The credit assignment trap
Attribution models allocate credit across touchpoints that were observed. They cannot see the conversions that would have occurred without any advertising at all, so they systematically flatter retargeting and branded search.
Incrementality as the tiebreaker
Geo holdouts, PSA tests and switchback experiments answer a different question: what changed because we spent. Run them quarterly on your largest line items and accept the answer even when it is unflattering.
A workable stack
Attribution for daily optimisation, marketing mix modelling for quarterly allocation, experiments to calibrate both. Anything simpler is a guess with a dashboard.