What Efficiency Actually Looks Like

I have too many cables. USBs, HDMIs, a graveyard of wired AirPods that mostly don't work anymore.

I don't collect them on purpose. Every time I can't find a cable, I buy another one. Now there's a box at the top of my cupboard with dozens of tangled cables in it, and I still can't find the right one when something needs charging.

You'd call this Marie Kondo's nightmare. Grab a rubbish bag. Start throwing cables out until the number feels reasonable.

That gets you fewer cables, but it doesn't get you a cupboard where you can find the one you need.

For most organisations right now, efficiency means fewer people, and AI has become the justification for arriving at that number. "AI-driven efficiency" sounds better than "we needed to hit a budget target," so it's the AI tagline that ends up in the announcement.

Efficiency isn't about headcount, though. It's about friction.

Cut ten people from a team and approvals still takes four days, because the approval chain was never about headcount. It's about a process someone built in 2015 that nobody's owned since. The issue is the missing decision-maker, not a missing person.

Removing people who weren't causing friction doesn't remove the friction. It just means fewer people managing the same amount of it. That shows up a few months later as burnout, then attrition, then, often, a hiring plan to replace the people who left, doing the same friction-heavy work under a new title.

Headcount is easy to see. It's a number on a slide. Friction isn't. It's measured in how long approvals take, and how many meetings exist to compensate for a decision nobody's empowered to make. None of that shows up on an org chart. 

What the org chart doesn't tell you.

Ask what happened to the work after the re-org.

If it got redesigned, reallocated, automated, or correctly identified as work nobody needed to be doing, that's efficiency. If it stayed in the same place and is delivered by the same team, that's a cut. 

Gartner's research into customer service functions found something interesting about this. Where AI tools were introduced, most organisations kept staffing stable and handled more volume. The tools were absorbing the extra work, not extra people. Gartner expects roughly half of those post-cut organisations to rehire by 2027 (often under a different job title), which is an expensive way to find out the cuts were premature.

I used to think this was mostly semantics: the story leaders told to give a restructure announcement the right rationale. I don't think that anymore.

Wayne Cascio tracked downsizing outcomes across thousands of companies over several decades and found headcount cuts don't reliably improve financial performance. The correlation with better outcomes is much less dramatic: redesign how the work actually gets done.

That's 1993 research, and it still holds. A 2003 follow-up tracked over 6,000 instances of workforce change across the S&P 500; the finding didn't move. A 2024 meta-analysis revisited the whole body of downsizing research and landed in the same place again. Three decades, thousands of companies, and the correlation between headcount cuts and improved performance still hasn't turned up. At some point you'd think someone would stop looking for it.

If AI were replacing labour at scale, you'd expect that to show up in the productivity figures. Mostly, it isn't. Companies that announced layoffs in 2025 named AI as the reason for just 5% of them. The other 95% had other ordinary reasons, the kind that don't make it into a press release.

Organising your cables.

Friction lives in the approval that exists because of a policy from 2015, or the handoff that nobody owns. Often it's the lack of a defined decision-maker, or too many purposeless meetings (I bet you could name at least one you’re part of). 

This is where design principles earn their place. They force the friction question before the headcount question. Rather than landing on a number and hunting backward for an AI-shaped rationale to justify it, principles make you articulate, in advance, what you're actually optimising for and why.

Sometimes the honest answer is "we have too many people for this workload." Sometimes it's "we have exactly the right number of people doing the wrong things." Either way, the headcount number should fall out the other end of the design work, not get decided in a budget meeting and justified backward from there.

That means going into the room without the number already decided, which is a hard thing to do when everyone else in it is waiting for one.

But it's the difference between a cupboard full of cables you can never find, and one where you can put your hand on the right one when you actually need it.