Issue 283,  published November 17, 2020

Another Local-National Bundle to watch

Here’s a technical experiment that caught my attention recently. For about a week and a half starting on Election Day, the Washington Post ran automated AI-powered audio updates reporting local and national election results at the top of episodes from three of its podcasts: Post Reports, Can He Do That?, and The Daily 202’s Big Idea.

Think of these updates as a kind of audio ticker tape that came bundled with episodes of those podcasts as they were downloaded during the immediate post-election period. As a listener, you essentially experience them as a variation on the pre-roll spot, in which a mechanical — though female — voice reads out the latest real-time results from both the presidential race as well as races that were local to your area. The spots were short and punchy, typically clocking in at around a minute or so.

This dynamically-inserted election update project was the latest effort from the Post’s Strategic Initiatives team, which focuses on using emerging technologies to explore experimental storytelling approaches in order to reach more people. Past projects have included efforts around virtual reality, augmented reality, and artificially intelligent automation.

Audio was always an area they were interested in, as Elite Truong, the recently-appointed director of strategic initiatives who oversaw the project, told me last week. “The main question for us was: how do we provide more of a service through audio — whether it’s through podcasts or audio embeds on web pages?” she said. “How do we make that more accessible?”

The project was primarily powered by an in-house product called Heliograf which Truong describes as a cluster of machine-learning algorithms built to process high volumes of data from various APIs and convert them into information-rich scripts that can be edited down into publishable journalism. When the tech was used to cover DC-area high school football games every week back in 2017, Jeremy Gilbert, Truong’s predecessor, said that the technology was “creating a new model for hyperlocal coverage.”

This isn’t the first presidential cycle in which Heliograf was used to automate the Post’s local race coverage. In 2016, the Post tested it out to track results for 500 races around the country, and in this cycle, that number of covered races was greatly expanded. The idea to try delivering those race results through the Post’s existing podcast products, Truong tells me, came out of an impulse to serve those results to audiences where they already are.

The actual deployment of those audio updates depended on a mix of new and old technology. The new being the combination of Heliograf-generated results with text-to-speech automation tools, and the old being the manipulation of dynamic ad insertion solutions to deliver those generated audio update reads to listeners based on their geographic location. There’s a touch of the familiar to this particular composition of tools and ideas: between the bundling of local content with fundamentally national products, the general thinking of meeting audiences where they are, and the use of ad tech as means to those ends, this Washington Post experiment sounds like NPR’s recent announcement that it will start shipping episodes of its daily afternoon podcasts, Consider This, with dynamically-inserted local news segments from select cities.

Two doesn’t necessarily make a trend, but it’s a start, and I really do hope to see more podcast teams dip their toes into this whole local-national bundle thing. There’s a ton of service (and entrepreneurial!) potential in this lane. I’d also be curious to see what other audio experiments that the Post will come up with once the dust settles around this election project.

But that might have to wait. The automated audio update only ended last Friday, and Truong’s team has yet to fully debrief on the findings. That said, she notes that, at this point, there seemed to be very few reports of failed local results deliveries, which is a very encouraging data point. It might be enough to greatly expand the program.

“I want to put this everywhere,” she said.

 

I run this thing.