Most marketing "experiments" aren't experiments. They're campaigns with a hopeful name attached. A real one has a hypothesis you could be wrong about, and that single difference is what turns activity into compounding knowledge.
Here's the framework I use to keep tests honest. Save it for Monday.
Five steps to experiments that pay
- Start with one hypothesis, not a campaign brief: not "let's try a loyalty programme." Instead: "If we add a QR code to the bill, at least 10% of anonymous customers will scan it and identify themselves." Specific, measurable, and capable of being proven wrong.
- Pick the smallest possible test: one channel, one segment, one location, one week. The goal is to learn fast, not launch big. If it works small you scale it. If it fails small you learned cheaply.
- Always have a control group: no control, no learning. You need a group that doesn't receive the intervention, or you're measuring activity and calling it impact.
- Measure what matters commercially: not open rates, not impressions. Revenue per head, repeat visit rate, cost per acquisition, margin. If a test can't be tied to a commercial outcome, the test is the problem.
- Document and share the result: win or lose, write it up in one paragraph. What you tested, what happened, what you'll do next. This is how a team builds institutional memory instead of relearning the same lesson every quarter.
💡The discipline isn't in running the test. It's in being willing to write down a prediction you might get wrong, in front of other people.
That willingness is what separates teams that compound from teams that just stay busy. Each documented test makes the next one sharper. Skip the rigour and every quarter starts from zero, however much you spend.
🧪Test small. Learn fast. Scale what works.
