[Infowarrior] - WaPo ‘Bandito’ Tool Optimizes Content For Clicks

[Richard Forno]

Washington Post’s ‘Bandito’ Tool Optimizes Content For Clicks

System promotes version of a story with most popular headline, images
Jack Marshall
Feb. 8, 2016 2:34 p.m. ET

http://www.wsj.com/articles/washington-posts-bandit-tool-optimizes-content-for-clicks-1454960088

The Washington Post is experimenting with technology to automatically optimize articles on its website for maximum readership.

A new internally-developed tool, dubbed “Bandito,” allows editors to enter different article versions with varying headlines, images and teaser text into its content management system. The technology then detects which version readers are clicking or tapping on more, and automatically serves that version more frequently on the homepage and other areas of the Post’s site.

Publishers frequently use so-called A/B tests to compare different versions of articles and to establish which headlines and images appeal to readers, but the Post’s tool is particularly interesting because it automatically implements changes based on the information it collects. This allows editors to essentially “set and forget” the tool, the company said, which makes the process more efficient.

“We have terrifically reported stories and I just want to best explain to readers why they should be interested,” said Eric Rich, editor of The Washington Post’s Universal News Desk.

Bandito was built in-house at the Post and is evidence of the influence its owner, Jeff Bezos, is having on the company. In the more than two years since Mr. Bezos bought the Post from the Graham family for $250 million, the billionaire Amazon.com Inc. founder has made his mark by championing technological experimentation, and a focus on customer experience and data-driven decision making.

According to Mr. Rich, the tool is now driving editorial staffers to present articles in ways they may not have considered previously. For example, it’s helped them understand that images with human faces often result in more clicks than those without, and that headlines beginning with the word “how” often drive greater readership.

The first iteration of the tool optimizes content based purely on the number of clicks articles receive, but the company plans to soon factor in other metrics also, including how much time readers spend with articles after they click.

“This isn’t just about clicks. Clickbait works for a while, but not long-term,” said Sam Han, engineering director of data science at the Washington Post.

“Engagement is a better metric to focus on, but we’re not going to optimize just to that. We’ll find a good balance.”

According to Mr. Han, the company plans to develop the tool to help power other features of the site, such as its video players, and to further optimize content for different types of users. For example, it plans to optimize for international audiences versus domestic readers, and for desktop users versus mobile users.

Bandito may even begin to factor in other more detailed information about Post readers into its decision-making, such as whether or not a user is an existing subscriber, and types of content they’ve engaged with in the past.

“We will be more careful in that area because of privacy, and because we need to be sure we’re providing right-leaning and left-leaning content equally,” Mr. Han said, in reference to articles of a political nature.

Beyond regular editorial content, the Post said it also plans to use the tool to help boost the readership of sponsored posts created by its BrandStudio ad division.

Instead of working with vendors to power features on its site, Mr. Rich said the company sees advantages to building its own tools.

“We’re building everything in-house now. We control what we’re making; we can make it to specifications we understand and can iterate and evolve,” he said. “I don’t have to call a vendor and rely on customer support. I have the engineer sitting 40 feet away from me.”

Write to Jack Marshall at Jack.Marshall at wsj.com

--
It's better to burn out than fade away.