A campaign that could not be measured

An illustrative scenario, not a customer account. Ravaki is in
closed testing.

The situation

An advertiser runs a campaign across several channels. Each platform reports conversions.
Added together, the platforms claim more conversions than the business actually had.

This is extremely common and it is not fraud. Every platform is claiming credit for touchpoints
that overlapped, under attribution models that are each internally consistent and mutually
incompatible.

Why more reporting makes it worse

The instinct is to get finer-grained data and reconcile. But the disagreement is not about
data; it is about the causal model. Two platforms observing the same conversion will both claim it
under their own rules, and no amount of additional detail resolves which one was decisive.

How we would approach it

1. Stop adding platform-reported conversions together. The sum is not a
   quantity.
2. Hold out a region or an audience segment. A geographic holdout is clumsy,
   cheap, and closer to true than any attribution model.
3. Measure at the business, not the channel. Total revenue against total spend
   over a period is crude and has the virtue of being real.
4. Treat platform attribution as directional. Useful for deciding what to test.
   Not evidence.

What this costs us to say

A platform arguing for holdout testing is arguing for a method that will sometimes show its
own channel did less than its dashboard claimed. We think that is the correct position, and we
would rather be measured honestly than rank well in a comparison nobody believes.