Why “the algorithm changed” is almost never the answer

When performance drops and nobody can say why, the explanation that arrives is usually the platform. It is comfortable, it is unfalsifiable, and it is occasionally even true. It is also the most expensive first guess available, because it is the only one on the list you cannot act on.

The useful discipline is to rule out the cheap explanations in cost order. Here they are, roughly from cheapest to most expensive to check.

1. Something in the measurement moved

Before believing that performance changed, establish that measurement did not. A deduplication setting, a consent banner rollout, a tag that stopped firing on one template, a domain change that reset attribution: any of these produce a clean step change in reported results with no change in the underlying business at all.

The tell is a step rather than a slope, and a step that lands on a deploy date rather than a spend change. If your reported conversions fell 30% overnight and your order count did not, the platform is not the problem and neither is your creative.

2. The mix changed underneath a flat total

Account-level acquisition cost is an average, and averages move when the weights move even if nothing inside them does. A campaign that took a larger share of budget without anyone deciding it should, a new geography entering rotation, a retargeting audience shrinking as a promotion ends: each moves the blended number while every individual component holds steady.

This one is worth checking second because it is pure arithmetic and takes minutes. Break the total into its parts, hold last month's weights constant, and see whether the change survives. Often it does not.

3. You are in a learning phase you caused

Edits reset optimisation. Most buyers know this and most accounts still carry a week of self-inflicted volatility every month, because the edits are small and individually reasonable and nobody counts them together.

The question is not whether you made a significant edit. It is how many edits, across how many ad sets, in the last fourteen days. A list of every change with its timestamp answers it immediately, and almost nobody keeps one.

4. Creative is doing what creative does

Fatigue is the most common real cause and it is fourth on this list rather than first, because it is the most expensive of the four to diagnose properly. Doing it properly means separating fatigue from audience saturation, which look similar at the account level and call for opposite responses.

The distinction is where the frequency sits. If two or three variants carry most of the impressions, the audience is fine and the creative is tired. If frequency is spread evenly across a healthy variant count, the creative is fine and the audience is too small. Shifting budget fixes the first. Broadening targeting fixes the second. Doing the wrong one makes the number worse and costs you the week it takes to find out.

And then, sometimes, it is the platform

Platforms do change. Delivery systems get retuned, auction dynamics shift, an inventory source opens or closes. When it happens it usually shows up across every account in a vertical at once, which is precisely why it is the hardest explanation to confirm from inside a single account.

So the honest form of the conclusion is not that the algorithm changed. It is that the four cheaper explanations have each been checked and ruled out, on this date, with these readings. That is a statement someone can disagree with, which is what makes it worth writing down.