A brand I know turns over €30 million a year. This year, they want €100 million. Same team, same headcount, not one new hire on the plan. That’s 233% growth in twelve months, and they intend to get there on efficiency and technology alone.
I didn’t laugh. I did the maths.
If you want to grow revenue without hiring, 233% in a year isn’t a simple stretch target. It’s a different category of ambition, and it’s worth working out exactly what has to be true for it to happen before you commit a year of your life to it.
So I went looking for every serious piece of research I could find on companies that grew revenue fast through AI and automation, without adding people, and I compared what I found against what this brand is attempting. Some of it is encouraging. Most of it is a warning.
How to Grow Recenue Without Grow Headcount
The number that made me check my working
Growing from €30 million to €100 million with a flat headcount means revenue per employee has to roughly triple in a single year. There’s no way around that arithmetic. If you have 100 people generating €300,000 each today, you need those same 100 people generating just under a million each by December.
The best documented example of a company pulling off something like this is Klarna, and even Klarna needed three years, not one. Klarna’s own Q4 2025 investor release states that since Q4 2022, revenue grew 104% while operating expenses fell 8%, and revenue per employee reached $1.24 million, a 3.6 times increase since 2022. Headcount fell by 49% over that same period. So Klarna’s efficiency miracle involved a 3.6x jump in revenue per head, spread across three years, achieved partly by cutting the workforce in half.
This brand wants a 3.3x jump in one year, with the headcount untouched in either direction.
Here’s how the two compare side by side.
| Metric | Klarna, 2022 to 2025 | This client’s 2026 target |
|---|---|---|
| Revenue per employee, multiple | 3.6x | 3.3x |
| Timeframe | 3 years | 1 year |
| Headcount change | Down 49% | Flat |
| Approximate annual pace | About 53% a year, compounded | 233% in one go |
That table is why I didn’t dismiss the target, and also why I didn’t cheer it on unquestioned. It’s not impossible. It’s rare enough that when you go looking for real, named, audited examples of it, you mostly find caveats.
Why AI on top of a shaky process just makes the mess faster
Here’s the uncomfortable finding sitting underneath almost every serious study on this topic published in the last two years.
AI does not fix a process. It amplifies whatever is already there. Feed it chaos, get faster chaos. Feed it a system that already works, get that system running at a scale no team of humans could manage on their own.
The Boston Consulting Group’s 2024 study of 1,000 executives across 59 countries found that 74% of companies have yet to show any tangible value from AI. The 4% getting the most out of it, which BCG calls the leaders, follow what it terms a 10-20-70 rule: 10% of the effort on algorithms, 20% on technology and data, and 70% on people and process. Read that split again. The tool is the smallest line item. The process work is everything.
McKinsey’s 2025 global AI survey found a similar pattern from a different angle. Of the nearly 2,000 organisations surveyed, 88% now use AI regularly somewhere in the business, but only 39% can point to any measurable impact on profit, and most of those put the number below 5%. The small group McKinsey labels high performers, roughly 6% of respondents, are described as nearly three times as likely as everyone else to have fundamentally redesigned their workflows before layering AI on top.
Then there’s the study that made headlines for the wrong reason. MIT’s NANDA initiative published a report in 2025 titled The GenAI Divide, and the figure everyone quoted was that 95% of generative AI pilots show no measurable financial return. Worth knowing before you repeat that stat at a dinner party: it’s preliminary, based on interviews and surveys rather than audited financials, and it isn’t peer reviewed.
It’s still useful, because the report’s stated reason for failure lines up with everything else here. The tools couldn’t retain feedback, adapt to context, or fit into how the business actually worked day to day. That’s not an AI quality problem. That’s a process problem wearing an AI costume.
One more piece worth having in your back pocket, because it will come up whenever you’re trying to grow revenue without hiring and someone tells you it usually fails. Researchers from the University of Toronto, using US Census Bureau data on tens of thousands of manufacturers, found what they call a productivity J-curve. Firms typically get worse before they get better after adopting industrial AI, and the losses concentrated among older firms, with roughly a third of that dip explained by companies abandoning the structured management practices they already had, in favour of the shiny new tool. Firms that were already operationally mature before adopting AI captured the bulk of the eventual upside.
The 70% statistic you’ve probably heard, and shouldn’t repeat
While I was digging through this, I kept bumping into the claim that 70% of digital transformations fail. It’s everywhere. Consultancy decks, LinkedIn posts, probably a slide in one of my own presentations from a few years back. So I traced it.
It doesn’t hold up. McKinsey has used the figure in change-management writing since at least 2015, without citing a source for it. Harvard Business Review repeated it in 2019, tracing back not to a study but to a 2018 opinion column by a software company executive. Even the 1993 origin, from Michael Hammer and James Champy’s reengineering work, was described by the authors themselves as an unscientific estimate. A peer-reviewed 2011 paper in the Journal of Change Management went looking for empirical evidence behind the number and found none.
BCG’s actual research on the topic, a 2020 study of 825 executives, found that 70% of digital transformations fell short of their stated objectives, but 44% of those still created some value along the way. That’s a more honest number, and a more useful one. Falling short of an ambitious target and achieving nothing are not the same thing, and the industry has spent a decade blurring the two.
What really works: the study I’d put my own money on
If I had to point to one piece of research that best explains how you grow revenue without hiring, it’s the one on customer support agents, not because it’s the flashiest but because it’s the most rigorously done.
Economists Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied the staggered rollout of a generative AI assistant to 5,179 customer support agents at a Fortune 500 company, published in the Quarterly Journal of Economics. Average productivity, measured in issues resolved per hour, rose 14%. For the newest and least experienced agents, it rose 34%. For the most experienced agents, barely at all.
Here’s the part that matters for anyone chasing outsized growth. The researchers found that the AI had effectively learned the habits of the company’s best performing agents and was spreading those habits to everyone else. The tool didn’t invent a new way of working. It took a system that already existed, quietly, in the heads and habits of the top 10% of the team, and rolled it out to the other 90%.
That’s the entire thesis in one paragraph.
AI is a phenomenal distribution mechanism for a system that works. It has no system to distribute when none exists.
Klarna, the cautionary half of its own success story
Klarna’s efficiency numbers are real and well documented, but the company’s own experience also shows exactly where the approach breaks. In May 2025, CEO Sebastian Siemiatkowski told Bloomberg that the company’s AI-first customer service push had produced lower quality service, and Klarna began rehiring human agents. The company later said an overemphasis on cost, rather than the technology itself, was to blame.
Put the two halves of Klarna’s story together and you get a really useful lesson. The AI performed brilliantly on high volume, standardised queries, the part of the process that was already codified and consistent. It struggled on the complex, judgement-heavy cases, the part of the process that had never been fully systemised in the first place.
Same company, same technology, wildly different results, depending entirely on how mature the underlying process was in each area.
Systemise first, then automate: the version of this thesis I’d defend
There’s a real counter-argument worth taking seriously here, and I’d rather deal with it properly than pretend it doesn’t exist.
Wharton’s Ethan Mollick argues that waiting for a fully documented playbook before you start using AI means you’ll be waiting long after everyone else has already figured it out through experimentation. His point, in short, is that employees using AI daily often discover the better workflow themselves, in the doing, rather than needing it handed down from above.
He’s not wrong, and the Brynjolfsson study actually supports both sides of this at once. The playbook that AI distributed to Klarna’s junior agents was never written down anywhere. It existed only in the behaviour of the best performers. In a sense, the AI did the codifying. The system existed before the automation, just not on paper.
Process mining tools, the kind that reconstruct how work actually happens from the data trail a business leaves behind rather than from what’s written in a manual nobody reads, are built on the same insight. You can discover your system with AI’s help, rather than only building one by hand and then bolting AI on afterwards. Fine. I’ll take that. But notice what both routes have in common.
Somewhere before you scale, a repeatable, measurable way of working has to exist and has to be surfaced, whether it’s mined from data, distilled from your best performers, or written down the old-fashioned way. The order can vary. The requirement doesn’t.
Codify first, or codify at the same time you deploy, but codify. That’s the thesis I’ll actually stand behind, and it’s a more defensible one than the version that says never touch AI until you have a binder full of documented processes.
So what does this brand need to hit 233%?
Bringing this back to the business at the centre of this piece. To go from €30 million to €100 million in a single year with the same headcount, three things have to be true, and I’ll say them plainly rather than dressing them up as a checklist.
First, whatever is already working needs to be identifiable and repeatable, not just present in the heads of a few strong performers. If the best salesperson, the best account manager, or the best delivery lead is quietly carrying results that nobody else can replicate, that’s the exact pattern AI is good at spreading, provided someone notices it exists and points the tool at it.
Second, the growth needs to come disproportionately from the parts of the business that are already standardised and measurable, the Klarna lesson in reverse. Trying to triple revenue per head in the messiest, most judgement-heavy corner of the operation is where quality collapses first. The standardised corners are where efficiency gains actually stick.
Third, and this is the one people skip, someone needs to measure the baseline honestly before switching anything on. Part of why MIT’s researchers found so many pilots showing no measurable financial return is that plenty of those businesses never had a clean before-and-after comparison to check against in the first place. You can’t prove 233% growth came from the system if you don’t know what the system was producing on day one.
None of that guarantees the target lands. It’s a really aggressive number, and I’ll be honest with the brand about that, the same way I’ve been honest with you about it here. But it also isn’t fantasy, and the businesses in the research that came closest to something like it, the small operators growing 200% to 300% in a year, all had one thing in common.
Every single one of them could point to a specific, repeatable way of working that existed before the tools arrived, even if it had never been written down.
Leave a Reply