# Design a quick experiment to test a hypothesis about a work problem

Turns a hunch about a problem or fix into a testable hypothesis and the fastest, cheapest experiment that gives a clear answer.

- Topic: Thinking & Decisions
- Framework: Few-Shot
- Level: Advanced
- Tags: Write, Brainstorm, Fill-in Template
- Source: https://www.thinkwithprompt.com/prompts/design-a-quick-experiment-to-test-a-hypothesis-about-a-work-problem

## Prompt

```text
Help me test an idea about a problem with a fast, cheap experiment before we commit to a big fix.

**The problem:** [WHAT IS GOING WRONG, with any numbers we have]
**My hunch about the cause or the fix:** [e.g. "onboarding calls would cut early churn", "approvals are slow because forms are incomplete"]
**What we can change or try:** [LEVERS WITHIN OUR CONTROL]
**Data we can measure:** [METRICS, SYSTEMS, HOW OFTEN]
**Time and resources available:** [e.g. "2 weeks, one person part-time"]

Steps:
1. Write the hunch as a testable hypothesis: "If we [change], then [metric] will [move by roughly how much] for [group], because [reason]." Offer 2-3 alternative hypotheses if the hunch is vague.
2. Design the fastest experiment that could prove it wrong: a pilot with one team, a before-and-after check, a split between two groups, a manual mock-up of a solution, or a quick data pull. Say which design fits and why.
3. Define success before starting: the primary metric, the result that would confirm or reject the hypothesis, and a guardrail metric that must not get worse.
4. Point out what could distort the result (small sample, seasonality, people knowing they're being tested, other changes happening at the same time) and how to reduce it.

Return:
1. Hypothesis.
2. Experiment design: what we do, with whom, for how long, and who runs it.
3. Success and failure criteria, set in advance.
4. Risks to a clean result.
5. What we'll do next if it works, if it fails, and if it's unclear.

Rules: keep it small enough to start this week. Don't invent baselines; mark them [ADD BASELINE]. If the experiment affects customers or employees, check it with the right owner first.
```

## What to fill in

- WHAT IS GOING WRONG, with any numbers we have
- LEVERS WITHIN OUR CONTROL
- METRICS, SYSTEMS, HOW OFTEN
- e.g. "2 weeks, one person part-time"
- change
- metric
- move by roughly how much
- group
- reason
- ADD BASELINE

Free to copy and use. From ThinkWithPrompt (https://www.thinkwithprompt.com/), a library of 2,663 curated AI prompts.