Lean Startup

Lean Startup

Overview

A startup is a temporary organization searching for a repeatable, scalable business model under extreme uncertainty (Steve Blank). Most early-stage failures are from building something no one wanted because the demand assumption was never tested.

Eric Ries (2011): name the riskiest assumption, build the smallest test (MVP), measure real behavior, decide to pivot or persevere — the Build–Measure–Learn loop, run as fast as possible.

Compose: first-principles to find what the model truly depends on; probabilistic-thinking to calibrate experiments; inversion before each Build phase; business-model-canvas to surface the riskiest assumption blocks.

When to Use

Apply when: high uncertainty + limited capital; a team is about to build before testing demand; a pivot-or-persevere decision is on the table; you're building an AI feature on a foundation-model API and worried "the next model release will commoditize us" / "are we just a GPT wrapper?"; no clear answer to "what is the load-bearing assumption and how would we know if it's wrong?"

When NOT to use: known business model in known conditions (execution, not search); decision is not business-model-level; cannot ethically run a test with real customers; using "lean" as a schedule excuse to ship buggy software.

Coaching Novices (Adaptive Front Door)

• Engine mode: user has a concrete hypothesis → run The Process directly.
• Coach mode: no concrete hypothesis or signals unfamiliarity → guide step by step.

In Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.

1. One-line what-it-is. Most startups fail by building before knowing if anyone wants it; lean startup names the riskiest assumption, tests it with the smallest MVP, measures real behavior, and decides pivot or persevere — fast.
2. Check fit. Match against When to Use / When NOT to use; if low uncertainty + known model, redirect.
3. Elicit their real hypothesis. Force them to name one load-bearing assumption — specific segment, specific value, specific willingness-to-pay.

[WAIT — do not advance until user responds]

1. Walk the loop step by step. Name assumption → design MVP → define metric → set threshold. Pause at each.

[WAIT — do not advance until user responds]

1. Close by naming the next-week experiment. One assumption, one MVP, one threshold, one date — not a strategy doc.

[WAIT — do not advance until user responds]

The Process

Run the Build–Measure–Learn cycle. Identify, test, decide.

1. State the load-bearing assumption. Specific segment, specific value, specific willingness-to-pay, specific timeframe. Not "users want X."
2. Pre-commit to a pivot-or-persevere threshold. Write the metric value before running the experiment. You will rationalize if you have not pre-committed.
3. Design the smallest MVP that tests the assumption. Often not a product — a landing page, concierge/"Wizard of Oz" version, or 3-minute video. Purpose is learning, not selling.
4. Build the MVP fast. Time-box. If an early-stage test takes more than 4–6 weeks, cut.
5. Measure real customer behavior, not stated intent. Actionable metrics (conversion, retention, willingness-to-pay) test the assumption. Vanity metrics (signups, likes) do not.
6. Compare result to the pre-committed threshold. Don't move the goalposts.
7. Decide pivot or persevere — explicitly. Persevere = assumption held; pivot = assumption failed in a specific way, change the load-bearing block and re-test.
8. Document and iterate. Write: assumption, MVP, threshold, result, decision, rationale. Each loop must produce a durable carry-forward learning.

Output: Experiment Card

Assumption: " will  at  for  by "
Threshold: Persevere if  | Pivot if 
MVP: 
Metric:  | Vanity to ignore: 
Result: 
Decision: [ ] Persevere  [ ] Pivot (type: ___)  [ ] Re-test
Validated learning: 

→ Method in Action: Dropbox's Video MVP (2007) · Votizen's Pivot Sequence (2010–2011) → 2026 lens: AI-native lean startups (2023–2026) — when the next model release commoditizes your AI feature, that's an invalidated assumption, not bad luck

Experiment Packs

Applying It Well

• MVPs are for learning, not revenue — the deliverable is evidence, not a launch.
• Pre-commit to the threshold or you will rationalize whatever you get.
• In B2B, talk to buyers (hold budget), not just users (love the product).
• Vanity metrics (signups, likes) ≠ actionable metrics (conversion, retention, willingness-to-pay).
• The MVP is disposable — a test instrument, not v0 of your product.

→ Primary sources: references/sources.md

Common Rationalizations

[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.

Red Flags

• The team is building for months with no MVP yet shipped
• "MVP" is a six-month build with full polish
• Vanity metrics dominate the dashboard; conversion/retention/willingness-to-pay are absent or untracked
• Customer interviews reported as "they love it" with no behavioral data
• Pivot decisions made on a single week's noise, or after founders simply got bored
• No pre-committed pivot/persevere threshold exists for any experiment
• "Lean" is being used to justify low-quality shipping rather than test-before-build

Verification

• The load-bearing assumption is named in specific segment/value/willingness-to-pay/timeframe form
• The pivot-or-persevere threshold is pre-committed in writing, before the experiment runs
• The MVP is the smallest test of the assumption (time-boxed ≤ 4–6 weeks early-stage)
• An actionable metric (not vanity) is pre-specified for evaluation
• Result is compared to the pre-committed threshold — without moving goalposts
• Pivot vs. persevere decision is made explicitly, with type if pivoting
• Validated learning is documented in one sentence carry-forward

---

Part of deciqAI Knowledge Skills — 227 open-source thinking skills that make rigor executable for AI agents. The same skills power every deciqAI agent, which runs them autonomously to operate your company. See it run → https://www.deciqai.com/c/lean-startup · ⭐ Star the repo → https://github.com/deciqAI/knowledge-skills · Contributions welcome.

Agents: latest version & machine-readable metadata → https://www.deciqai.com/s/lean-startup.json