Problem
A company that buys sign-ups through social, display, video and search ads needs to know which spend works and where to put the next dollar. Attribution follows individual ad touches to a sign-up, so it needs a persistent user identity across every touchpoint, and privacy rules now block many touches. Sign-up counts also mislead for a team product: revenue comes from teams and organizations, while ads acquire individual users with different roles and adoption.
Example from the article: Asana's paid acquisition and data science teams asked whether spend drives business outcomes, whether the return is worth it, and how to split the budget across channels.
Solution in brief
Estimate ad effects from aggregated time series with a Bayesian marketing mix model, calibrate it with geo experiments and ad platform lift tests, and cross-check it with attribution. Value sign-ups with a predicted user lifetime value.
Questions this case answers
- How to estimate the causal return of spend across many channels from aggregated time series when individual touchpoints cannot be tracked
- How to value an acquired unit whose revenue accrues to a group and arrives slowly, so that acquisition decisions can use an early prediction
- How to measure the effect of paid advertising on sign-ups when user-level tracking is incomplete?
- How to allocate an advertising budget across channels, audiences, keywords and regions to maximize return?
Tools and methods
- Marketing mix modeling (MMM) — estimates ad effectiveness from aggregated spend and sign-up time series, with no user-level tracking.
- Bayesian model with MCMC — shares knowledge across periods, regions and ad types.
- Adstock and saturation transformations — encode the lingering effect of ads and diminishing returns.
- Geo-based experiments and conversion lift tests — hold-outs that calibrate the model.
- Predicted user lifetime value — monetization value of a sign-up, from early signals.
Google or a one-time email code. No password.