Why "true"
Reported ROAS counts every order the platform attributes. True ROAS counts only the orders the ads caused. The gap between the two is the share of attributed orders the holdout shows would have happened anyway.
How it's computed
Run a holdout experiment. Compare orders in the exposed audience to orders in the control. Divide the lift by the spend on the exposed audience. The result is the true ROAS.
Why it matters
Reported ROAS rewards channels closest to the conversion. True ROAS rewards channels that actually caused the conversion. Operators who optimise on true ROAS allocate budget to the campaigns that pay their way.