The Lean Startup

Mental model

A startup is “a human institution designed to create a new product or service under conditions of extreme uncertainty” — the uncertainty is what defines it, not the size, the age, or the garage. Under that uncertainty a detailed plan is theatre, because you cannot forecast a market that does not exist yet. So the unit of progress stops being features shipped and becomes validated learning about customers, and every practice below is machinery for buying that learning cheaply.

Hypotheses, not ideas

Leap-of-faith assumption	The question it settles
Value hypothesis	Does this deliver value once used?
Growth hypothesis	How do new customers find it?
Customer segment	Who has this problem badly?
Willingness to pay	Will they give up money or time?

Steve Blank puts the correction bluntly: “new ventures don’t start with ‘ideas’, they start with hypotheses (a fancy word for guesses)”. His frame for a startup is a temporary organization searching for a repeatable and scalable business model, and customer development is “the process of how you get out of the building and search for the model”. Write down the two assumptions that kill the company if they are wrong, and test those first — not the ones that are easiest to test.

Gotcha: Interest is not evidence. “Would you use this?” is answered by politeness, and everyone says yes. Ask what they did last time they had the problem, or ask for a commitment — an email, a deposit, an hour of their time.

Build-Measure-Learn

Stage	Produces
Build	The smallest test of one hypothesis
Measure	Data on what people actually did
Learn	A pivot-or-persevere decision

The loop runs build, then measure, then learn — but it is planned in reverse: decide what you need to learn, derive the metric that would show it, then build the least thing that produces that metric. The number to optimize is total time around the loop, not the speed of any one arc. Blank’s summary of the intent is “to maximize learning through incremental and iterative engineering”, which he restates as hypotheses, experiments, tests, insights.

Plan backwards        Run forwards
  Learn what?           Build
  Measure what?   ->    Measure
  Build what?           Learn

Optimize the whole lap, not one arc.

Gotcha: Teams instrument the Build arc alone — velocity, cycle time, deploys per day — and leave a two-week reporting lag in Measure and a quarterly review in Learn. Halving build time then barely moves the loop it was supposed to shorten.

Minimum viable product

MVP type	What it tests
Landing page, ad smoke test	Demand, before anything exists
Concierge	The service, delivered by hand
Wizard of Oz	Real interface, manual behind it
Video	Whether people grasp and want it
Single-feature build	Value of the one core action

The MVP is “that version of a new product which allows a team to collect the maximum amount of validated learning about customers with the least effort”, and Ries is emphatic that “MVP, despite the name, is not about creating minimal products”. Minimum is measured against the hypothesis, not against a feature count — IMVU’s first MVP took six months. Blank’s version is the useful test in the room: the simplest thing you can show customers to get the most learning at that point in time.

Gotcha: An MVP that produces no measurement is not an MVP, it is just a small product. Ries’s own example is a feature that took two weeks to build when “a simple AdWords smoke test would have revealed how utterly bad the concept was”.

Validated learning

Activity	Progress?
Feature shipped on schedule	Only if it settled a hypothesis
Hypothesis disproved by data	Yes — a costly path closed early
Revenue up, cause unknown	No, nothing was learned
Roadmap delivered in full	This is how you achieve failure

Validated learning is “a rigorous method for demonstrating progress when one is embedded in the soil of extreme uncertainty”, and the validation “comes in the form of data that demonstrates that the key risks in the business have been addressed”. Ries states the priority plainly: “Given a choice between what a successful team has learned and the source code they have produced, I would take the learning every time.” The failure mode all of this exists to prevent is achieving failure — executing a plan faithfully, on time, on budget, for a product nobody wants.

Gotcha: A disproved hypothesis is a successful experiment, but it reads to the organization as a missed commitment. If the review only rewards shipped scope, teams will stop running experiments that can come back negative — which is all the useful ones.

Actionable metrics

Vanity metric	Actionable replacement
Total registered users	Activation rate, by cohort
Page views	Retention by signup week
Gross revenue	Revenue per customer, per cohort
Press mentions	Conversion from that channel

Vanity metrics are “numbers which look good on paper but aren’t action oriented: website hits, message volume, or ‘billions and billions served’”. They only go up and to the right, so they cannot tell you whether the last change helped. Ries asks that every metric be actionable, accessible, and auditable, and that cause and effect be established by cohort analysis and split testing rather than by reading a cumulative chart.

Gotcha: Gross numbers invite a bias Ries names directly — “when the numbers go up, we tend to take credit. But when the numbers go down, we tend to blame someone or something else.” A per-cohort rate removes the argument, because each cohort’s curve stands alone.

Innovation accounting

Milestone	The work
Establish the baseline	Ship an MVP, measure where you are
Tune the engine	Move that baseline toward the ideal
Pivot or persevere	Judge whether tuning still pays

Conventional milestones assume you know the destination, and a startup’s metrics are all effectively zero at the start — so innovation accounting substitutes learning milestones: how to measure progress and prioritize work while the business model is still a guess. The baseline is deliberately unflattering, being the honest current conversion, retention, or price. Tuning means every experiment must move that number, and the size of the moves tells you whether the hypothesis has life in it.

Gotcha: A team that keeps hitting small improvements can still be failing. If a dozen experiments each add a fraction of a percent and the gap to a viable business is a factor of ten, the engine is tuned and the hypothesis is still wrong.

Engines of growth

Engine	The number that governs it
Sticky	Retention above churn rate
Viral	Viral coefficient above 1.0
Paid	Lifetime value above blended CPA

Growth is not a mood; it is one of three mechanisms, each with a single governing number. Sticky growth compounds when new customer acquisition outruns churn. Viral growth needs a coefficient above 1.0, because “if the coefficient is > 1.0, you generally have a viral hit on your hands”. Paid growth lives on the spread, which “determines either your profitability or your rate of growth”. Ries’s instruction is to specialize: “Every startup needs to ‘pick a major’ among these three drivers of growth. It’s simply too hard to focus on more than one.”

Gotcha: Word of mouth is not the viral engine. Viral means transmission is a necessary byproduct of ordinary use, the way sending mail once carried Hotmail’s signature — happy customers recommending you is sticky growth wearing a flattering name.

Pivot or persevere

Pivot	What changes
Zoom-in	One feature becomes the whole product
Zoom-out	The product becomes one feature
Customer segment	Same product, a different buyer
Customer need	Same buyer, a different problem
Platform	Application becomes platform, or back
Engine of growth	Sticky, viral or paid is swapped

A pivot is “a structural course correction to test a new fundamental hypothesis about the product, strategy and engine of growth” — the vision usually survives it, the strategy does not. What separates a pivot from a restart is continuity: “successful startups change directions but stay grounded in what they’ve learned. They keep one foot in the past and place one foot in a new possible future”, where unsuccessful ones “jump outright from one vision to something completely different” and discard the learning they paid for. Value capture, channel, technology and business architecture pivots complete the book’s catalogue. Schedule the decision as a standing meeting, so it is taken on evidence rather than on exhaustion.

Gotcha: Runway measured in months is the wrong unit — the real question is how many pivots you can still afford. Cutting burn buys more of both, but shortening the loop increases the runway without additional cash, which is the cheaper lever.

Small batches and Five Whys

Practice	Effect
Small batches	Defect found next to its cause
Continuous deployment	The loop closes in hours
Five Whys	The fix lands on the cause
Proportional investment	Response sized to the damage

The method inherits its factory floor from lean manufacturing: work in small batches, stop the line when something is wrong, and treat unused inventory — features nobody asked for — as waste. Five Whys is the tool for the stop: ask why five times, on the premise that behind every supposedly technical problem is a human problem, and always include “why didn’t our tests catch and prevent the problem?”. Then invest at every level in proportion to the damage — “don’t do too much, and don’t do nothing” — with everyone affected in the room.

Gotcha: Five Whys turns into five blames the moment a name arrives before a system does. Ries offers no shortcut — “there’s no easy fix to this problem. Trust takes time to build up” — so start with small problems and let the practice earn its credibility.

Further reading

• The Lean Startup principles
• Minimum Viable Product: a guide
• Validated learning about customers
• The three drivers of growth
• Pivot, don’t jump to a new vision
• Beware of vanity metrics
• How to conduct a Five Whys analysis
• Steve Blank on Build, Measure, Learn