The Washington Post unveils Bandito, a custom-built, real-time content testing tool that allows Post editors to experiment with how a story is presented on The Post’s site. Bandito lets editors create multiple experiences for a story, varying the headline, blurb, and thumbnail photo for articles that appear on The Post’s site. Once a test goes live, Bandito detects which version readers prefer and automatically serves that version more frequently.
“Compelling presentation is really important in the digital world, and often there is more than one way to write a headline or more than one image available,” said Eric Rich, editor of The Washington Post’s Universal News Desk. “Bandito gives us immediate feedback from readers and helps give important stories their best chance to be read widely. Early results are very promising with some stories showing more than double the response rates.”
“Bandito greatly expands our testing capabilities and gives us the flexibility to test more of our content, both on and off our site,” said Nikhil Muralidhar, data scientist at The Washington Post. “This is just our first iteration on Bandito. We are planning to expand this technology to video and make it available to advertisers to maximize engagement with BrandStudio content. We are also working to automatically serve the best experience based on personalization algorithms. For example, the experience that works best for our international readers may be very different than the one that works for our local audience. The same is true for readers coming to The Post via social media. Bandito would automatically detect and optimize the experience based on multiple personalization signals like social channel, device type, time of day, and more.”
Bandito’s framework is based off of the multi-arm bandit (MAB) paradigm and is part of The Post’s Arc custom publishing suite. In a post for The Washington Post’s Engineering Blog, Nikhil describes how Bandito works within the newsroom:
“…as soon as an editor, adds a variant to a site module and publishes the page, the bandit algorithm automatically registers a new test with a new variant and the default variant. The publish event also causes the changes on the page to be pushed to the live site. The bandit algorithm, after registering a new test, starts tracking user events related to the test, on the live page. Initially the bandit starts with exploring the default variant and new variant experiences. At any point in time, using user engagement data from the two experiences, the bandit determines the best variant to showcase. A combination of exploring all variants as well as exploiting the best variant, not only allows the bandit to maximize user engagement with the site module by serving the best possible experience to users, but also gives new and less explored variants a chance to be served.”
About Arc: Arc is a powerful suite of publishing tools custom built for the modern newsroom by engineers at The Washington Post. Each of the tools work seamlessly together and can function individually, based on a newsroom’s needs. Each tool in Arc has a team of dedicated engineers devoted to advancing the product and responding to changes in the digital news landscape as quickly as possible. Learn More.