Everyone tracks the same startups. The ones with a funding round, a Crunchbase page, a Sifted headline. For a few weeks I've been tracking a different category, and I can't find it in any database: companies created from zero by large French groups.

Not acquisitions. Not internal brands. Not the twentieth innovation "initiative" with a slide deck and no P&L. Real companies — born in a corporate studio or a corporate programme, with their own legal entity, their own accounts, and enough size to matter.

They are invisible by construction. No funding announcement, because the group financed them. No M&A press release, because nobody bought them. No database entry, because databases are built on those two signals. And yet some of them weigh more than most of the seed-stage startups that get written about every week.

What I found so far

I posted the question on LinkedIn two days ago. Twenty-five comments later, the list is growing faster than I can verify it. Here is what I had going in — public information, or things I've had direct echoes of:

Two exits. Tens of millions in cumulated revenue. Hundreds of jobs. And no overview anywhere. Several of these came from readers in the last forty-eight hours; every one of them will be verified before publication.

The criteria are already doing their job. One reader pointed out that Free2move (Stellantis) grew by acquisition, not from zero — so it comes off the list. Astore, Accor's procurement platform, is the opposite case: 250 people, thirty offices, more than €3 billion of purchasing managed, third-party hotels as clients. If it is a separate company, it belongs here. If it is a division with a brand, it doesn't. That is the line, and it is worth being strict about it.

I have around twenty candidates now, across EDF, Crédit Agricole, AXA, Michelin, BNP Paribas, SNCF, Accor and others. I'm certain many are missing.

Make, invest, or partner

I wrote a few years ago that large groups have three ways to innovate, and that they rarely see them as one system.

In most groups these three levers sit in three different teams — business units, M&A, strategy, innovation, intrapreneurship — with nobody holding the whole picture. There is no control tower. Each team pulls its own lever for its own reasons, and the group ends up with a portfolio of minority stakes, a graveyard of POCs, and a lab that has never incorporated anything.

Corporate venture building is make, with entrepreneurs. It is the most demanding of the three levers, because it requires the group to do something it is structurally bad at: create an independent entity, hand it to people who are not employees, and let it compete. It is also the only lever that produces an asset the group did not have before.

Why nobody measures it

Here is what struck me while compiling the list. France measures corporate innovation quite well — as long as it is investment.

Orange Digital Ventures and Deloitte publish an annual barometer on corporate venture capital. France Invest has a white paper on corporate venture. Bpifrance tracks collaboration between large groups and startups. All useful. All about the invest and partner levers.

Nobody measures make. Nobody counts the companies that large groups actually create. The lever that demands the most, and the only one that builds something new, is the one with zero data.

Which means that when an innovation director walks into a board meeting and proposes to build a venture rather than buy a stake, she has no benchmark. No survival rate. No typical time to first revenue. No sense of what a good cap table looks like, or what it costs. Every conversation starts from zero. That is a large part of why so few of these companies get built.

The Observatoire

So we are going to count them.

Honey Lab is compiling the first map of corporate venture building in France. The criteria are deliberately strict:

Excluded: acquisitions, internal brands, and incubation programmes that never incorporated anything.

For each company we want the group, the year of creation, the ownership structure, and at least one hard number — revenue, headcount, or the exit. Where accounts are public, we will use them. Where they are not, we will say so.

An imperfect map beats no map. The corrections are where the learning is.

Version one will be published in the first quarter of 2027. It will be imperfect, and it will be published anyway. The Free2move correction taught me more about what counts than the six confirmed entries did.

If you know one

Founders of these companies, people running corporate studios, intrapreneurs who incorporated something: I want your venture on the list.

Send me the name, the group, the year, and one number. That's it. And if you run a group that has built one of these and never talked about it — you are exactly who this is for.