Failure is a Signal

Definitions

Let’s start by breaking the question down to its essentials.

What is better: a first time entrepreneur with no previous startup employment or a second time entrepreneur from a failed startup?

For any discussion to make sense, we must agree on the following:

1. Better, in what sense?
2. What exactly are we comparing?
3. What defines a second timer?
4. What constitutes success or failure in a startup?

Here’s my interpretation:

An attempt is a non-negligible effort to accomplish something. For startup founders, that could be working on something full-time for one year.

Success is defined as generating a good ROI, and failure is defined as generating a poor ROI (taking into account all stakeholders and their alternatives). Both are subjective, and can only be evaluated in retrospect. Attempts can also end in limbo, meaning neither succeed nor fail.

With that in mind, we can examine the original question from a statistical modeling perspective. Each of us has an implicit model for predicting the likelihood of success for a startup, in which the team often plays a major role. For example, one may believe that, with all else equal, founder experience increases the probability of success.

The question can thus be rephrased as follows: based on your model, what is the marginal effect of knowing a founder had a previous failed attempt? Put differently, having just learned that a first-time founder actually failed a previous attempt, would you adjust the likelihood of their success now to be higher or lower?

A first-order model of startup success

The three core elements of every startup are team, product, and market. Our main focus here is on the team.

Marc Andreessen writes:

The caliber of a startup team can be defined as the suitability of the CEO, senior staff, engineers, and other key staff relative to the opportunity in front of them.

You look at a startup and ask, will this team be able to optimally execute against their opportunity? I focus on effectiveness as opposed to experience, since the history of the tech industry is full of highly successful startups that were staffed primarily by people who had never “done it before”.

The ability of a founder to execute is largely determined by two factors: their talent and their experience.

Bill Aulet writes:

While I find many elements of entrepreneurship that draw from empirical processes, I also find many others that require creativity. It is neither science nor art. Instead, it’s a craft — a process that draws from both.

Creativity, art… talent. Empirical processes, science… experience. Let’s examine how previous failures relate to each.

Previous failures and talent

Talent is a strong causal factor for a founder’s ability to execute, but it cannot be measured directly. In other words, talent is a latent variable in our model.

The best we can do is find proxies for talent: evaluate how smart a founder seems to be, look for a history of excelling in other activities, perform reference checks, etc. Note that these do not cause success by themselves — in information theory terms, they signal talent.

A previous failure is one such signal. Indeed, if talent causes success, a previous failure implies less talent by Bayes’ theorem.

Asking how likely a founder is to succeed if their last attempt failed is similar to asking how likely a possibly biased coin is to land on heads if its last toss landed on tails: given the founder’s true talent, or the coin’s theoretical probability of landing on heads, we’re dealing with independent events and history carries no additional information; but lacking that knowledge, Bayesian inference tells us that the coin is now less likely to land on heads, and that the founder is now less likely to succeed — compared to our prior estimate.

Taken to the extreme, consider a coin that landed on tails 100 times straight, or a founder whose last 5 attempts failed. It doesn’t take a probabilist to realize there’s an underlying factor at play here, other than pure chance.

Of course, while this is true from a probabilistic perspective, each case should be judged on its own merits. For example, a prior failure that is fully explained by external factors can be safely ignored, as it carries no information.

Previous failures and experience

Experience is also a strong causal factor for a founder’s ability to execute, because experienced founders are likely to execute faster and make less mistakes. As Rob Fitzpatrick writes:

Startups are more craft than science. It’s something you learn by doing.

However, not all startup experience is equally valuable: while failure is a good teacher, success is a much better one. Paraphrasing Tolstoy, this is because successful startups are all alike, but every failed startup is failing in its own way. At best, you learn one way to fail out of a thousand (Quora has more on that subject).

Bottom line, experience is generally a strong positive signal, but founders of failed startups did not necessarily gain much of it. Obviously, as always, every case must be judged on its own merits as the specific details make a huge difference.

Tying it all together

So back to the original question — how would I update my prediction if I suddenly discovered that a first-time entrepreneur had previously failed?

In a vacuum, I’d take this as-

• a signal that the founder is less talented than I originally thought, and therefore less likely to succeed.
• a weak signal that the founder is more experienced than I originally thought, and therefore more likely to succeed.

Combining the two, I’d adjust my prediction of the startup’s success probability to be slightly lower than before.

Empirical data

Studies of this topic should be taken with a grain of salt: they are all correlational and vary widely in definitions of success and failure as well as population.

Still, two studies on startups deserve mention:

• Published in Journal of Financial Economics, “Performance Persistence in Entrepreneurship” found that first-time entrepreneurs and those who have previously failed are just as likely to succeed (both were much less likely to succeed than entrepreneurs with a track record of success).
• A discussion paper by Centre for European Economic Research, “If You Don’t Succeed, Should You Try Again?” looked at nearly 8,400 German startups and found that failed entrepreneurs did not do any better than first-timers, and in fact had poorer outcomes the second time around.

Prof. Francis Greene, who performed the latter, concludes:

Part of the folklore about successful entrepreneurs is that they succeeded because they first failed. This is a myth.

While second-chance stories are comforting, my research shows that entrepreneurs don’t learn from their mistakes. In fact, it’s the opposite: fail once and you’re most likely to fail again. Believing in the myth only sets entrepreneurs up for more failure — and leads to disappointment and frustration.

To be sure, there are some quantitative studies that show serial entrepreneurs have greater success than first-timers. But the overwhelming majority of the studies don’t show that failed entrepreneurs specifically are more successful. In my opinion, the studies that do touch on failure usually come with limitations or caveats.