Should you trust in-platform attribution models?
TL;DR: probly not.
Attribution is tricky. Last-click attribution is also, uh, sketchy. Why?
It often confuses correlation with causation. Did views CAUSE sales or just correlate with them?
It rarely accounts for 10 or 12 other factors that are likely involved.
It’s got conflict-of-interest built in: of COURSE a platform wants to claim brand lift or sales impact or whatnot.
Econometric models are better (the good ones are). They factor in LOADS of things to tease apart effects.
Is it using correlation to assume causation? Mostly yes. But it’s miles better than naïve attribution models.
magic numbers compared 5 years of Google attribution model data to their own econometric model.
According to Google, ~33% of sales could be attributed to someone clicking a google ad. Woop!
According to magic numbers, only ~9% was rillly incremental. Oopsie.
The rest was people who were about to buy anyway, but The Goog was taking credit for them.
Some lessons:
Be wary of what a platform claims it delivers: Mandy Rice-Davis Applies
Use a good econometrics model to get a better picture of incrementality & attribution.
Look up Mandy Rice-Davis.