[caption id="attachment_1983" align="alignleft" width="234" caption="Not Eric Ries."][/caption]

The most important thing I learned from reading The Lean Startup is the idea that “learning is the essential unit of progress for startups” (from the “Learn” chapter).

I think what I find fundamentally profound about this is the idea that there is a unit of progress for a startup. Ries defines a startup as “a human institution designed to create a new product or service under conditions of extreme uncertainty” (from the “Define” chapter). This makes sense as a definition, but if there’s extreme uncertainty about the universe, how can we know what the right direction is at any given moment, what direction we ourselves are progressing in, or even what constitutes “knowledge”?

Ries deals with this by applying the scientific method. The scientific method is also set up to deal with conditions of extreme uncertainty – i.e. the universe, which is very difficult (fundamentally impossible?) to observe and understand. Because confirmation is difficult (or impossible), the scientific method works by disproving hypotheses, rather than proving them.

In the context of a startup, disproving a hypothesis means learning, and it means validated learning, which is what Ries uses to gauge progress; failure is OK, as long as something has been definitely learned about the customer, or business model, or process technology, etc. Of course, there is in this context no “definitely”, but as humans we use “definitely” to mean “very, very probably.”

Disproving hypotheses seems like a difficult way to build a company – after all, there are an infinite number of possible hypotheses that should need to be disproved. But all this method does is take into account the reality of today’s production circumstances. Just as, in the universe, there is an infinite number of potential fact patterns, “in the modern economy, almost any product can be imagined and built” (“Lessons Beyond IMVU”). For a startup, the way to constrain the number of possible hypotheses to test is to stick to a particular vision of what the company needs to become, and then test hypotheses about ways to achieve that vision.

As a nice side effect, this method helps prevent a startup from focusing on good news that is satisfying, but actually irrelevant (or bad news that discouraging, but irrelevant). Take, for example, Ries’ discussion of metrics in the “Measure” chapter. He gives three guidelines for metrics that are worth measuring (and implicitly, metrics that are worth testing against): whether they are can be acted upon, can be understood, and can be trusted.

For example, a lot of metrics that are popular with startups, such as website visits, are actually not always valuable since they can’t be acted upon – an increase in hits on the website doesn’t by itself tell us about conversion rates or actual interest in using the product.

Without an “essential unit of progress” to track, the temptation is to use these metrics instead, since they help in raising money and provide a strong feeling of progress if nothing else is available. By providing the validated learning alternative it’s much easier to know what to measure, why, and how to act on it.