Geo lift tests

Causal experiments calibrating the model's channel priors

Experiments

5

4 read out · 1 running

Significant lifts

3/4

p < 0.05

Test spend

$1.68M

across all markets

Measured incremental

$1.86M

1.11x blended iROAS

CTV national holdout — market map

Treatment markets got the media change; control markets ran business-as-usual. Lift is the revenue gap between them.

Treatment marketsControl markets
Phoenix — controlTampa — controlColumbus — controlSalt Lake City — controlAustin — controlCharlotte — controlIndianapolis — controlMinneapolis — controlSt. Louis — controlSacramento — controlChicago — controlDallas — controlSeattle — controlBoston — controlNew York — controlLos Angeles — controlHouston — controlPhiladelphia — controlSan Antonio — controlSan Diego — controlSan Francisco — controlJacksonville — controlDetroit — controlMemphis — controlBaltimore — controlMilwaukee — controlAlbuquerque — controlAtlanta — controlMiami — controlOklahoma City — controlLas Vegas — controlLouisville — controlNew Orleans — controlPittsburgh — controlCincinnati — controlCleveland — controlOmaha — controlTucson — controlRichmond — controlBuffalo — controlBoise — controlDes Moines — controlDenver — treatmentDenverPortland — treatmentPortlandNashville — treatmentNashvilleKansas City — treatmentKansas CityRaleigh — treatmentRaleigh

Measured lift

7.4%

CI 3.1% – 11.6%

Incremental ROAS

1.40x

Connected TV experiment

p-value

0.014

significant

Markets

5 vs 42

test vs control

Readout

Test markets: Denver, Portland, Nashville, Kansas City, Raleigh.

CTV drove a measurable, delayed lift. The MMM coefficient (0.94 iROAS) understated CTV by ~33%; the model has been re-primed with this test as a Bayesian prior.