What Is Lean Startup?

By his own account, Ries first used the term in a blog post on 8 September 2008. He gave it two senses there: low burn, and an application of Lean Thinking. He named three drivers: free and open-source platforms, agile development and rapid iteration with customers through the Customer Development process. His book "The Lean Startup" came out in September 2011.

The word "lean" is older. John Krafcik, of MIT's motor vehicle programme, used it in autumn 1988 in "Triumph of the Lean Production System" for plants with a "lean" production policy. Mark Graban credits him with the phrase and points out that the practices had already developed at Toyota.

On his website, Ries sums up the method in five principles:

• Entrepreneurs are everywhere: you don't have to work in a garage.
• Entrepreneurship is management: a startup is an institution and needs a kind of management of its own.
• Validated learning: a startup exists to learn how to build a sustainable business, and experiments can verify that learning.
• Build, measure, learn: turn ideas into products, measure how customers respond and decide whether to pivot or persevere.
• Innovation accounting: measure progress, set milestones and decide what comes first.

Steve Blank tells it differently in Harvard Business Review (May 2013). He defines a startup as a temporary organisation searching for a repeatable and scalable business model, and gives three principles: summarise your hypotheses in a business model canvas, go out and ask customers and partners, and develop in an agile way.

The misreading is to take the method for a way of spending less, or for another name for an MVP. An MVP is the tool for one test, and the method decides what gets tested and what follows from the result. Spending little is in the name and in none of the five principles. Ries didn't promise more either: in 2008 he left open whether such companies would bring investors higher returns.

First, write down your assumptions. Each block of a business model canvas is one, such as who buys and how much they pay. Word them so that they could turn out to be false.

Second, pick the riskiest: the one that sinks the business if it fails, even when everything else goes well.

Third, design the smallest test that gives you data. That could be an MVP, a page with a price and a buy button, or a service you deliver by hand at first. If you already have enough traffic, an A/B test will do.

Fourth, write down the metric and the threshold before the test runs. Without this, any result ends up looking good.

Fifth, launch, measure with real customers and talk to them. Blank calls this getting out of the building.

Sixth, decide and record it: persevere, pivot to another hypothesis or drop the idea. Then start again with the next assumption.

An invented example, with every figure invented. A company wants to deliver lunch to offices, and its riskiest assumption is that offices of 20 to 50 people will pay €8 per meal in return for ordering the day before. Instead of building a kitchen and an app, it offers the service to 10 offices with a form and a contracted restaurant. Before starting, it writes down the threshold: 3 of the 10 offices order two weeks running. Only 1 does, and two say they would order for company meetings. The team pivots to those meetings and tests again.

The method tries to make being wrong cheaper: finding out early that an assumption is false costs less than finding out with a finished product. There are trials that say something about this, within limits:

• In a randomised trial published in Management Science in 2020, Camuffo, Cordova, Gambardella and Spina followed 116 Italian startups for about a year. All received 10 training sessions, and some were also taught to form and test hypotheses the way scientists do. That group performed better, pivoted more often and was no more likely to drop out at that early stage.
• A replication published in Strategic Management Journal in 2024, with 759 firms in four trials, found that more ideas get dropped. The effect on pivots isn't linear: treated firms made a few, as opposed to none or repeated changes.
• Leatherbee and Katila (2020) studied 152 teams from the US National Science Foundation's I-Corps programme. The parts of the method linked up as expected, and team members with an MBA resisted using it, although they saw its value afterwards.

Two cautions. The first two studies test a scientific approach to decisions, similar to Ries's, and leave his book aside. And only the abstracts were read here, without the full papers. None of those abstracts gives a figure you could promise a founder, and Leatherbee and Katila warn that the method had been presented as a solution for everyone with little rigorous testing behind it.

It also has a blind spot. Gans, Scott and Stern wrote in Harvard Business Review (2018) that entrepreneurs often run with the first plausible strategy they identify. Testing an idea well doesn't tell you whether a better one existed, so compare several before you start, for example with a go-to-market plan and a TAM SAM SOM estimate. Design thinking asks how someone uses a solution, and Lean Startup asks whether the business holds up.

• Write each assumption as a sentence that could turn out to be false.
• Start with the assumption that would sink the business if it failed.
• Set the metric and the threshold before you see any data.
• Measure what customers do, such as paying or coming back, as well as what they say.
• Talk to customers so that you understand the numbers.
• Record each decision and what you learned in each cycle.
• Accept dropping an idea as a valid result.

• Calling a half-finished product an MVP with no question behind it.
• Moving the threshold after seeing the result.
• Measuring visits or sign-ups when the assumption is about paying.
• Pivoting after every test without leaving time to measure.
• Testing minor details before the main assumption.
• Treating the method as a guarantee of success.
• Showing real customers a product that could cause harm before meeting the rules of that sector.