.svg){kind=icon}
.webp)
On Monday I was talking with a group of founders, all running businesses of under 100 people: a consultancy, a software product and others.
We got onto the age-old question of when you add a management layer to scale through growth: a head office, people and culture, a CFO and the rest. One founder asked where his next hires should be. My answer was simply: whatever will unlock you.
The conventional wisdom is to stay small or have a plan to scale through those costs, but I'm not sure that's right anymore. If you ask an AI about this, it will give you the conventional wisdom. So this week I put the same two questions to ChatGPT and Gemini, three times each in fresh chats: one about a 16-person consultancy wanting to grow to 25, the other about where its next hires should be. Eleven of the twelve answers recommended a head-office role: an operations manager, people and culture, a fractional CFO. One mentioned AI, in a single clause. ChatGPT searched the web for four of its answers and came back with consulting benchmarks.
Then I asked each one whether AI changed any of its advice. All twelve said yes, and most cut or reshaped the hiring plan. When I added four words to the question, "given what's possible today", both apps put an AI role at the top of the list.
The answers are cheap and consistent now, but they're based on what's already happened. Ask about a business and you get advice about all the businesses that have existed up to this point, not how businesses should be built from here. Judgement and intuition are the scarce, valuable things, and they have to be developed.
Where the numbers come from
The textbook has numbers for when to add that layer, and they're shakier than they look. Larry Greiner's 1972 model of how organisations grow has no numbers on its size axis. The growth "barriers" at 8 to 12 and 40 to 70 people that turn up in scaling books come from multiplying an assumed span of seven to ten direct reports. Dunbar's 150 comes from comparing brain size and group size in primates, and a 2021 re-analysis concluded that specifying any one number is futile.
Venture capitalists, who see more growing companies than anyone, do no better. Their advice on the first HR hire ranges from the tenth employee to the 125th. Index Ventures, one of the few to cite data, recommends hiring a people leader "sooner than the data provided above indicates". Practitioners trace the stock answer of 50 to American employment laws that start at 50 staff; Australia's step up at 15. And founders hire later than they're told: on average at 40 to 50 people, according to an HR software firm's study of 973 UK startups.
The payroll records show something steadier. Studies of French and Portuguese firms find they add management layers one at a time, at thresholds of their own, usually after a run of growth. The force underneath is the one Edith Penrose described in 1959: a firm can only grow as fast as its managers can absorb. AI may change where the rungs sit. Nobody has measured that yet.
Whatever will unlock you
The companies building these models are hiring people anyway. This month City AM reported Anthropic offering £145,000 for an administrative business partner in London. This week Anthropic's jobs board listed 667 open roles, including 51 in finance and 16 in people, and OpenAI's listed 807, including 39 in finance and 19 in people. Both are building AI for their own people teams and still hiring people to run them.
If you're talking about a real person, it's judgement, not automation. It's where the judgement is needed, or where your judgement can be delegated to someone else's judgement. That person then has to consider how they exercise their own judgement, using all of the modern tools available to them. I think that's the duty incumbent on any leader now. ASIC has said much the same to directors: their duty of care and diligence extends to adopting and using AI.
The tools don't do the judging for you. When researchers gave small-business owners in Kenya an AI adviser, the weakest did worse and the strongest appear to have done better, mostly on similar advice.
In an age where information and capability are flattening, the focus has to be on what's happening now. You need people in your organisation who understand what's going on. Process is the part software encodes most cheaply, so optimise for intuition and curiosity: people who understand the forces that are constant and can reassess them for what needs to be navigated now.
(Declaring my interest: I co-founded New Dialogue, which sells software that lets Australian organisations use AI over their own information, under their own control.)
This month the Fair Work Ombudsman updated its guidance: AI answers to workplace questions can be out of date, mix up countries and time periods, and include changes that haven't started yet.
One more thing
In 2002 the Oakland A's won 103 games, the same as the New York Yankees, on less than a third of the payroll: US$39.7 million against US$125.9 million. They bought what the rest of baseball overlooked: players who got on base. Two economists, Jahn Karl Hakes and Raymond Sauer, later checked what the market paid for that skill. It had been underpaid since the 1980s. By 2004 the price had caught up.
"By the time people started reading Lewis' book," Sauer said, "the market had already adjusted."
Author: Matt Vitale
.avif)
