Ad Measurement for PMs: Funnel Metrics, Attribution and Incrementality Tests

You are a product manager working on advertising at Rakuten. Explain how you use data to make advertising decisions that actually pay off: which decisions data should drive, which data and metrics you rely on for different campaign goals, how you tell whether an ad caused a result rather than merely preceding it, and how what you learn changes the next decision. ```hint Start from the decision, not the dashboard Pick one concrete decision, such as where the next block of budget should go, and ask which number would change it. Be wary of metrics that are easy to count but cannot show whether the ad changed anyone's behavior. ``` ### Clarifying Questions - Are you making these decisions for advertisers deciding how to spend, or for the ad platform deciding on formats, placements and how many ads to show? - What is the campaign meant to achieve: awareness, consideration, or purchases and sign-ups? - Which channels are in scope: display and video on websites and apps, TV or streaming, or ads shown inside a shopping site? - What data is available: only delivery logs from the ad server, or also the advertiser's conversion and sales data? ### What a Strong Answer Covers - Tying each advertising decision to a business objective and choosing metrics that fit the funnel stage, from reach and frequency to conversions and brand lift. - The data sources involved and their limits, including delivery logs, conversion tracking, surveys and privacy constraints. - Distinguishing correlation-based attribution from causal measurement such as randomized holdouts, lift studies or media mix models. - A decision loop that turns results into concrete changes to budget, targeting, creative, format and frequency, with guardrails for the user experience. - Pitfalls such as optimizing for clicks, crediting ads for conversions that would have happened anyway, and over-reading small tests. ### Follow-up Questions - Last-click reports say retargeting is your best campaign. How would you check whether that is true? - A TV or streaming campaign produces no clicks. How would you judge whether it drove store visits or online sales? - How would you set a frequency cap, and what data would tell you it is too high?