Amazon 1P Data Drives 30M Installs, 900% Ad Spend Growth for Streaming Service

Context

A direct-to-consumer streaming service sought a data-driven digital advertising strategy to acquire new customers at scale while maintaining strong return on ad spend (ROAS). The goal was to identify the customer segments and audience profiles most likely to install and register for the service, and to deliver 30 million ad-attributed installs within three years. The partner platform provided an extensive first‑party data ecosystem across content viewing, device usage and cross‑service engagement that could be leveraged to compensate for the lack of historical campaign data from the streaming client.

Challenges

The streaming service faced three core challenges: limited and sparse historical install and conversion data to train reliable acquisition models; the need to demonstrate measurable, privacy‑safe incrementality to convince the client that the partner platform was the right acquisition partner; and an ambitious install target that required maximizing efficiency under budget constraints so that scale did not erode ROAS. Additionally, the team needed an attribution approach that could validate installs directly attributable to the partner platform across multiple ad formats and touchpoints without exposing sensitive user-level data.

Implementation

With no historical customer-level conversion dataset from the streaming service, the team treated the partner platform’s first‑party signals as proxy labels and engineered a proprietary feature set from content consumption and device engagement. Signals included streaming viewing behavior on the platform’s video service, third‑party app usage on connected TV devices (competing streaming apps), and cross‑entity engagement such as music listening patterns, voice assistant interactions, and smart speaker usage. Transfer learning techniques were applied to map these proxy behaviors to the streaming service’s likely audience; XGBoost was used to build high‑precision propensity models that scored platform customers on download and registration likelihood. Our Fractional Head of AI led the modeling work and established an iterative training pipeline so that models were retrained as real client registration data accumulated post‑launch, progressively improving segmentation accuracy over time. A unified attribution framework integrated the partner platform’s demand‑side advertising stack with the streaming service’s mobile measurement partner (MMP). Using a privacy‑preserving data clean room, ad exposure logs were matched to registration events to enable accurate attribution without compromising user privacy. A multi‑touch attribution methodology was applied to correctly weight each ad format’s contribution and surface the end‑to‑end path to install. Given early budget limits, the activation strategy prioritized precision over sheer reach: highest‑scoring audience segments received the majority of spend, minimizing wasted impressions. The team adopted a test‑and‑learn approach across the advertising stack—initially concentrating on lower‑funnel, high‑intent formats (sponsored product placements and search) to validate creative and audience hypotheses, then scaling winners programmatically via the DSP. Continuous ROAS monitoring and rapid budget reallocation ensured spend followed performance as campaign learning matured.

Results

The data‑driven approach delivered well beyond expectations. The streaming service’s target of 30 million ad‑attributed installs within three years was met by the end of Year 2, driven by high‑precision audience targeting and ongoing model refinement. Advertising investment with the partner platform scaled from $5 million in Year 1 to $50 million by Year 3, representing a 900% increase in ad spend and demonstrating both confidence in the channel and the cost‑efficient acquisition capability. Attribution fidelity from the unified framework gave the client clear visibility into the partner platform’s direct contribution to installs, and the proprietary audience segments derived from first‑party signals created a competitive targeting advantage no other partner could match. As a result of the success, the streaming service expanded the relationship and appointed the partner to support acquisition initiatives across the wider entertainment company.