Design your first growth experiment

Begin with an observation

A useful experiment starts with something you can point to: visitors arrive on a service page but rarely enquire, a newsletter gets replies but few clicks, or a community repeatedly asks a question your site could answer.

Write the observation before proposing a fix. Check that the data or feedback actually supports it. “Our headline is bad” is an opinion; “three people could not tell what the service includes” is an observation you can investigate.

Write one testable hypothesis

Use this structure: “If we change __ for __, we expect __ because __.” For example: “If we explain what happens after an enquiry, we expect more suitable visitors to complete the form because the next step will feel clearer.” This is a hypothetical example, not a reported result.

Choose one primary measure and one guardrail. Completed enquiries might be the measure; the share of enquiries that are relevant might be the guardrail. That helps you notice if a change increases the number while reducing its usefulness.

Set the boundaries before starting

Decide the audience, the page or message, the effort limit, and the review date. Keep other major changes steady when practical. Record anything else that changes during the test, such as a promotion or a new traffic source.

Do not set an arbitrary short deadline and assume it creates enough evidence. A low-traffic site may learn more from a usability session and a longer observation period than from an underpowered A/B test. Choose the method to fit your traffic and question.

• Question: what are you trying to learn?
• Change: what exactly will be different?
• Measure: what outcome will you count?
• Guardrail: what must not get worse?
• Review: when and how will you decide?

Keep a record you can learn from

At review time, write the actual counts and the context. Note what you changed, what else happened, and what remains uncertain. A before-and-after comparison can suggest a direction, but it cannot isolate cause when audience, season, or other conditions also changed.

Choose to repeat, revise, pause, or collect more evidence. “Inconclusive” is a legitimate result. Resist rewriting your original hypothesis to make the outcome look successful.

Save the learning where you will see it before your next campaign. Even a test that does not improve the metric can reduce uncertainty about the audience or reveal a problem in the journey.