New ZealandMedia and entertainment2023 – ongoingNZME
Karma machine-learning article recommendation engine for the NZ Herald
NZME's recommendation engine lifted click-through on recommended articles by 21% and session time by 9%
Problem
NZME wanted to turn casual readers of the NZ Herald into paying subscribers and earn more ad impressions. Its earlier recommendation engine was a heavy workload that took about two hours to run on the old cluster, which left little time to iterate.
Approach
NZME's data team built a first recommendation engine with BigQuery ML, with help from Google's news initiative. It then designed a second version called Karma. The engine suggests articles based on readers with similar reading history, uses natural language processing to read the context of the current article and removes articles the reader has already seen. Results are cached and served to the article page in about 80 milliseconds.
Results
Takeaway
NZME tied a recommendation model directly to reader engagement, which feeds both ad impressions and subscriber conversion. The company is now testing time-of-day models and how quickly different kinds of content lose value.
Categories
- Industry
- Media and entertainment
- Business function
- Marketing and sales
- Innovation type
- Product and service innovation
- AI technology
- Recommendation and personalization
- Adoption
- Built in-house
- Evidence
- Vendor case study
Sources
Source summaries
Google Cloud's customer story describes how the New Zealand publisher NZME built the Karma recommendation engine on BigQuery ML and Google Analytics 360. It reports a 21% rise in recommendation click-through, a 9% rise in average session duration and over one billion recommendations delivered. The page carries no date; the Internet Archive first saved it on May 15, 2023.
Full details are in the original sources.
