Learn Faster Than Anyone Else

The Core Insight

Eric Ries spent six months building an instant messaging add-on at IMVU, shipped it, and nobody downloaded it. The plan promised 300 dollars of revenue in month one, and the number stalled near 500 dollars a month. Teenage testers ran eight IM clients at once and wanted to make new friends, so the team discarded the interoperability code.

The Lean Startup, published in 2011, turns that scar into a method. A startup is a human institution built to create a new product under extreme uncertainty, and its unit of progress is validated learning. Learning counts as validated when a core metric moves and real customer data backs the move. Working code that teaches nothing is waste.

Most founders believe startups win on vision, timing, and heroic execution of a good plan. Ries argues entrepreneurship is management, which means the process can be learned and taught. A good plan needs a long, stable operating history to stand on, and a startup has neither.

IMVU rebuilt around what the failure taught and passed 50 million dollars of 2011 revenue, with more than 60 million avatars.

The Framework

The loop is Build-Measure-Learn, and the planning runs in reverse. Decide what you need to learn, work out what measurement settles it, then build the smallest product that produces the measurement. Every practice in the book serves one clock, total time through the loop.

Before the first turn, strategy rests on leap-of-faith assumptions, and two matter most. The value hypothesis claims the product delivers value once people use it. The growth hypothesis claims new customers arrive through a named mechanism. Facebook cleared both within a month of its February 2004 launch. More than half of users returned every single day, and almost three-quarters of Harvard undergraduates joined with no marketing spend.

Innovation accounting turns learning into milestones a board can audit, in three steps.

• Establish the baseline: ship an MVP and measure real conversion, sign-up, and repeat-use rates, testing the riskiest assumption first.
• Tune the engine: aim every initiative at one growth driver, and count a redesign that fails to move its number as a failure.
• Pivot or persevere: numbers converging on the plan say persevere, and numbers stalling below it say change strategy.

Key Ideas

Sell It Before You Build It

An MVP is the version of the product that completes one full turn of the loop with the least effort and development time. The sizing rule: remove any feature, process, or effort that adds nothing to the learning you seek.

Zappos began as a hypothesis that people will buy shoes online. Nick Swinmurn photographed the stock of local stores, posted the pictures, and bought each pair at full retail when an order arrived. The test produced real payments, real returns, and real support problems, which no survey touches. Sales grew past a billion dollars, and Amazon bought the company in 2009 for a reported 1.2 billion dollars.

Early adopters prefer an 80 percent solution and distrust polish. IMVU's avatars did not walk, so the team shipped instant teleportation, and customers ranked it among their top three likes. The quality rule follows: if you do not know who the customer is, you do not know what quality is.

Fake the Back End, Keep the Measurement Real

Dropbox ran the video MVP. Drew Houston's all-engineer team sold file sync, a problem most people did not know they had. He narrated a three-minute screen recording, seeded with in-jokes for technology early adopters. The beta waiting list went from 5,000 people to 75,000 overnight.

Food on the Table ran the concierge MVP with one customer, one store, and no software. CEO Manuel Rosso visited her weekly with a hand-prepared list and recipes and collected a check for 9.95 dollars. Automation waited until the founders were too busy to add another customer. Ries flags the concierge trap: the hand-run service turns profitable, the growth model fails, and the founder settles for a small business.

Aardvark ran the Wizard of Oz MVP, where customers believe the product works and humans do the work backstage. Max Ventilla and Damon Horowitz tested prototypes in two-to-four-week cycles on one hundred to two hundred invited friends, and the sixth became Aardvark. Humans answered queries for nine months, through eight hires, a seed round, and a Series A. Google bought the company for a reported 50 million dollars.

Cohorts Grade the Work, Totals Flatter It

Vanity metrics are the totals: registered users, cumulative revenue, hits. They rise while the engine turns at all, and they license success theater. Actionable metrics show cause and effect, and the gold standard is the cohort, each period's new customers graded as their own report card.

IMVU bought its cohorts for five dollars a day on AdWords, 100 fresh clicks every day. Of the customers who joined in February 2005, about 60 percent logged in at least once. The share who used the product five or more times grew fourfold, from under 5 percent to almost 20 percent. The share who paid sat near 1 percent the entire time. The same board deck carried a hockey-stick revenue graph drawn from the same period.

Ries asks three things of any metric. Actionable means clear cause and effect, so a rising number credits a change instead of somebody's ego. Accessible means concrete, people-based units, since losing 50,000 customers is legible where a dip in hits is noise. Auditable means anyone who loses an argument can spot-check the report against real customers.

Grockit shows both halves. Farbood Nivi began by teaching test prep over WebEx with no custom software, collecting 10,000 to 15,000 dollars a month. Vanity metrics then hid a flat product until the team moved to cohorts and split tests. One test forced a cohort to register before touching the product, and both cohorts behaved identically on registration, activation, and retention. Most features engineers consider improvements change no customer behavior.

Runway Is the Number of Pivots Left

A pivot is a structured course correction testing a new fundamental hypothesis about product, strategy, or engine of growth. Ries names the biggest destroyer of creative potential: the misguided decision to persevere.

Votizen is the worked example. David Binetti built a social network of verified voters and mapped four leaps of faith to registration, activation, retention, and referral. The first MVP cost about 1,200 dollars and three months, and the funnel read 5 percent registration, 17 percent activation. Eight months and 20,000 dollars of split tests later, retention sat at 8 percent and referral at 6.

He zoomed in: the voter-contact feature became the whole product, and the engine went from sticky to paid. The new MVP took four months and 30,000 dollars and hit 42 percent registration and 83 percent activation. Fewer than 1 percent of users were willing to pay. A customer segment pivot to businesses died when the letter-of-intent companies stalled and passed, because learning pays no salaries. The platform pivot took one month, lifted referral to 64 percent, and found 11 percent of users paying 20 cents a message. Each MVP arrived faster: eight months, then four, then three, then one.

The book names ten pivot types. Four cover most founders.

• Zoom-in: one feature becomes the whole product, the move Votizen ran first.
• Customer segment: the product solves a real problem, for a different buyer than the plan named.
• Engine of growth: the company switches among viral, sticky, and paid, and value capture usually moves with it.
• Platform: an application becomes a platform other people build on, or the reverse.

Runway counted as cash over burn says a million dollars at 100,000 dollars a month buys ten months. Ries recounts it as the number of pivots left, and reaching each pivot faster buys more of them. The pivot-or-persevere meeting goes on the calendar in advance, somewhere between a few weeks and a few months apart. Product development brings metric trends against the plan, and business leadership brings customer conversations.

IMVU skipped its own medicine past a million dollars a month in revenue. Months of individually successful A/B wins moved aggregate activation only a few percentage points, because the early-adopter market neared exhaustion. The mainstream overhaul that followed more than doubled revenue past 25 million dollars a year by 2009. Ries counts the delay as his own failure.

Small Batches Surface Problems Sooner

Lean manufacturing supplies the physics. In the envelope-stuffing race from Lean Thinking, a father finishing one envelope at a time beats his daughters batching by stage. Piles of half-finished work hide defects, add handling, and feel efficient the whole way. Shingo cut Toyota's die changeovers from hours to under ten minutes, and Toyota became the world's largest automaker in 2008.

IMVU ships about fifty product changes every day. Apple boasted more than 1,500 changes at the iPhone 4 release, one batch. Wealthfront pushes more than a dozen releases a day inside an SEC-regulated investment business.

Software batches grow, because no physical limit caps them and moving one forward creates rework, until a release becomes a bet on the company. The guard is an immune system that watches business metrics, since a checkout button rendered white-on-white passes every functional test while sales die. A bad change reverts on its own, and new work stops until the root cause is fixed.

Pick One Engine and Read Its Number

Sustainable growth obeys one rule: new customers come from the actions of past customers. Three engines organize that, and startups run one at a time.

The sticky engine compounds at the natural growth rate minus the churn rate. One Ries client retained 61 percent of customers while adding 39 percent new, a compounding rate of 0.02 percent. The prescription ran against instinct: improve retention instead of buying more customers.

The viral engine spreads as a side effect of normal use, and the coefficient sets the speed. At 0.1, one hundred customers bring ten, and the loop fizzles. Above 1.0, growth turns exponential, and a coefficient of 0.9 or better puts a startup on the verge of success. Hotmail crawled until the founders appended a one-line postscript inviting recipients to get free e-mail. Within six months the line added more than a million users. Eighteen months after launch, Microsoft paid 400 million dollars for 12 million subscribers.

The paid engine runs on the spread between lifetime value and acquisition cost. An ad buy of 100 dollars that signs up fifty customers costs two dollars each, and the engine turns while lifetime value clears that. IMVU planned on viral growth, but its customers wanted new friends and had no reason to invite old ones, so it switched engines. Its teenagers looked unlucrative until the company billed through mobile phones and took cash in the mail, a monetization edge rivals lacked.

Practical Applications

Write both hypotheses down before the next sprint. The value hypothesis names a repeat behavior you can count. The growth hypothesis names the mechanism that turns customer one into customer two.

Buy a baseline with the cheapest MVP that tests the riskier hypothesis. A preorder page is a smoke test. A demo video, a hand-served customer, or humans behind a curtain all count as products, provided each produces a number.

Instrument cohorts before adding features. Grade each week's signups on the same three or four behaviors and read the columns side by side.

Launch every feature as a split test with a named hypothesis, and remove features that fail their test. Grockit's board rule: no story counts as done until it produces validated learning.

Who This Is For

Founders before product-market fit get the most, because every chapter assumes the customer is still unknown. The startup definition carries no size clause, so the method serves teams inside large companies too. A five-person team inside Intuit shipped SnapTax this way, against the company's own flagship.

Skip it if you are cloning a proven business, a case Ries excludes because success there rides on execution. Operators past fit need scaling machinery this book does not carry. Readers fluent in the vocabulary still gain from the exact mechanics, because the slogans shed the numbers long ago.

Read the cases with the disclosures page open. Ries discloses a relationship or an equity interest in eight of the principal case companies. Several headline outcomes are reported or rumored rather than audited, the Dropbox valuation and the Aardvark price among them. The cases aged unevenly after 2011: Google shut Aardvark down, Groupon shrank after its IPO, and IMVU stayed a niche business. The method also has a failure mode the book itself names as pseudoscience: learning cited as an excuse for shipping nothing.

The Decision

The test fits inside one week and needs no new code. Write the single leap-of-faith sentence your product depends on, in present tense, as a claim about customers who already exist. Name the number that proves it wrong and the day you will read it.

Buy the reading with the smallest artifact that counts as a product: a preorder page, a screen recording, one customer served by hand. If the number moves toward the plan, persevere. If it stalls, the week bought you a pivot. Zappos started with photographs of someone else's shoes.