Due diligence ends on the day of signing. That's when the part begins where the investment thesis either turns into a result or stays on a pitch slide. The first 100 days in a portfolio company are the narrow window in which the operating partner still has the board's attention, a fresh mandate and permission to change things — and in which that mandate is easiest to burn. The most common way to burn it is to announce an "AI program."
In the note on AI due diligence we set out five questions worth working through before signing, and the register of red flags that stays in the fund's hands after the transaction. This post is about what the operating partner does with that register in the first 100 days — before anyone in the company hires a Head of AI.
Why an "AI program" is a mistake, not an ambition
A program sounds serious: a budget stream, an eighteen-month roadmap, workshops for departments, sometimes a new director. The trouble is that a program consumes exactly the resource that's scarcest in the first 100 days — the board's attention — and scatters it across a dozen initiatives, none of which delivers a result before the first review. A year later the fund has a presentation about "AI transformation" and not a single process it could show a buyer at exit as a real piece of value creation.
The alternative is less impressive and more effective: one pilot, wired into the value map, driven to a measurable result within the quarter. Not because one is enough for the whole hold. Because the first proven result buys the mandate for the second, while ten parallel initiatives with no result spend that mandate up front. It's the same conclusion Bain reached in its midyear report this summer: in a tougher market, the firms that win are the ones that steer scarce resources toward countable bets rather than a broad front.
From the red-flag register to a 100-day backlog
The due-diligence register isn't a document for the archive — it's a ready backlog, just written in the language of risk instead of the language of tasks. The first 100 days are about translating it into three items, in this order.
One process with an owner. Not "the sales department," but one process that has a person accountable for the outcome, not for the tool. If DD raised the flag "automation maintained by one person with no documentation," that's the first candidate: a process important enough to be worth taking ownership of, and narrow enough to close within the quarter.
One workflow with a before-and-after measurement. A pilot with no baseline is an anecdote, not a result. Before anything starts, you need a recorded baseline: how long it takes today, what it costs, who signs off on the outcome. Without it, at exit there's nothing to prove that AI changed anything, and the buyer will subtract it from the price as "unconfirmed."
Defusing vendor concentration. If the margin thesis rests on a single model vendor hard-wired in, the first 100 days are the moment to separate that out — before the product grows around one API. It isn't about rewriting everything, just a thin layer that lets you swap the vendor without touching the product. It's the same plan B we asked about in due diligence, moved to the delivery side.
Let us flag plainly that this is an operator's thesis, not a universal methodology: the order "process, measurement, vendor" comes from what, in our practice, most often blocks the second pilot — not from a ranking of importance detached from the specific company.
What not to do in the first 100 days
- Don't hire a Head of AI before there's anything to give them. A headcount before the first proven process is a cost still looking for a problem. The owner of one process is enough to start; the role appears when the backlog outgrows one person.
- Don't mistake a demo for production. A model that "answers from the knowledge base" in a meeting but has no owner, no data classification and no decision log isn't a pilot — it's a risk waiting for the first compliance review.
- Don't defer measurement to later. A baseline collected after the pilot has already started is contaminated. If it isn't there beforehand, it won't be there at all.
What this means for a fund, a large organization and a mid-sized company
Fund (PE). The first 100 days aren't the time for the company's "AI strategy," but for one move that at exit can be told to a buyer as repeatable. An operating partner who comes out of the 100 days with one proven process and an orderly backlog for the rest of the hold stands stronger than one with a presentation about transformation. That's exactly how we set up our work with funds: from the value map to the first pilot, not the other way around.
Large organization. If the company is part of a larger group, the same mechanism guards against drift between companies: one process with an owner and a measurement is a unit you can later copy into the next portfolio company, instead of inventing a transformation from scratch in each one.
Mid-sized company. Here the first 100 days are the most unforgiving, because there's no slack of people to fight fires. The good news is that a smaller company doesn't need a program — it needs one process that frees up the owner's or a key team's time. The rest can wait until that first one delivers a result.
Bring the value map
If you're closing a transaction right now, or you're early in the hold and hold a register of red flags from DD, bring it. We start from something concrete — we choose the first pilot together, from your value map, not from a list of fashionable use cases. Describe your case: mailto:[email protected]?subject=Rozmowa%20z%20Aurora%20AI.