案例库 · 软件与 IT · 技术决策 · 2015-2017
这条还没译成中文，下面是英文原文。
Vungle's millisecond ad choice lifted revenue per impression 23%
A machine-learning ad-serving algorithm personalises a mobile ad in milliseconds; an A/B test showed +23% revenue per 1,000 impressions.
Vungle
那一手
Vungle, one of the largest mobile ad networks, shows video ads to hundreds of millions of people a month. When a user enters the network, the ad-serving decision must be made in milliseconds, too fast for heavy computation, and a purely rule-based choice leaves revenue on the table.
Researchers designed and implemented an ad-serving algorithm that incorporates machine learning to personalise ad selection in real time, while addressing user fatigue, budget restrictions, and campaign pacing. The model's predictions pick the best ad for this user right now.
In an A/B test against Vungle's legacy algorithm, the new system increased revenue per 1,000 impressions by 23%, worth about $1 million in additional monthly revenue across the network.
为什么管用
- Personalisation at the single-impression level beats campaign-level rules.
- The algorithm respected budgets and pacing, so it was deployable.
- A/B testing gave a clean, credible measurement of the gain.
- Millisecond decisions scale: 23% per impression, millions of times a day.
可以搬走什么
When a decision must be made in milliseconds, learn the model offline and make the tiny real-time choice well. Small per-impression gains compound into millions across a large network.
后来呢
The case was published in Interfaces (2017) and is used as an example of analytics in real-time digital advertising; Vungle continued as a major in-app ad platform after the work.
资料来源
- Vungle Inc. Improves Monetization Using Big Data Analytics
- Vungle Inc. Improves Monetization Using Big Data Analytics
发现哪里写错了？告诉我们。