lean-startup（精益创业）

Note: This skill is independent analysis and commentary, not a reproduction of the original text. It synthesizes the book's core ideas with modern startup practice, surfaces where frameworks are outdated or incomplete, and integrates perspectives from adjacent disciplines. For the full argument and context, read the original book.

The Lean Startup

"The Lean Startup is not a collection of individual tactics. It is a principled approach to new product development." - Eric Ries

Should You Use This Skill?

Are you building something under conditions of extreme uncertainty?
|-- YES --> Do you know who your customers are?
|   |-- NO  --> Start with Four Steps (Customer Discovery), use Lean
|   |           Startup for iteration speed within that process
|   +-- YES --> Do you have product/market fit?
|               |-- NO  --> THIS SKILL. Build-Measure-Learn loop.
|               +-- YES --> Use Crossing the Chasm for mainstream scaling
+-- NO  --> Are you optimizing an existing product in a known market?
            |-- YES --> Lean Startup principles apply (small batches,
            |           Five Whys) but you don't need the full framework
            +-- NO  --> Rethink what you're doing

The Core Insight

Most startups fail not because they can't build a product, but because they build something nobody wants. The default response to failure is: we didn't plan well enough, execute hard enough, or have the right vision. Ries argues the real problem is the absence of a management framework designed for uncertainty.

A startup is: a human institution designed to create a new product or service under conditions of extreme uncertainty. This definition applies to garage founders, corporate intrapreneurs, and government innovators alike.

The Five Principles

1. Entrepreneurs are everywhere - any organization creating under uncertainty
2. Entrepreneurship is management - not just "a cool product" but a discipline
3. Validated learning - not "we learned a lot" but learning backed by empirical data
4. Build-Measure-Learn - turn ideas into products, measure response, learn whether to pivot or persevere
5. Innovation accounting - hold innovators accountable with a new kind of accounting designed for uncertainty

Build-Measure-Learn

The fundamental activity loop. Minimize total time through the loop.

         IDEAS
        /      \
       /        \
  LEARN          BUILD
      \          /
       \        /
        DATA--PRODUCT
        (Measure)

Critical insight: Although the loop reads Build-Measure-Learn, you plan in reverse: figure out what you need to LEARN, determine what DATA will tell you that, then BUILD only what's needed to get that data.

"We need to focus our energies on minimizing the TOTAL time through this loop."

Leap-of-Faith Assumptions

Every startup rests on two untested assumptions:

Both must be tested empirically, not assumed. Use analogs (similar successes) and antilogs (similar failures) to sharpen assumptions before testing.

"The two most important assumptions are the value hypothesis and the growth hypothesis."

Minimum Viable Product (MVP)

The MVP is the fastest way to get through the Build-Measure-Learn loop with minimum effort. It is NOT the smallest product. It is the smallest experiment that tests your leap-of-faith assumptions.

WHAT AN MVP IS:                    WHAT AN MVP IS NOT:
- A learning vehicle               - A crappy v1.0
- Tests one specific assumption    - The smallest feature set
- Designed to maximize learning    - A prototype to show investors
- May lack features, polish, UX    - A proof of concept
- Can be embarrassingly simple     - A demo

MVP Types

MVP Quality Concerns

"If we do not know who the customer is, we do not know what quality is."

Customers don't care about quality dimensions you're imagining. Build the MVP, ship it, and let customer behavior (not opinions) tell you what quality means.

Innovation Accounting

Traditional accounting can't measure a startup. Revenue is near-zero. Forecasts are fiction. Innovation accounting provides an alternative.

Three Learning Milestones

1. ESTABLISH THE BASELINE
   |-- Build an MVP
   |-- Get it in front of real customers
   +-- Measure current state of the engine (conversion, retention, revenue)

2. TUNE THE ENGINE
   |-- Run experiments to improve metrics from baseline toward ideal
   |-- Each experiment tests one assumption
   +-- Track whether changes actually move the numbers

3. PIVOT OR PERSEVERE
   |-- Is tuning working? Are you making progress toward the ideal?
   |-- YES --> Persevere. Keep tuning.
   +-- NO  --> Pivot. Change strategy fundamentally.

Vanity Metrics vs. Actionable Metrics

The Three A's of Good Metrics:

1. Actionable - demonstrates clear cause and effect. If you change X, metric Y moves.
2. Accessible - everyone in the company can understand them. Use cohort reports, not cumulative.
3. Auditable - you can trace the data to real humans. Talk to the customers behind the numbers.

Pivot or Persevere

A pivot is a structured course correction designed to test a new fundamental hypothesis about the product, strategy, and engine of growth. It is not failure. It is the mechanism that makes startups robust.

The Pivot Meeting

Hold regularly (monthly or quarterly). Bring product development AND business leadership. Review:

1. Are our experiments moving metrics toward the ideal model?
2. Is our progress sufficient given the time and resources invested?
3. What have we learned about our assumptions?

Ten Types of Pivot

Runway = Pivots Remaining

"The true measure of runway is how many pivots a startup has left."

Not months of cash. A startup that can test more hypotheses before running out of money has a longer runway than one burning cash on a single bet.

Three Engines of Growth

Every startup's growth is powered by one dominant engine. Focus on ONE.

"Startups don't starve; they drown." - in too many simultaneous growth strategies.

Engine Selection

Is your product inherently shareable / visible to non-users?
|-- YES --> Test VIRAL engine first
+-- NO  --> Do customers use it repeatedly (daily/weekly)?
            |-- YES --> Test STICKY engine first
            +-- NO  --> Test PAID engine first

Important: engines eventually run out. When they do, pivot or find a new engine.

Small Batches

Borrowed from Toyota Production System. Smaller batches = faster learning = fewer wasted resources.

"The biggest advantage of working in small batches is that quality problems can be identified much sooner."

The Large-Batch Death Spiral

Large batches look efficient but create a death spiral: the bigger the batch, the longer to test, the more rework, the bigger the next batch needs to be to "catch up," the longer to test...

Pull, Don't Push (from Toyota JIT)

Work in progress is inventory. In startups, features built but not validated are WIP. Only build what's needed for the next experiment.

Five Whys

Adapted from Taiichi Ohno's Toyota Production System. At the root of every technical problem is a human problem.

The Method

Ask "Why?" five times to trace symptoms to root causes. Make a proportional investment at each level - small fix for small problem, bigger investment for deeper cause.

The Five Blames (Anti-Pattern)

When Five Whys goes wrong, it becomes finger-pointing. Prevent this:

• Everyone affected by the problem must be in the room
• Senior people go first with "shame on us for making it so easy to make that mistake"
• Focus on bad process, not bad people
• Appoint a Five Whys master
• Start with a narrow, specific class of problems
• Never start with legacy "baggage" problems

Decision Trees

"What should our MVP be?"

What do you need to LEARN?
|-- "Do customers want this at all?"
|   +-- Smoke test (landing page) or Video MVP
|-- "Will customers pay for this?"
|   +-- Concierge or Wizard of Oz (deliver manually, charge real money)
|-- "Can we build the technology?"
|   +-- Technical prototype (not an MVP - engineering risk, not market risk)
+-- "Which features matter?"
    +-- Single-feature MVP + split testing

"Should we pivot?"

Are experiments moving metrics toward the ideal?
|-- YES, meaningfully --> Persevere. Keep tuning.
|-- YES, but very slowly --> Investigate. Are you out of easy optimizations?
|   |-- YES --> Consider pivot
|   +-- NO  --> Keep tuning, but set a deadline
+-- NO --> Have you exhausted experiment ideas for current strategy?
          |-- YES --> PIVOT. Change a fundamental hypothesis.
          +-- NO  --> Run more experiments, but set a time box.

After pivoting:
- Acceleration test: is the new direction producing faster learning?
- If MVP cycles aren't getting shorter, something is still wrong.

"Are we using vanity metrics?"

Does this metric go up and to the right no matter what you do?
|-- YES --> It's vanity. Switch to cohort-based or per-customer metrics.
+-- NO  --> Can you trace a specific change to movement in this metric?
            |-- YES --> It's actionable. Keep it.
            +-- NO  --> Probably vanity. Test with a split experiment.

Critical Numbers & Rules of Thumb

Common Failure Patterns

Modern Relevance (2011 --> 2026)

Where Lean Startup Still Applies

• Pre-product/market-fit startups of any kind
• Corporate innovation teams testing new business lines
• Hardware and physical products (with longer cycle times)
• Any team that doesn't know if what they're building will work

Where It Shows Its Age

• AI-native products - the feedback loop can be automated in ways Ries didn't anticipate
• PLG/viral-first products - the MVP concept is well-understood; the harder question is distribution
• Hypergrowth VC model - "runway = pivots remaining" conflicts with "blitzscale or die" pressure
• No-code/low-code - building an MVP is now so cheap that the bottleneck is finding users, not building product

What Ries Got Permanently Right

• Validated learning as the unit of progress, not features or code
• Build-Measure-Learn as the fundamental loop
• MVPs as experiments, not small products
• Vanity metrics as the default trap
• Pivots as structured hypothesis changes, not random flailing
• Small batches beat large batches in nearly every context
• Five Whys for proportional investment in root causes

Supporting Files

• frameworks.md - Build-Measure-Learn detailed breakdown, leap-of-faith assumptions, MVP selection, innovation accounting milestones, vanity vs. actionable metrics, pivot catalog, engines of growth mechanics, small batches, Five Whys, innovation sandbox, adaptive organization
• cases.md - IMVU (founding story + continuous deployment), Zappos (Wizard of Oz MVP), Dropbox (video MVP), Groupon (MVP origin), Grockit (innovation accounting), Votizen (3 pivots with metrics), Wealthfront (platform pivot), QuickBooks (large company transformation), IGN Entertainment (Five Whys), SGW Designworks (physical product small batches)
• examples.md - MVP selection worksheet, innovation accounting setup template, pivot-or-persevere meeting template, engine of growth diagnostic, Five Whys session template, cohort analysis template, Build-Measure-Learn cycle planner
• integration.md - Relationship to Four Steps (Lean Startup is direct descendant), relationship to Mom Test (conversation technique for the Learn phase), relationship to Crossing the Chasm (Lean Startup stops at product/market fit), conflicts with $100M Offers (validation-first vs. offer-first), master sequence

Honest Scope of the Book

• Published: 2011
• Examples: Mostly 2004-2010 tech (IMVU, Dropbox, Groupon, Zappos, Votizen). Some are now household names; others pivoted or died.
• Empirical base: Author's experience at IMVU + consulting/advising. Anecdotal case studies, not statistical research. Ries acknowledges this directly.
• Where it shines: Early-stage startups, corporate innovation, any team testing whether something should exist.
• Where it's weak: Post-product/market-fit scaling, marketplace dynamics, deep infrastructure products where MVP approach is dangerous (medical devices, aircraft software). The book is light on HOW to talk to customers (use Mom Test) and silent on Market Type (use Four Steps).
• Intellectual lineage: Direct descendant of Steve Blank's Four Steps to the Epiphany + Toyota Production System (Taiichi Ohno). Ries was Blank's student and implemented Customer Development at IMVU. The Build-Measure-Learn loop owes a lot to Boyd's OODA loop.