What I learned shutting down a company
We shut Foundry down in the autumn of 2021, fourteen months after starting. We returned roughly 60% of the seed round. I have written and deleted this post about six times since then; enough distance has passed that I think I can be honest about it.
What went wrong
We built for the demo, not the Tuesday. Foundry gave data teams a shared workspace, and it demoed beautifully. People said "this is great" in every single call. What nobody ever said was "when can I have it." We mistook enthusiasm for demand, and those are not the same signal — enthusiasm is about the idea, demand is about the pain.
I confused activity with progress. We shipped constantly. Weekly releases, visible changelog, real velocity. It felt like the opposite of failing. In retrospect the shipping was a way to avoid the harder work of finding out whether anyone needed the thing.
We never found a user who would be upset if we disappeared. This is the whole test. We had users. We had people who logged in weekly. We had zero people whose week would have been worse without us — and I knew that, in some pre-verbal way, for about six months before I said it out loud.
What we got right
Two things, both of which I would do again.
- We decided fast once we decided. From "this is not working" to wound down was seven weeks. I have watched friends spend two years on that transition. The money is not the expensive part; the time is.
- We told our investors the truth early. The conversation I dreaded most turned out to be the easiest. They had seen it before. One of them has since introduced me to two jobs.
The thing nobody tells you
The failure was not the hard part. The hard part was the six months before the failure, when I could see the shape of it and kept looking for a reading of the data that let me not see it. That was the genuinely miserable stretch. Once we admitted it, the actual shutdown was mostly logistics.
If you are in that stretch right now, the question that finally cut through it for me was not "can this work?" — you can always construct a story where it works. It was:
If this were someone else's company, and they showed me exactly this data, what would I tell them to do?
I knew the answer immediately. It took me another two months to take my own advice, which is roughly the human discount rate on unwelcome truths.
I do not regret it. I learned more in fourteen months than in the three years before, and nearly everything I now believe about building products came from getting it wrong at my own expense. But I would not romanticize it either. It was a bad year, and the lesson would have been cheaper if I had talked to more people before writing code.