How Mark Kelly used conversational AI to help win a Senate seat

Conversational artificial intelligence has rapidly smartened and scaled since chatbots first entered mainstream social media in 2016. The first few iterations of chatbots on Facebook Messenger were simple, enabling restaurant reservations, flower deliveries, and other structured calls to action. Now, roughly one U.S. presidential term later, conversational experiences are increasingly intuitive. The AI technologies behind them can manage additional individual complexity, contextualize language more readily, and better simulate human reality — even when talking about politics.

Amplify.ai, an enterprise-level conversational AI platform, has helped 2020 senatorial campaigns drive engagement with local constituents. It uses natural language processing (NLP) and machine learning to attach to existing social media pages, analyze public sentiment and intent, and field individual questions through humanlike interactions with AI chatbots.

Multiple Democratic candidate campaigns incorporated the platform into their existing digital strategy, though Amplify.ai CEO Mahi de Silva told VentureBeat that contractual obligations prevented him from sharing a full client list. Senators-elect Mark Kelly (D-AZ) and John Hickenlooper (D-CO), however, have publicly partnered with the company. Jaime Harrison (D-SC), associate chair of the Democratic National Committee and recent challenger to incumbent Senator Lindsey Graham (R-SC), has too.

Mark Kelly was not a typical candidate: Before running for office, he served as a United States Navy captain, completed missions as a NASA astronaut, and launched a political action committee with his wife, former Congresswoman Gabrielle Giffords. The election also proved atypical. After John McCain passed away in 2018, his senate seat was held by two different Arizonian Republican appointees — Jon Kyl, then Martha McSally — in a two-year period. Kelly challenged incumbent McSally for McCain’s remaining term. Kelly’s campaign also needed to account for factors such as COVID-19 restrictions and Arizona’s history as a swing state.

In a statement to VentureBeat, Justin Jenkins, the digital director for Kelly’s campaign, commented on his team’s adoption of conversational AI. “When the pandemic hit, the campaign quickly began exploring new and creative ways to replicate the in-person conversations that we traditionally had at the doors. We chose to test Amplify’s conversational AI because of its ability to scale and customize the user experience based on the user’s history with the campaign.”

Kelly’s campaign couldn’t risk spreading COVID-19 by visiting constituents. But, in some ways, chatbots’ accessibility and slight personalization parallel the door-to-door, in-person canvassing that campaigns relied on for voter education and engagement before the pandemic.

Older chatbots were primarily based on strict inputs and outputs. For example, if a user typed “what is the capital of Arizona” into a bot on Facebook Messenger four years ago, the bot might have replied “Phoenix.” The conversational AI used in the most recent election goes further, working to interpret the user’s intent, or the different phrases people may use to ask about one topic. It then seeks to assemble a helpful, friendly, relevant response — and mirror the back-and-forth exchange of a human conversation.

Like a house call, individual messages from a chatbot could more strongly connect users to a candidate’s platform and allow a campaign to recruit them as donors, volunteers, and voters. The Mark Kelly campaign reported that it engaged with over 180,000 voters via Facebook Messenger in the first month of its conversational AI program.

De Silva differentiates these chatbot conversations from those people have with virtual voice assistants like Siri or Alexa, in which the user receives precise information the product has amassed from a database. In an interview with VentureBeat, de Silva said AI messaging creates “a consumer- or citizen-to-brand organization conversation … so it’s highly dynamic, it’s not trying to put all the resources into one system.” These dynamics are amplified by machine learning that tracks user behavior.

For example, the platform might decipher a comment along the lines of “Mark Kelly rocks” on the campaign’s public Facebook page and autonomously reach out to the poster on Messenger. The chatbot would thank the user for engaging with Kelly’s campaign and lean into the positive sentiment by asking if they are open to talking. If the platform analyzes a less positive comment, it may express interest in understanding different points of view.

Amplify.ai also tabulates comments and reactions that flow to a campaign’s Facebook account. The platform then performs sentiment and intent analysis on each data point and visualizes it on a dashboard so team members can closely track the audience’s interactions as a campaign unfolds. “You could imagine that a smiley face is pretty easy to find, you know, to associate positive sentiment with,” de Silva said. “But if you get a comment, we actually have to process that in context of what the post was trying to achieve.”

In addition to the insights campaign workers gained through Amplify.ai’s analysis, conversational AI can engage with people at a speed and scale unmatched by human teams. If, for example, a campaign received over 100,000 written engagements in one month, that would translate to over 3,000 individual messages or comments per day, which would require at least six full-time volunteers or staff members to reply to an average of a message a minute. The right AI could, in theory, reduce and manage this task while engaging constituents and inspiring them to volunteer, vote, and donate. According to de Silva, Amplify.ai has created over 10 billion engagements with over 500 million consumers since its launch.

Conversational AI will likely see gains in intelligence, credibility, adoption, and deployment speed in upcoming years. Startups such as Hyro, Pypestream, and Orbita are also working to provide businesses with conversational AI solutions for customer engagement. Hyro’s clients include government agencies, and marketing head Aaron Bours told VentureBeat that “if the AI is smart, fast, and human enough, you might actually have a discussion with it over key issues [such as] tax programs or foreign policy.”