Business Core · Innovation · Module

Experiments

Breaking an idea into the assumptions it rests on and testing the riskiest one first, with the smallest test that could change the decision.

The idea

How it works

An idea is a bundle of assumptions: that the problem exists, that people will pay, that it can be delivered at a cost that leaves a margin. Each can be wrong, and they are not equally risky. The work is to name them and test the one most likely to end the idea before spending on the others.

The boundary with neighbouring tests is scope. Comparing two versions of an existing interface is handled under UI and UX Optimization, and the statistics of controlled experiments sit in Performance Analysis. Here the question is whether something new should exist at all, and many useful tests are not statistical — a landing page, a manual version run by hand, ten conversations with a price on the table.

Working with it

In practice

  1. 01

    Write the assumptions down

    Desirability, viability and feasibility, each as a sentence that could turn out to be false.

  2. 02

    Rank by risk and by ignorance

    The assumption that would end the idea, and about which least is known, goes first.

  3. 03

    Set the threshold before the test

    What result would count as a pass, agreed in advance. A threshold chosen afterwards will fit whatever happened.

  4. 04

    Record the finding either way

    A failed test that is written down saves someone else from running it again next year.

One level in

The components of experiments

A component is something that exists afterwards which did not exist before — a deliverable or a mechanism, not an intention.

  1. Assumption register

    Every assumption an idea depends on, categorised and ranked by how much damage it would do if wrong.

    Learn
  2. Test card

    One test, written before it runs: the hypothesis, the method, the budget and the pass threshold.

    Learn
  3. Learning record

    What each test found and what was decided because of it, including the tests that failed.

    Learn

If a test could not change what happens next, it is a demonstration. Ask what result would stop the idea before running it.