A jewellery brand with a wide catalogue

The story

Meta was not behaving consistently. A campaign would do well for a while, and then the numbers would move: CPM, click-through rate, CPC and cost per acquisition. The first reaction in a performance account is that something is wrong and needs fixing. We did not want to react every time the dashboard moved, because fluctuation is not automatically a problem. Meta is an auction, and the cost of reaching an audience changes with competition, demand, placement, creative fatigue and season.

The first thing we did was separate noise from real performance problems. Instead of comparing yesterday with today, we looked across longer windows and at the relationship between metrics. Was CPM rising? Was click-through rate falling? Was CPC rising because the creative had weakened? Or was traffic healthy while the website’s conversion rate fell? Those are four completely different problems.

We stopped treating ROAS as the diagnosis. ROAS is the output; it tells you what happened, not why. So we read the account down the funnel: CPM, click-through rate, CPC, landing page view, add to cart, checkout, purchase, cost per acquisition and ROAS. If CPM rose but click-through and conversion held steady, pausing the campaign because ROAS dipped would have been the wrong decision. If click-through fell while frequency climbed, that pointed to creative fatigue. If Meta’s numbers looked healthy but purchases did not arrive, we looked at the website, the offer, checkout and tracking instead of blaming Meta.

The biggest change was how we managed the account. Not: ROAS is down, so change the campaign. Instead: what changed, why did it change, is the change persistent, and is there enough data to act? Every unnecessary change resets something. Change the audience, the creative, the budget, duplicate the campaign, pause it and launch another, and eventually you are not optimising Meta. You are interfering with it.

The catalogue gave us another lever. Not every product deserves equal media support. Some attract attention, some generate clicks but do not convert, some convert well at low volume, and some work best as combinations. So we judged performance at the level of product, product set, creative angle and new against returning customers, and asked what had earned more money behind it. Keep what works, fix what can be fixed, and stop what consistently does not.

The lesson is that the numbers will not always go up, especially on Meta. The job is not to make the graph straight. It is to know when to act, and to make decisions commercially even when reading the account technically. Meta does not care about your ROAS. Your profit and loss does.