Here is a number almost nobody in my position will say out loud: over a three-year period, my marketing company went from 320 people to 38. We do over $10 million in revenue. Those two facts sit in the same sentence, and the distance between them is the entire story of what AI actually does to a business — not the conference-keynote version, the inside version.
I'm telling it because the industry is full of people theorizing about "AI transformation" who have never run one. This is what it looks like when it's your company, your people, and your name on the door. I started writing about this rebuild in Hands Up — chapter seven, "The Restructure" — this essay is the part that has happened since.
What actually happened
The lazy reading is "she cut costs." That is not what happened. What happened is that we stopped organizing the company around people doing tasks and rebuilt it around an AI infrastructure and intelligence layer — with people operating it.
Work that used to take a department — production, fulfillment, reporting, coordination, the endless glue work of a marketing agency — now runs through systems: data connected end to end, intelligence applied to every decision the operation makes, agents carrying the coordination load. The 38 people who run this company today are not doing the jobs the 320 did, smaller. They are doing a different job entirely: they build, steward, and improve the engine. It is the SME Inversion — expert to engineer to steward — running in production, at company scale.
The economics nobody wants to say plainly
This is the labor-hours-to-tokens shift as a lived case study, not a thought experiment. The output that once required 320 people's hours now ships from a fraction of the headcount, because the cost of a widening band of work moved from salaries to systems — from hours to tokens. Over $10 million in revenue with 38 people is a revenue-per-person ratio that the traditional agency model cannot approach, and that gap is the competitive story of the next decade in every service industry, not just mine.
When I debate agency owners who insist the old model is fine, this is the number I'm holding while they talk. You cannot argue with a cost structure. You can only adopt it or compete against it.
The human cost, honestly
Now the part that would be dishonest to skip. Going from 320 people to 38 means hundreds of people who worked here don't anymore. Some of that was the natural churn of three years. Some of it was the hardest category of decision an owner ever makes. Anyone who tells you an AI transformation is painless is selling something; anyone who pretends the pain is a reason not to transform is selling something more expensive.
What I owed people — what every owner owes their people in this transition — was the truth, early: the industry we knew was not coming back, the company would be rebuilt around the future instead of defended against it, and every person here would get the honest chance to evolve with it. Some did, and they became the 38 — the builders and stewards, the authors of the systems. Some built their next chapter elsewhere, carrying skills and clarity I hope served them. I wrote "Optimize for Both" because of exactly this tension: the technology story and the human story are the same story, and you do not get to choose only the comfortable half.
What I'd tell another owner standing where I stood
- This transition will be run — by you or on you. The market does not ask permission. The only choice is whether you lead the rebuild deliberately or absorb it late, at someone else's price.
- Start with the layer, not the tools. Subscriptions don't transform anything. Infrastructure, intelligence, agents — in that order, as one system.
- Tell your people the truth before the market does. The cruelest version of this transition is the one where leadership pretends, and people find out from the layoff instead of from the plan.
- The destination org chart is small teams on a big engine. Hire and promote for stewardship — the people who can encode what they know and run what they build.
I've spent twenty years being early on what's next — 2008 taught me what that costs and pays. This rebuild is the biggest bet of all of them, and it is why, when I stand on a stage and talk about AI in the trades, it isn't theory. The receipts run the company. If your audience needs to hear what this actually takes, that's the keynote.