Business Success: Luck, Not Merit.

Most people think startup failure is a problem of execution. Of course, that's part of it. However, more often than not it’s also a problem of information.

1. A new founder with scarce financial capacity need to find the right product-market fit in order to achieve compunding distribution.
2. However, to get the right product-market fit, they need to test their hypothesis against real users to learn what works and what doesn't.
3. To get real users, they need to get product-market fit or nobody will be interested to try anything.

It's a chicken and egg problem. The loop closes before the experiment can even start.

The founders who succeed are partly the ones who happened to know the answer (someone on the team or themselves had built something similar in another company), or they guessed, and guessed, and eventually the guess held up.

Those who have more money, either by coming from a wealthy family, by building it incrementally over decades or selling an asset like a house, have the ability to try more guesses. The more guesses they try increases their chances of winning. Those who are financially constrained have limited guesses and therefore less chances of winning.

Building a business is a game of chance. We overestimate merit and underestimate luck.

About 42% of failed startups in the CB Insights post-mortem analysis built something nobody wanted. The founders had no way to know what their customers would say without shipping. By the time they shipped, the money was running out.

The information that would have saved them already exists. It’s sitting inside the companies that already failed, and inside the ones that succeeded and decided not to publish. A failed venture is the only place the most valuable data lives: they spent the money, ran the experiment, and watched it fail.

The findings can be specific: pricing pages that destroyed conversion, onboarding flows that dropped half the users at step three, acquisition channels that looked great for a month and went flat, features people swore they wanted in interviews and ignored after signup. Those findings sit in someone’s head, or in a Slack archive that gets deleted six months after the company shuts down.

The knowledge that would help the next founder is the same knowledge the last founder paid to produce.

Universities used to be where this kind of knowledge accumulated. A researcher would publish, peers would replicate, the field would advance. The trouble is that universities now operate closer to being a businesses, far away from being recongised as a pillar of wisdom. They chase grants, file patents, and protect findings the same way a company would. Even setting that aside, a research team without a product on the market can’t generate the kind of data that matters for building one.

Customer behaviour shows up in production environments, not in clean experimental designs.

This is the biggest trap of the 21st century.

The institutions designed to share knowledge (universities) can’t produce the most useful kind. The institutions that produce the most useful knowledge (businesses) have no reason to share it.

Not everyone retains knowledge, of course. In 2008 Netflix had a three-day database outage that prevented DVDs from shipping to customers. They started moving to AWS soon after. The migration took years and they hit reliability problems other companies hadn’t faced at that scale. They built a tool that deliberately killed instances in production. The point was to find weak spots in advance of real failures. They called it Chaos Monkey. They published the concept on their blog in 2011. They open-sourced the code in 2014.

The practice spread and other companies adopted chaos engineering without reinventing it.

That work was operational. Netflix wasn’t selling chaos engineering, and no part of it was patentable in any useful way. Sharing it cost them nothing competitive and saved other companies years of incident-driven learning.

Open source is one version of this pattern: a company solves a problem internally, decides what they built isn’t what they sell, and releases it. The cost of producing the knowledge stays with the original company. The benefit spreads.

Open Source has solved knowledge sharing for technical domains

The problem is that open source is mostly about code. The hardest learnings in a startup are not technical. They are about the gap between what customers do and what they say, which acquisition channels look real and which collapse after week one, what kinds of pricing pages destroy conversion.

That data is what matters most. Almost none of it gets shared.

A small group of companies do publish operational findings. GitLab keeps its company handbook public.

Basecamp wrote down their product development process and called it Shape Up. Buffer published every salary in the company along with the formula used to calculate them, starting in December 2013.

None of those publications hurt the businesses that produced them. Several of them turned into recruiting advantages. More companies sharing operational findings would lower the cost of starting a business across the board. More businesses would survive, benefiting everyone. More tax revenue would follow. Fewer founders would burn two years rediscovering what three other founders had already worked out.

Productivity would skyrocket.

Productivity and rework in a society is proportional to the incentives organizations have to share their findings to the world.

There's a simple reason none of this is widespread at scale: there’s no incentive for it! Publishing operational findings has no obvious commercial return. Keeping everything as trade secrets, even when they aren’t really trade secrets, "feels" safer.

A government that wanted more successful companies in its borders would treat shared business findings the way it treats published research.

• Tax credits for publishing case studies.
• Grants for companies that release post-mortems of their failed product lines.
• A public registry of what worked and what didn’t, with anonymised data where required.

The same logic that funds basic science would apply. The social return is higher than the private return, so the public covers the gap.

None of this requires giving away the product. It requires giving away what the company learned on the way to building the product.

Startup success looks like merit from the outside, but in reality it's just a game of chance. To reduce the reliance on luck, businesses need to start publishing more of what made the skilled people skilled.

If you liked this, you might like readplace.com, built for exactly this kind of reading.

Thanks for reading. If you have some feedback, reach out to me on LinkedIn, Reddit or by replying to this post.