Most marketing teams don’t lack ideas for experiments. They lack the discipline to run them well.
Three rules separate the teams that consistently win.
1. Pre-register the hypothesis
If the test design doesn’t state, before launch, what success looks like, the team will reframe success after the fact. That’s human, and it’s also why most A/B results don’t replicate.
Write it down before you launch, in one sentence, with a number in it. "Variant B will lift lead-to-SQL by at least 10% over two weeks." That’s a hypothesis. "We think this hook will perform better" is not.
Marketeer does not hold a hypothesis field, and it will not stop you creating a test without one. That part is on you: put the sentence in the test name, or in the campaign brief, so it is in front of you again when the numbers land.
2. Pick one primary metric
Tests with three primary metrics aren’t one test, they’re three tests with shared traffic. The math falls apart.
Pick one. Track the others as guardrails: secondary metrics that can fail the test but can’t pass it.
Marketeer asks for it when you set a test up: "What counts as a win? e.g. orders". Leave it blank and it records "Conversions" and carries on, which is a sensible default and a bad habit. Type the metric you actually care about.
3. Pre-set the stop-loss
Auto-stop rules protect spend. "If CPL on any cell exceeds £140 with at least 1,200 impressions, pause that cell" is a stop-loss rule. It runs without the team thinking about it.
Without stop-loss, losing cells consume budget that should be flowing to the winning hook. Stop-loss is the cheapest decision optimisation in marketing.
Be clear about where that rule lives today: on the ad platform where the money is actually spent, and in your own head. Marketeer has no stop-loss field and cannot pause a cell for you. Set the rule in the ad account when you launch, and write the number into the brief so the next person can see it.
What changes when this is real
Teams that run ≥4 disciplined tests per quarter compound. Their winning hook from Q1 anchors Q2 messaging. Their losing audience from Q2 prunes their Q3 exclusion list. Their best-performing channel mix from Q3 informs the Q4 budget.
The teams that run undisciplined "let’s try this" tests don’t compound. The data doesn’t survive past the campaign.
Marketeer’s experiments step is built to make these rules easy to keep, and it enforces exactly one of them. Before you start, the planner tells you how many people you need per version and roughly how many weeks that will take at your traffic, so you size the test instead of guessing. Afterwards, it will only call a winner at 95% confidence with at least 100 people per version, and it shows you the range the true result is likely to sit in. Below that bar it says there is no clear winner yet, however much you want one. That is the guard rail that stops you crowning a fluke, and it is the only one the software imposes.
The other two, the written hypothesis and the stop-loss, are yours. No field asks for them and nothing blocks you without them. That is honest, and it is also the point: the discipline is a habit, not a form. Keep it and you compound.