How do I know if this is actually working?

At the account level, compare ad spend, ad-attributed sales and ACOS over mature, comparable windows. Add TACOS only when matching total-sales data is available, and consider contribution margin, stock, price changes and promotions.

At the change level, distinguish Verified from effective. Verification confirms that an expected change is present on Amazon. Impact's kept, reverted and pending records describe the evaluation outcome under your rules; they do not prove a causal profit gain.

Look for a repeatable pattern with enough data, rather than treating one good week or a fixed number of weeks as proof. Record what else changed in the account. Our worked audit example shows how to keep observed evidence, a proposed action and measured results separate.