The Hypothesize-Design-Test-Learn Loop

Definition

Olsen renames and re-splits the Lean Startup build-measure-learn loop into four steps: hypothesize (formulate problem-space beliefs), design (create the artifact or product that will test them, crossing into solution space), test (expose it to customers and observe), and learn (extract validated learning that revises the hypotheses for the next lap). The change isolates "learn" (what did we observe) from "hypothesize" (what do we now believe), which build-measure-learn collapses into one step.

In the Book

Chapter 10 argues "measure" is too narrow a label because qualitative customer observations — not just statistically significant A/B data — count as valid input, so "test" replaces it; and "learn" actually does two jobs (extracting the finding, then reformulating the hypothesis), which the loop splits explicitly. Olsen compares the loop to passing "Go" in Monopoly: you collect validated learning every time around, and your speed through the loop is usually gated by how fast you can design and test, not by hypothesize or learn. He ties the loop directly back to the Product-Market Fit Pyramid: every revised hypothesis should be mapped to the specific pyramid layer it belongs to, and issues at the lowest layer (e.g., a wrong target-customer hypothesis) should be fixed before issues higher up (e.g., UX polish), since fixing UX won't repair a broken value proposition underneath it. A worked example tracks four waves of user testing where feature-set, UX, and messaging complaints are tallied wave over wave until they drop to zero.

Why It Matters

Separating "what did we observe" from "what do we now believe" prevents a subtle failure mode: silently updating your working theory in the same breath as reading the data, without ever stating the theory explicitly enough to be falsified next time. Pairing that discipline with a layered model of your assumptions (fix the lowest broken layer first) applies to any iterative practice — research, policy, engineering — where cheap surface fixes can mask an unaddressed foundational error.