The Open Curriculum

The Firm · Level 1

The Lean Startup

Eric Ries · 2011

Thesis

Startups fail mainly by building products nobody wants, and they can avoid it by treating every plan as a set of untested hypotheses: building the smallest product that can test them, measuring real customer behavior and deciding quickly whether to change course, so that progress is measured in validated learning rather than in features shipped.

The main ideas

Validated learning

Eric Ries defines a startup as "a human institution designed to create a new product or service under conditions of extreme uncertainty." The definition covers a team inside a large corporation as much as a company in a garage. What distinguishes it is uncertainty: neither the customer nor the product is known in advance.

Ries developed his method as co-founder and chief technology officer of IMVU, a company founded in 2004 that sold three-dimensional avatars for instant messaging. He combined Steve Blank's method of customer development with the principles of lean manufacturing from the Toyota Production System.

His own founding mistake is the book's starting case. IMVU spent months building its avatars as an add-on to existing instant-messaging networks, on the theory that users would bring their friends. Nobody downloaded it. When the team finally talked to customers, they learned that people did not want to risk looking foolish in front of their friends with an unfamiliar product, but were happy to use a standalone service to meet strangers. Most of the integration work had been wasted.

The lesson Ries draws is that the measure of a startup's progress is not output but validated learning: knowledge about customers demonstrated by real behavior, not opinion.

Implication: judge the progress of any new venture by what it has proved about customers, not by how much it has built.

The minimum viable product

Every plan, Ries argues, rests on leap-of-faith assumptions, of which two matter most. The value hypothesis asks whether the product delivers value to customers once they use it. The growth hypothesis asks how new customers will discover it.

The tool for testing them is the minimum viable product, which Ries defines as the version of a new product that allows a team to collect the maximum amount of validated learning about customers with the least effort. It is often far less than a working product. Dropbox tested demand with a short demonstration video aimed at an online technology community; the waiting list for its beta jumped from 5,000 to 75,000 people overnight. Nick Swinmurn tested whether people would buy shoes online, before founding Zappos, by photographing shoes in local stores, posting the pictures and buying each pair at full price when an order arrived.

Other forms include the concierge minimum viable product, in which founders deliver the service by hand to a first customer, as Food on the Table did with its meal-planning service, and the Wizard of Oz test, in which people secretly perform what customers believe is automated, as the question-answering service Aardvark did.

Implication: before building, identify the assumption most likely to sink the plan, and find the cheapest way to test it with real customers.

Build-Measure-Learn and innovation accounting

Ries describes a startup's activity as a feedback loop: build a product, measure how customers respond, learn whether to change course. The aim is to minimize the total time through the loop.

Measuring requires what he calls innovation accounting, built on three learning milestones: establish a baseline with a minimum viable product, tune the product toward the ideal, then decide whether to pivot or persevere.

The obstacle is vanity metrics, such as total sign-ups or cumulative revenue, which rise over time almost regardless of what a company does and so make every team look successful. Ries recommends actionable metrics, which show cause and effect, and particularly cohort analysis, which follows groups of customers who joined at the same time, so that a team can see whether each new group behaves better than the last. Good reports, he argues, are actionable, accessible and auditable.

His example is Grockit, an online test-preparation company that had been adding features while its gross numbers rose. When it switched to cohort analysis and split tests, comparing versions of the product on separate groups of users, it found that many features made no difference to customer behavior.

Implication: measure the behavior of successive cohorts of customers, not totals, and test each change against a control group.

Pivot or persevere

Ries defines a pivot as "a structured course correction designed to test a new fundamental hypothesis about the product, strategy, and engine of growth." A pivot keeps what the company has learned and changes one element of its strategy, rather than starting over.

He lists ten kinds: zoom-in, where a single feature becomes the whole product; zoom-out, the reverse; customer segment; customer need; platform; business architecture; value capture; engine of growth; channel; and technology.

His case is Votizen, which began as a social network for verified voters. Its first minimum viable product cost about $1,200 and took three months. When the numbers stayed flat, its founder, David Binetti, pivoted several times, eventually turning it into a platform for citizens to lobby their representatives.

The mechanism is that innovation accounting makes the decision possible. Without actionable metrics, a team cannot tell whether a strategy is failing slowly or about to work, and the usual response is to persevere too long.

Implication: schedule regular decisions on whether to pivot or persevere, and base them on cohort data, not on hope.

Small batches and engines of growth

Ries borrows from lean manufacturing the case for small batches. His illustration, taken from James Womack and Daniel Jones's Lean Thinking, is a father and two young daughters stuffing newsletters into envelopes. The children want to fold every letter, then seal every envelope; doing one envelope at a time turns out to be faster, because problems surface immediately and no work piles up. For software, this means releasing small changes continuously, so that each mistake is found quickly.

He identifies three engines of growth, each with its own measure. The sticky engine depends on retaining customers: growth is the rate of new customers minus the rate of those lost. The viral engine depends on customers bringing in others: growth becomes self-sustaining when each customer brings in more than one new one. The paid engine depends on revenue: the value of a customer over their lifetime must exceed the cost of acquiring them.

Implication: identify which engine your business runs on, and focus measurement on its one key number.

The adaptive organization

To keep learning fast as a company grows, Ries adapts Toyota's Five Whys: when something goes wrong, ask why five times to reach the root cause, and make an investment at each level proportional to the problem. He warns that the method can turn into the Five Blames if it becomes a search for someone to punish.

He argues that large companies can use the same methods if leaders protect experimental teams. His example is Intuit's SnapTax, which let users file simple tax returns by photographing a wage statement with a phone. A small team built it inside the company with senior management's protection, and it was downloaded about 350,000 times in its first three weeks after the national launch.

Implication: when a problem recurs, trace it to its root cause and fix the system rather than the individual, and give experimental teams inside established organizations the autonomy of a startup.

Limits

Related