Following its own success, Spain’s Politibot moves to help other outlets build chatbots

by Shan Wang
Oct 30, 2018 in Miscellaneous

Chatbots need vacations, too.

First launched last June during Spain’s general election as a Telegram bot experiment, Politibot tried to offer Spanish audiences a more whimsical and comfortable way to keep up with a tense political cycle and dissect facts and figures politicians and their parties spouted. (Quartz’s news app and Purple’s messaging app were among the inspirations.) Promoted only through its founders’ networks, the spunky Telegram bot built up a reliably loyal following of 8,400 users.

Then a couple of the Politibot project’s founders, María Ramírez and Eduardo Suárez (who launched Politibot after they parted ways with the Spanish digital news startup El Español, where they were also cofounders) left for Univision to cover the U.S. election, and the news and information side of Politibot “went on holiday” last summer to the Spanish seaside resort city Torrevieja.

In Politibot’s case, it’s been more of a working vacation, as the team behind it began building out a full-fledged bot-making platform based on what it had learned from its Spanish election summer experiment, with the help of a EUR50,000 Google DNI grant from November 2016. The completed “bot management system” that other media organizations can use to build custom chatbots includes analytics more useful for newsroom purposes than either Telegram (through Yandex) or Facebook Messenger offer now, and more personalization on the messaging journey of individual users. The makeup of the eight-person team suggests the founders’ ambitions: On board are journalists, a designer, engineers who’ve worked on natural language processing, and political and data scientists. (Politibot chief of technology Miguel Gil Biraud detailed these new platform developments to me during a pitch event at the Global Editors Network conference last month in Vienna.)

“Someone describing Politibot once said it was almost Socratic; he had the impression he understood things better this way,” Suárez told me. “We have the platform and we have the expertise. Problems news organizations will run into, we’ve run into before. We can help them with that, and we can hopefully help them get the same retention and engagement we did.”

About two thirds of the Google DNI money went into platform development, and a third to producing original content for Politibot. After an overhaul of the backend bot platform, the team resuscitated Politibot for Telegram and added a Facebook Messenger version for a limited run in the spring. It began testing expanded editorial content again, including news digests, analyses of issues outside of Spain, newsy charts for Instagram, and an original politics podcast. It had to give up its audience from last year after it closed down the original bot, but the bot’s second season attracted more than 6,200 users across the two different platforms, about 3,400 of whom opened the bot once each week and 2,400 of whom opened it every day — a prime user base to target for any potential paid subscription or membership program, Suárez said.

Just this month is started to raise funds through a Patreon drive to build out the editorial side of the company. Suárez said the team would love to hit US$2,000 in monthly reader support to add a journalist and help pay for its servers, and is working on exclusive content for paying supporters (it’s about a fifth of the way to that goal).

The company’s other planned source of revenue will be the clients who pay to use its bot-making platform, and for installation, troubleshooting, and advice. It’s also considered advertising (a narrative, serialized ad-bot, Suárez suggested), but hasn’t built anything. At the moment, it’s in talks with a couple of publishers in Spain, and according to Suárez and Biraud is looking at tiered monthly payment levels tailored to the organization, based in part on the number of messages the customized bot sends and the complexity of the bot required by the news organization.

“We’re more of a consultancy at the moment, and having conversations with newsrooms about installing the platform. The first part of the conversation is always, what do you want a chatbot for?” Suárez said. “If you just want a chatbot that is an RSS [feed], that gives you content from your news organization tailored to a user according to the user’s tastes, and pushes notifications around that, sure, that’s one option. Our experience says it’s probably more useful for an interested news organization to build this kind of interactive conversation through a chatbot, a conversation that really gives the reader a reason to come back.”

Suárez and Biraud both acknowledged that alternatives like Chatfuel or the free Wit.ai exist, but said that the various features in Politibot’s bot-making platform were honed with news organizations’ needs in mind. For its Spanish election experiment, for instance, Politibot profiled users on their age and gender, and then showed them how their specific demographic was voting, based on polls. In another chat interaction, it asked users for their location, and then returned a chart showing the results of their constituency. And if you wanted to offer different extra content to a segment of paying subscribers, the Politibot platform can help with that, too.

“You need different analytics for chatbots, since pageviews don’t make sense. It took us a while to find metrics that wouldn’t be easy to game,” Biraud said. “Now we have a way of showing a heatmap of the conversation tree, showing where an organization’s bot lost users, what branches of the conversation individual users followed, what individual users are interested in. We might have one conversation option on Brexit and another on French politics, and if the user never takes the Brexit branches, we need to think, what else can we show them?”

Some larger news organizations like The New York Times or the BBC will inevitably go in-house anyway when it comes to building bot experiences, Biraud said, but “smaller media groups are barely scraping by. They don’t have the resources. We can solve some problems for them.”

This article originally appeared on Nieman Lab and is republished on IJNet with permission.

Main image CC-licensed by Flickr via andreavallejos