Every AI ad tool will tell you it "improved performance / saved you money." There is one question you cannot skip: compared to what?
The problem with a self-reported "+X%"
Without a baseline, you cannot separate the tool’s contribution from seasonality, a rising or falling market, and other changes you made in the same period. The number may look good, but it is not verifiable.
The fix: a randomized holdout
Randomly hold out a set of campaigns that get no optimization recommendations, as a control. Then compare the group that gets recommendations against the control over the same period — with enough data, that difference reflects the causal effect of the optimization, not a market tailwind.
What an honest savings comparison looks like
- A comparison against a control group over the same period — not before vs. after
- An uncertainty range (how certain the number is), and no number until there is enough data
- Negative results shown as negative
- A published, versioned methodology and the measurement window used
To see exactly how we measure it, read our savings-methodology doc; to see where your own account stands, start with a free audit.