Build-Measure-Learn

• You are building something under genuine uncertainty, where nobody can yet know whether customers want it, and the biggest risk is building the wrong thing well.
• A team is heading into months of development on the strength of internal conviction, and you want a cheap test between them and the spend.
• You need a shared discipline for product experiments: a stated hypothesis before building, agreed metrics before launch, and an honest verdict afterwards.
• You face a pivot-or-persevere decision and want it grounded in cohort evidence rather than in the sunk cost of the current strategy.

And when it doesn't

• The requirement is genuinely known, as in a compliance build or a well-specified integration. Iterating towards a fixed specification is waste dressed as learning.
• Experiments are prohibitively expensive or irreversible: implanted medical devices, aviation hardware, one-shot infrastructure. The loop assumes cheap, repeatable trials.
• A visibly unfinished product would do lasting damage to a trusted brand or breach regulation. MVP scope must respect what a market will forgive.
• The team intends to ship regardless. Running the loop without a real willingness to pivot is theatre, and expensive theatre at that.

Before starting, gather the inputs the analysis depends on:

• Explicitly stated leap-of-faith assumptions: the value hypothesis (do customers want this?) and the growth hypothesis (how will it spread?).
• The smallest product that can genuinely test those assumptions, which may be a concierge service or landing page rather than software.
• Access to real customers whose behaviour, rather than politeness, can be measured.
• Actionable metrics agreed in advance, ideally per-cohort, with a stated threshold for what counts as validation.
• Leadership patience for verdicts that kill favoured ideas, without which the loop reports only good news.

1. 1

   State the hypotheses

   Write down the value hypothesis and the growth hypothesis before anything is built. Ries is explicit that the loop is planned backwards: decide what you need to learn, then what to measure, then the minimum thing to build. Teams that start at Build are doing ordinary development with extra vocabulary.

2. 2

   Build the minimum viable product

   Create the fastest artefact that gets you through the loop: a concierge version, a demo video, a single-feature product. The MVP is defined by the learning it enables, and any effort beyond what the test requires is waste, however satisfying to build.

3. 3

   Measure with actionable metrics

   Put the MVP in front of real customers and measure behaviour against the thresholds set in advance. Ries's innovation accounting favours cohort analysis and split tests over cumulative totals, because vanity metrics rise regardless of whether the product is working.

4. 4

   Learn, and call pivot or persevere

   Compare results with the hypothesis and extract validated learning. If the drivers of the model are not moving, schedule the pivot-or-persevere meeting and make a structural change of strategy with the vision held constant: a zoom-in pivot, a customer-segment pivot, a channel pivot.

5. 5

   Go round again, faster

   Feed the learning into the next loop's hypotheses. The unit of progress is validated learning per unit of time, so the operational goal is minimising total time through the loop, which is a property of the whole cycle rather than of the build step.

A stream of validated learning: tested hypotheses with evidence-backed verdicts, a pivot-or-persevere decision trail, and a product shaped by what customers demonstrably do. The loop's health is read in cycle time and in how often evidence changes the plan.

• Judge progress by hypotheses settled per month rather than features shipped per month; the two diverge more often than teams admit.
• Read pivots as findings rather than failures. A loop that has never produced a pivot is either a validated strategy or an unfalsifiable one, and the metrics will say which.
• Watch cycle time. If a single trip round the loop takes a quarter, the MVPs are too big or the metrics too slow, and competitors running weekly loops are learning ten times faster.

• Starting at Build. The loop is planned in reverse, from the learning needed, and a build without a stated hypothesis is just shipping.
• Measuring with vanity metrics. Cumulative sign-ups and page views rise almost regardless of the truth; cohort behaviour against pre-agreed thresholds is what settles a hypothesis.
• Confusing minimum with shoddy for the market in question. MVP scope is relative to what customers will tolerate, and in trust-sensitive markets the minimum is higher.
• Letting the loop stall at Learn. Data gets collected, decks get written, and the pivot-or-persevere meeting never happens; Ries's advice is to put it on a standing cadence.
• Iterating into a local maximum. Small loops optimise what exists, so pair the cadence with occasional bigger bets or the product converges on a polished version of the wrong thing.

• Ries, E. (2011) The Lean Startup. New York: Crown Business. ↗
• Blank, S. G. (2005) The Four Steps to the Epiphany: Successful Strategies for Products that Win. Pescadero, CA: K&S Ranch.
• Blank, S. (2013) 'Why the Lean Start-Up Changes Everything', Harvard Business Review, May 2013. ↗
• Felin, T., Gambardella, A., Stern, S. and Zenger, T. (2020) 'Lean startup and the business model: Experimentation revisited', Long Range Planning, 53(4). ↗

Pairs well with Lean Canvas·PDCA Cycle·OODA Loop·Design Thinking·compare side by side

Near neighbours (computed from shared tags)·Business Model Canvas·Double Diamond·Marketing Mix (4Ps)