Lean Startup by Eric Ries

"You too can achieve fame and fortune with a great product and hard work."

This is the grand myth that the media loves to proclaim. However the stories we hear are a result of both selection bias (major success stories are newsworthy, the countless failed attempts are not) and after-the-fact rationalization (data is rarely available to determine true cause-and-effect).

So how can startup success move from being an art to being an engineering process? Lean Startup begins by recognizing that entrepreneurs are people who create new products and services under extreme uncertainty. The problem for a startup is figuring out the right thing to build and how to build a sustainable business around it.

Part I - Vision

Since, by definition, a startup doesn't know whether its product idea will be sustainable in the marketplace, the key insight is to rapidly expose the core assumptions about the product and then systematically test them. Each tested assumption is something new that is learned. If the assumption is validated, then the startup can persevere with the product direction; if it's not, then the product direction may need to change (called a pivot). Each stage of learning leads to new hypotheses to be tested and the opportunity to learn more about the product.

This process is a feedback loop called Build-Measure-Learn (see the diagram above). To be most effective, the feedback loop should be as short as possible. For example, an assumption might be that the world needs a new social media app. One way to test this is to spend $10 million of venture capital and two years building the app. Unfortunately, if the assumption is mistaken, then a lot of time and capital has been wasted. However, if a simple prototype were developed, or a video produced, or an Adwords test done, then it might be possible to learn of this mistake much sooner. This simplest version of the product is called a Minimum Viable Product (MVP), a minimal product that allows you to validate an assumption about the product. If the product is not meeting customer needs, then it may be possible to modify it in some way or target it at a different audience or decide that the whole idea needs to be scrapped. Each of these ideas are new hypotheses that can be tested.

The purpose of the feedback loop is to learn the crucial things you need to know without wasting resources. This does not occur by simply asking customers (they may be mistaken) or viewing sales metrics (which hide the specific causes of what is or is not working). Validated learning only occurs by testing a single assumption and carefully measuring the outcome.

Part II - Steer

Ries describes a number of MVP types. A concierge MVP is a product aimed at one or a few customers to see if they will buy it. A Wizard-of-Oz MVP is a prototype where functionality is manually provided. An MVP allows a startup to establish real data as a baseline. Then the startup engine can be tuned from the baseline to the ideal, persevering or pivoting as necessary.

Instead of relying on general metrics such as sales data or hits (termed vanity metrics), focus on metrics identifying cohorts (groups of users with common characteristics, such as first-time users, paid users or corporate users) that use the product in different ways. Metric reports should be actionable (demonstrating clear cause and effect), accessible (as simple as possible, understandable and concrete) and auditable (testable by hand).

A sign that a startup needs to pivot is when product experiments become ineffective or development seems unproductive. Note that any new design should always be tested against the old design to ensure that it is an improvement. Types of pivots include zoom-in (one feature becomes the whole product), zoom-out (the product becomes a single feature), customer segment (different customers need the product), customer need (customers have more important problems to solve), platform (app to platform or vice-versa), business architecture (B2B or consumer, complex or volume), value capture (type of revenue model), engine of growth (viral, sticky or paid growth), channel (internet, traditional retail) and technology pivots.

Part III - Accelerate

The purpose of the feedback loop is to determine which activities create value and which are a form of waste. Value is not found in the creation of stuff, but rather in the validated learning about how to build a sustainable business.

Ries recommends testing in small batches, for example, rapid releases to a small amount of users. Startups don't starve for ideas, they drown (most ideas have marginal benefit, hence the need to focus on validated learning).

Growth models include the sticky engine of growth (don't focus on the number of customers - a vanity metric, instead measure customer acquisition and churn), viral engine of growth (customers cause growth as a side-effect of use, for example, Facebook and Tupperware) and the paid engine of growth (revenue per customer exceeds cost of acquiring the customer).

To accelerate, Lean Startups need a process that provides a natural feedback loop. When a problem occurs, use the method of Five Whys (a method similar to a child repeatedly asking "Why?") to identify the root cause. Then make incremental investments and evolve the startup’s processes gradually. For example, if a problem ultimately occurred because an employee wasn't trained correctly, then the best solution might be a one hour training session rather than developing an eight-week training course. Also, look to improve processes rather than blaming personnel. For example, if a build was broken, is it possible to make the build process more robust?

The culture of a lean startup is one of data-driven decision making that focuses on early customer involvement and rapid iteration.