AI Doesn't Level the Field. It Sorts It.
By Elisabeth Virgilio, Founder, Maison Virgilio · November 2026
It is nine o'clock on a Monday morning at a mid-size beauty brand, and twelve people are sitting around a table with three dashboards on the screen, trying to answer a single question: why is the new serum underperforming in travel retail?
Sales has a number, marketing has a different one, and finance has a third, together with a spreadsheet to prove it. After forty minutes someone suggests a deep dive, everyone nods with some relief, and the decision quietly moves to next month's meeting, where it will meet the same three numbers again.
Meanwhile, two floors down, the company's first AI project has just gone live, and by all accounts it is a success: it processes supplier invoices faster and more accurately than ever before.
I have sat in some version of that Monday meeting for the better part of twenty years, sometimes as the person presenting the numbers and sometimes as the person asked to make sense of them, and I have come to believe that this room, rather than the back office, is where the AI race for mid-size companies will be won or lost.
The numbers finally caught up with the hunch
For a long time that belief was a hunch built on experience, but in September KfW Research gave it numbers, comparing roughly 6,000 German mid-size companies that use AI with otherwise similar firms that do not.¹
2024 was not a kind year for the Mittelstand: the typical company without AI saw its revenue shrink by 0.7%, while the typical AI user grew by 0.4%, earned roughly a quarter more margin, and generated around EUR 7,600 more revenue per employee.¹
Those averages are pleasant enough, but the more interesting story lies in who actually captured the gains. The productivity improvement was spread broadly across AI users, whereas the gains in growth and profitability were concentrated in companies that were already growing and already profitable, and among the fastest growers the gap to non-users widened to 7.6 percentage points.¹
KfW's own data don't prove this, but the study offers a reading very close to the one I would have given in that Monday meeting: growing companies tend to point AI at the top line, at marketing, pricing, customers and new markets, while struggling companies tend to point it at cost, which means the same technology, asked a different question, can produce a very different result.
Why I believe the moment is now
The first reason is that the gap compounds. AI is not lifting all boats equally; it is sorting them, and every quarter in which a competitor uses it to price more precisely, target more accurately and launch more intelligently widens the distance in a way that becomes harder to close the longer you wait.
The second reason is the J-curve. New technologies rarely pay off on the first day, because the upswing is preceded by months of unglamorous work in cleaning data, connecting systems and teaching people to trust a new way of deciding, as it was with the electric dynamo and later with the personal computer. The clock therefore starts when you start rather than when you feel ready, and the companies that began in 2024 are already climbing out of the dip.
The third reason is that difficult markets reward growth-oriented AI at least as much as good ones. In a tough 2024 the AI users still grew while their peers shrank, and when demand is tight, knowing which customer, which channel and which price point to fight for is not a luxury but the difference between holding your ground and slowly losing it.
The fourth reason is the one I find most encouraging. Cost-saving AI is easy to approve, which is precisely why almost everyone already has some, whereas revenue-driving AI requires a leader who is willing to change how commercial decisions are made, and that is why it is still rare. KfW makes the point explicitly when it concludes that management's direction and follow-through decide whether AI succeeds, and for mid-size brands between Paris and Geneva, whose data is typically scattered across distributors, e-commerce and travel retail, I see that less as a threat than as a genuine opening.
Start with one decision, not one more dashboard
So let us go back to that Monday morning and imagine the same meeting six months from now. The serum question comes up again, but this time there is one shared view of sell-out, pricing and traveller profile by airport instead of three competing numbers, and the team spends its energy arguing about what to do rather than about whose spreadsheet is right, so that the decision is made before the coffee has gone cold.
That, to me, is what a revenue-driving AI project looks like. It is not a moonshot but a single stalled commercial decision, named out loud, with the data and the model built around it and a leader who takes ownership of what happens next.
If you lead a mid-size brand, my suggestion is therefore a simple one: before you approve the next efficiency pilot, find the decision your Monday meeting keeps postponing and start there. The invoices can wait, but growth, as the numbers now show quite clearly, will not.
Elisabeth
The beauty of Data. Designed for Growth.
References
1. Volker Zimmermann, “Mittelständische KI-Nutzer sind erfolgreicher,” KfW Research, Fokus Volkswirtschaft No. 558, 15 September 2026. Based on wave 23 of the KfW-Mittelstandspanel (survey spring 2025, AI use 2022–2024, performance 2024). Associations, not proven causal effects. https://www.kfw.de/PDF/Download-Center/Konzernthemen/Research/PDF-Dokumente-Fokus-Volkswirtschaft/Fokus-2026/Fokus-Nr.-558-September-2026-KI-Wirkung.pdf
