Campaign-attributed revenue vs incremental revenue: how do you measure win-back lift?

I’m working through the measurement layer of a lifecycle recovery framework.

The question:

If Klaviyo attributes revenue to a win-back flow, how much of that revenue was actually incremental?

My preferred structure, where audience size allows it, would be:

→ define eligible customers
→ random treatment / holdout split
→ treatment receives the recovery workflow
→ holdout receives no recovery intervention
→ compare repeat purchase rate
→ compare revenue per recipient

The attached example shows the framework using entirely synthetic data.

It is NOT a real client result.

The goal is to separate:

“Revenue happened after the campaign”

from:

“The campaign actually changed behavior.”

For people running real lifecycle experiments:

What minimum audience size / test duration have you found practical before a holdout becomes too noisy to interpret?