Every advertising platform now gives you more data than anyone can read. That is useful, but it creates a quiet risk: with enough numbers, you can always find one that looks like a win. Optimisation then becomes a series of confident changes based on chance.

These are the habits we use to keep campaign decisions honest.

Decide what "better" means before you look

Pick one primary measure per campaign, and make it as close to business value as you can track reliably. For a lead-generation campaign that is usually qualified leads or cost per qualified lead, not clicks. Secondary measures help explain results, but they should not overrule the primary one.

Write it down before launch. If the goal changes every time you open the dashboard, you are not optimising; you are searching for good news.

Test one meaningful thing at a time

Testing a new headline, a new image and a new audience at once can tell you that something changed, but not what. Change one thing that you believe matters, keep the rest steady, and run both versions at the same time so that the season, the day of the week and the news affect them equally.

Good candidates for early tests are the big levers: the offer, the core message and the audience. Button colours can wait.

Give the test enough time and volume

Small numbers swing wildly. A version that is "40% better" after a few dozen clicks often ends up the same as the other once more data arrives. Before you start, agree on a minimum run time, usually at least one or two full weeks to cover weekly patterns, and a minimum number of the primary result per version.

If a campaign cannot reach that volume, test fewer things for longer, or use a more frequent signal that is closely linked to the primary goal.

Watch for the usual traps

  • Stopping early on a high. Results often look most dramatic in the first days. Wait until the agreed point.
  • Comparing different periods. "This month versus last month" mixes your change with everything else that changed.
  • Trusting one attribution view. Platforms tend to credit themselves. Compare with your own analytics and, where possible, with what customers tell you.
  • Optimising the middle of the funnel only. Cheaper leads that never buy are not an improvement.

A worked example

A campaign runs three creative concepts on Meta. After four days, concept B shows a cost per lead 40% lower than the others — on 23 leads.

That looks like a winner, but 23 leads split across three concepts is far too little to separate a real difference from chance, and the first days of a campaign are distorted by the learning phase. The honest move is to keep all three running to the planned sample size, then compare on qualified leads from the CRM rather than form fills. More often than not, the early leader regresses toward the others.

Move budget gradually

When something clearly works, scale it in steps rather than all at once. Many campaigns perform differently at higher spend as they reach broader audiences. Increasing gradually lets you see where the returns start to fall.

Keep a decision log

For each change, note what you changed, why, what you expected and what happened. It takes minutes, and after a few months it becomes the most valuable document in the account. It stops the team re-testing old ideas and makes it easy to explain results to the rest of the business.

Optimisation is a process, not a moment

Good campaigns are rarely right on day one. They get better through a steady cycle: measure, form a hypothesis, test, learn and adjust. The discipline is in letting the evidence, not the most recent screenshot, decide what happens next.

If you want a second pair of eyes on how your tests are set up, our conversion rate optimisation and marketing analytics teams can review the tracking and test design before more budget moves. Tell us about the campaign.