Why 80% of our ideas die before a line of code

A venture studio is not an idea factory. It is a filter. In any given year we will write up somewhere between forty and sixty theses, and we will incorporate three or four of them. The other fifty-odd do not fail — they get killed, deliberately, by a process we designed to be uncomfortable.

That ratio is the point. Every idea we do not build is capital and bench time redirected to one that survived scrutiny. The hard part is not generating ideas; it is being honest about them fast enough that the honesty is still cheap.

Kill criteria are written before the work starts

When a thesis enters the pipeline, the first artefact is not a deck. It is a one-page document naming the three things that would have to be true for the venture to work, and the specific evidence that would prove each one false. We sign that page before anyone talks to a customer.

Writing the criteria up front removes the most expensive failure mode in venture building: moving the goalposts. Once you have spent six weeks on a thesis, the temptation to reinterpret a weak signal as a promising one is overwhelming. A page you signed in week zero is much harder to argue with than a memory of what you meant.

"The cost of a wrong no is one idea. The cost of a wrong yes is a year, a team, and a fund's worth of credibility."

What actually kills a thesis

Across the last three cohorts, the reasons cluster more tightly than we expected. Four patterns account for most of the deaths:

• The problem is real but infrequent — buyers agree it hurts, then do nothing for eleven months of the year.
• The wedge has no owner — no single budget holder can say yes without three other departments in the room.
• Distribution is the whole business, and we have no unfair way in.
• The market is fine, but the version we can build is the fourth-best product in it.

Notice what is missing: "the technology did not work." In applied AI and fintech alike, technical risk is rarely what kills a studio venture. Demand risk and distribution risk are.

Making no cheap

The habit only compounds if killing an idea costs a team nothing socially. We review theses as a group, the person who authored it presents the case against it, and the memo that closes a thesis goes into the same folder as the ones that got built. Nobody has ever been penalised here for a well-argued no.

The result is that our bench treats validation as a search for the fastest disproof rather than a hunt for encouragement. That is the whole trick, and it is why 80% of what we write up never reaches a repository.