and then sit back and relax as each person gets a little faster at their own tasks until — hey presto — transformation.
Unfortunately, that’s not what happens.
And not for the reason you’d expect.
Most valuable work doesn’t happen inside one person’s head. It flows between people. Things like reference requests, onboarding programmes, right-to-work checks, grievances — each passes through several pairs of hands before it’s done.
Your organisation is a pinball machine
Picture those workflows as a pinball machine.
A piece of work is the ball. The flippers are your employees.
That ball pings from flipper to flipper — someone drafts, someone approves, someone chases, someone signs off. The value of that work gets created as the work travels between people, not while it sits with any one of them.
Now supercharge a single flipper. Make one person 100 times more productive and leave the rest of the machine exactly as it was. You haven’t sped up the game. You’ve just got one flipper hammering the ball in every direction while everyone else scrambles to keep up.
That’s what handing out personal productivity tools actually does.
And that’s mostly because people will adopt it unevenly. You’ll get a couple of super-users, a few occasional dabblers, and — this is the part leaders miss — a chunk of people who refuse to engage entirely.
One 2026 survey found nearly a third of employees admit to actively undermining their employer’s AI push. Among the youngest workers, it was closer to half.
So the gains don’t spread. They concentrate at the fastest flippers, who then batter everyone downstream.
Recently, I was talking to an area manager who was thrilled with what AI had done for his productivity. He was sending more emails than ever.
Then I spoke to one of his restaurant managers, who saw it differently: “He used to send me three emails a week. Now I get 200.”
Personal productivity AI, deployed like this, doesn’t remove work. It just moves work, and usually onto the person least able to absorb it.
Stop speeding up the work. Get rid of it.
By all means, use tools that make individuals more productive.
Just keep in mind: that’s the small prize.
The big prize is re-engineering how the work happens so the routine admin disappears altogether.
I’m not talking about a Copilot that helps someone write an employment reference faster. I’m talking about an agent that pulls the data, drafts the letter, and hands it to the employee on full self-service, so the request never reaches the People Team at all.
Not a nudge reminding an administrator to chase right-to-work renewals — an agent that chases every employee automatically and only escalates the 10% of cases that genuinely need a human.
That’s the difference between speeding up the work and getting rid of it. One saves a few minutes, the other removes whole categories of admin nobody enjoyed in the first place.
According to MIT, 95% of GenAI pilots fail. And the 5% that do work all share one trait: the AI was built into a workflow and adapted to it, rather than bolted onto individuals and left to sink or swim.
The winners re-engineered. The losers bought licences.
For deskless teams, this is the only model that works
If your people are on the frontline, the personal productivity model was never going to fly anyway.
A care worker, a chef, a warehouse operative — they aren’t going to craft prompts between shifts. And expecting them to become skilled AI users is a category error.
So flip it.
Instead of giving frontline teams a tool and just hoping they learn to use it, build them something that actually works for them. Make the experience so the intelligence is baked in and the adoption effort is designed out.
The employee asks for what they need, in plain language, on their phone, and an agent does the rest. No prompting, no inference, no training course — just a better experience that happens to run on AI underneath.
That’s not the same job as writing an AI policy. It’s product thinking. And it belongs to the People Team.
The real work is deciding what to automate
None of this happens by accident, and it definitely doesn’t happen by buying everyone a licence and hoping it sorts itself out.
It takes people leaders willing to look hard at how their workflows and employee experiences actually run, and to make deliberate calls.
What should be automated completely? Where can agents safely hand off to one another, and where does a human genuinely need to stay in the loop?
Get that right and you don’t just make your team quicker. You take the routine admin off their plate entirely and free them for the work that actually moves the business.
Because the organisations that win out won’t be the ones that adopted AI fastest. They’ll be the ones that were most deliberate about what they stopped doing.
Copilot for everyone feels like progress. But mostly it’s the road to hell, paved with good intentions and unread AI-generated emails.