30 million installs, delivered a year early.
Ad spend up 900%.
Propensity models built from Amazon’s first-party data — Prime Video, Fire TV, Alexa — found the customers most likely to install Disney+, and a unified attribution framework proved exactly which ones Amazon won.
hit a year early
ad spend with Amazon
up from $5M
three-year install target
Client
Disney+, a direct-to-consumer streaming service, running new-customer acquisition through Amazon’s advertising ecosystem.
Goal
Build a data-driven advertising strategy that finds the customer segments most likely to install or register for Disney+ — and prove Amazon’s direct impact on every one of those installs, without overspending or eroding ROAS.
Amazon had no Disney+ data to build on, and nothing yet to prove its worth as a partner.
Disney+ had little historical install or conversion data to train acquisition models on. Amazon had its own first-party data — Prime Video viewing behaviour, Fire TV usage of competing streaming apps like Netflix and Hulu, Amazon Music listening and Alexa/Echo engagement — but nothing yet tying any of it to Disney+ specifically.
The harder problem was proving it worked. Disney needed to see, without exposing user-level data, exactly which installs Amazon’s advertising had driven across Sponsored Ads, Search and DSP, before trusting Amazon as a serious acquisition partner. And the target was steep: 30 million installs within three years, without overspending or letting ROAS slip.
Three constraints were non-negotiable:
- Amazon had to prove its first-party data made it the right advertising partner for Disney+
- Attribution had to be tracked and validated across every ad touchpoint without exposing user-level data
- 30 million installs had to land within three years without overspending or hurting ROAS
Proxy signals stood in for the data Disney+ didn’t have yet.
With no Disney+ install history to train on, the team treated Amazon’s own first-party signals as proxies — Prime Video viewing behaviour, Fire TV usage of competing streaming apps, and cross-entity engagement through Amazon Music, Alexa and Echo. Transfer learning mapped these behaviours onto likely Disney+ audiences, and XGBoost built the propensity models that scored Amazon customers on download likelihood. As real registration data came in after launch, models were retrained iteratively, improving segmentation accuracy over time.
The data Amazon already had — what people watched, streamed and asked Alexa — turned out to predict Disney+ installs better than anything Disney could measure on its own.
A unified attribution framework connected Amazon’s DSP to Disney+’s mobile measurement partner, so every install could be traced to the ad that drove it. A data clean room matched ad exposure logs to registration events without exposing user-level data, and a multi-touch attribution model weighted each format’s contribution — giving Disney full visibility of Amazon’s direct impact on installs.
Bespoke audience segments built from Prime Video, Fire TV, Amazon Music and Alexa data gave Amazon a targeting capability no other advertising partner could match. With budget limited, spend concentrated on the highest-scoring segments rather than spreading for reach. Rollout followed a test-and-learn approach — starting with lower-funnel, high-intent formats like Sponsored Ads and Search, then scaling proven audiences through the DSP — with continuous ROAS monitoring driving rapid reallocation toward what was working.
A three-year target, hit in two.
Precision targeting and privacy-safe attribution let Disney see exactly what Amazon’s first-party data could do — proof that proprietary signals, applied carefully, can beat reach-based advertising even on an aggressive timeline.
* Case studies reflect work undertaken by our Heads of AI either during their tenure with Head of AI or in prior roles before they were part of the Head of AI network; they are provided for illustrative purposes only and are based on conversations with our Heads of AI.
More case studies.
Your biggest pain point.
Fixed in 14 days. 50% off.
This started with one conversation. Book a 30 minute brainstorm call — we’ll plan your first AI project together and issue your 50% discount code. No payment today.
*Case studies reflect work undertaken by our Heads of AI either during their tenure with Head of AI or in prior roles before they were part of the Head of AI network; they are provided for illustrative purposes only and are based on conversations with our Heads of AI.