A couple of years ago, most conversations about AI in the workplace started and ended with time saved. Roll out a Copilot, save everyone twenty minutes a day, multiply that across headcount, and celebrate the productivity gain. It’s a tidy story, and it made for an easy business case to get past finance. It’s also, in my experience running operations for a mid-sized UK professional services firm, the wrong one to lead with.
I say that having been fairly enthusiastic about the original pitch myself. We had the spreadsheet. We had the projected hours saved per employee, multiplied out across three hundred staff, turned into a headline number that looked impressive in the board pack. Eighteen months on, I’d write that business case very differently.
The twenty minutes rarely materialises the way the business case promised, and even when it does, it tends to evaporate into inbox management rather than anything the business actually notices. The real shift, when it happens, looks nothing like the pilot deck. It looks like people spending less time fighting systems and more time doing the parts of their job that are hard to automate: judgement calls, relationship management, the messy bits of a project that don’t fit neatly into a workflow.
Time Saved Is the Wrong Headline
Somewhere in year one, we stopped measuring “hours saved” and started measuring something closer to “friction removed.” It’s a subtler metric and a harder one to put in a board slide, but it’s far more honest. Ask a team where AI has genuinely changed their day, and they rarely mention the chatbot. They mention the report that used to take three people two days and now takes one person an afternoon. They mention the handover between departments that used to lose information every time and now doesn’t. Those are productivity gains too, just not the kind that show up in a usage dashboard.
See also: What online casino platforms can learn from better tech design
The Skills Conversation Nobody Wants to Have
The harder truth is that this shift exposes a skills gap that most L&D functions weren’t built for. It’s not a technical skills gap, mostly. Our people can learn to prompt a tool competently within a week. The gap is in judgement: knowing when to trust an AI-generated output and when to interrogate it, knowing which decisions still genuinely require a human in the loop, and knowing how to redesign a process around a colleague who happens to be a piece of software rather than a person.
None of that comes from a training video. It comes from managers who understand the technology well enough to coach their teams through it, and from a business that’s willing to redesign roles rather than just bolt AI onto the existing ones and hope for the best. That’s an uncomfortable amount of change management for what usually gets sold as a productivity tool rollout, and it’s the part most vendors quietly leave out of the pitch, because it isn’t theirs to solve.
Where the Bigger Shift Is Heading
It’s also, I think, where the broader industry conversation is finally heading. Microsoft’s own framing of what it calls the “Frontier Firm” treats productivity gains as something that comes from talent, process and technology working together, not from software alone — and Transparity has published a useful breakdown of how that productivity dimension is meant to work in practice, covering how AI-powered business applications are meant to connect people, data and automation rather than simply sitting on top of existing ways of working.
If I could go back and redo our own rollout, I’d spend less time in the first quarter talking about time saved, and more time talking to managers about what good judgement looks like when half their team’s tools can now think for themselves. The productivity gain was real in the end. It just didn’t look anything like the business case we originally wrote, and it took a lot longer to show up than the twenty-minutes-a-day pitch suggested it would.
If there’s one piece of advice I’d pass on to anyone starting this journey now, it’s to resist the urge to sell the easy number to your board. Sell the harder, truer one instead: that this is a people and process change wearing a technology costume, and it will take exactly as long as any other people and process change usually does. Anyone promising otherwise hasn’t run one yet.
