Why does my ad performance get worse right after I change something?

Because most ad platforms put a campaign into a learning phase after a significant edit, and during it performance is genuinely unstable and usually worse.

What the learning phase is. Delivery is driven by a machine learning model predicting which users are most likely to convert. That model needs outcome data to calibrate. When you make a significant change, the platform treats the setup as new and the model must gather fresh evidence — deliberately exploring a wider range of users, at higher and less consistent cost, to learn what works.

Meta typically describes exiting the learning phase at around 50 optimisation events within a seven-day window. Google's systems behave similarly without always naming it.

What resets it. Not everything, but more than people expect: changing the optimisation goal, significantly changing budget or bid, editing the audience, adding or swapping creative, and changing the ad set structure. Small budget adjustments generally do not; large ones do.

The most common self-inflicted problem is reacting to poor early performance by making another change, which restarts the learning phase, which produces more poor performance. Campaigns can be held in permanent learning this way, never accumulating enough data to stabilise. Platforms flag this as learning limited.

Practical guidance:

Wait. Give a change at least the full learning window before judging it, unless spend is genuinely running away.

Batch your changes rather than editing daily.

Consolidate. Splitting budget across many small ad sets means none accumulates enough events to exit learning. Fewer, better-funded ad sets usually outperform many thin ones.

Use a realistic optimisation event. Optimising for a conversion that happens twice a week cannot generate enough signal; optimise for an earlier, more frequent event instead.

Seasonality and auction competition also move performance independently of anything you did.