DoorDash built a brand-dish affinity metric to help CPG advertisers identify which consumer segments to target when running ads on new verticals like grocery and convenience stores. The metric uses a log-odds ratio comparing conversion rates inside vs. outside a dish segment, grounded in logistic regression. This approach sits between naive raw CVR ranking and complex ML models — interpretable enough for advertisers to reason about, yet statistically principled with confidence intervals and uncertainty quantification. The system surfaces affinity scores in a self-serve report, enabling advertisers to either concentrate spend on high-affinity segments or craft tailored campaigns for low-affinity segments with growth potential. Future extensions include adjusted affinity controlling for user activity levels and automated segment selection via regularized regression.
