Netflix business model is paid subscriptions, which means their marketing goal is to get sign-ups. How do they know they need to spend more ad dollars in the US vs. other counties? in Facebook or Google?
They’ve built a budget optimization system, and its architecture looks something like this:
(from Engineering to Improve Marketing Effectiveness (Part 3) — Scaling Paid Media campaigns)
They don’t explain in details how it works, but here is my guess:
1) Collect time-series performance data broke down by channel and country: for a given time window (30 days, three months, one year) they gather data about how each channel performed in each country (sign-ups, revenue, probably at the campaign level.)
2) Use a regression model to predict performance (machine learning): they use the amount spend, channel and country as variables to predict sign-ups and revenue (if they optimize at campaign level.)
3) Compare results and based on the confidence of the predictions adjust budgets per channel: depending on what were the predictions per channel and country, they automatically send write calls to each Publisher Ads API to modify the budgets and adjust the targeting specs (i.e. add/remove counties.)
The tech complexity is not in the automation but the potential aggregation bias: the system is treating each ad creative inside each campaign as they were equal; there is no learning process if you don’t know which message (original, video, story) is impacting the results the most.
