A major milestone for Conversational AI: 10 billion engagements with over a ½ billion people

John McCrea
Nov 4, 2020 · 4 min read

At Amplify.ai, we know we’re not a household name (yet), and that we share the field of Conversational AI with many other companies. So today, we are proud to announce a major milestone that puts us in a really interesting leadership position, with our Conversational AI platform having reached over 10 billion engagements with 500 million people around the world. And in the hard fought-battle to reach this milestone, we have now amassed one of the largest and most valuable conversational datasets — one that is now growing exponentially.

The first 10 billion engagements are the hardest

Efforts to achieve viable conversational AI have long suffered a classic “chicken-and-egg” problem: Machine learning only works once a large enough dataset exists to train on. But if your Natural Language Processing (NLP) engine isn’t already above threshold, people won’t engage with it deeply enough for you to amass a large enough dataset to unlock week-over-week improvement via machine learning. Making matters worse, traditional NLP systems are not very good at understanding live conversations, having been trained on “book quality” blocks of prose that are quite unlike the language of humans chatting with a brand. Knowing that, we took a flexible go-to-market approach in terms of both geography and vertical markets, and over time, we built out our own “messaging-first” NLP system, able to handle short phrases, slang, spelling irregularities, emojis, and more. 🍾 👍 👏

We achieved product/market fit in India first, helping media companies engage with their readers with messaging via the web and popular messaging services like Messenger and WhatsApp. In hindsight it makes sense that we saw traction first in the world’s second largest “mobile first” market. In India, we could get to billions of engagements with English speakers, while also mastering conversational engagement in Hindi and other Indic languages.

As our proprietary NLP engine improved, our market traction accelerated in the U.S., Europe, India and other parts of Southeast Asia. We built out solutions and partnerships to apply Conversational AI to marketing, e-commerce, and customer support. And we expanded the number of B2C “conversational surfaces” we work with to include social, messaging services, SMS/MMS, web chat, and search-driven conversations via Google Business Messages, making us the clear leader in omni-channel messaging.

Proven across many verticals

Early traction with e-commerce customers helped us tune our system to perform across marketing, re-engagement, and customer support use cases, and made it much easier to deploy solutions to new customers across that vertical. We’ve since made significant inroads into financial services, automotive and consumer-packaged goods, demonstrating the depth and breadth of our Conversational AI system and our ability to quickly adapt to new data sets.

Driving voter engagement at superhuman scale

And then came the 2020 U.S. election. Inspired by exciting engagement results in the 2019 national election in India, we found strong traction in the U.S. political market. We were excited about the scale of the opportunity, and a bit unsure how Conversational AI might fare at voter engagement in these highly polarized times. For Conversational AI, political discourse is the “deep end of the pool.”

The big idea: candidates for major political contests generate a ton of engagement on Facebook, in some cases hundreds of thousands of comments a day, but (absent Amplify.ai) those comments flow by without creating any value, like “water under the bridge.” What if every comment could be the start not just of a conversation, but a long-term conversational relationship that could drive voter engagement at superhuman scale? Well, it worked so well, driving list building and grassroots donations in a fully automated way, 24/7, that we ended up working with five Democratic presidential campaigns and seven Democratic Senate campaigns (including Mark Kelly and John Hickenlooper!), along with many other Democratic and progressive campaigns, committees, and organizations. (More on this to come, once all the votes have been counted.)

Helping battle COVID-19

Another accelerant in our climb to 10 billion engagements across 500 million people was the COVID-19 pandemic, which turbo-charged e-commerce, put enormous pressure on customer care operations, and challenged governments to respond at massive scale. We were proud to be selected by the Indian Government to help with their response to COVID-19 through AI-driven engagement with millions of citizens in need of information or assistance. Through the MyGov portal we delivered omni-channel civic engagement at superhuman scale, on the web; working with Facebook for solutions for Facebook, Instagram on Messenger and WhatsApp; and with Google on solutions based on Search and Maps leveraging Google Business Messages.

Onward and upward

The journey to get here has been full of learning — both for our human team and in the form of machine learning. And as we work through this time of transition, we look forward to more exponential growth and ever faster learning as more organizations realize that the future belongs to conversational brands.

Amplify.ai

Amplify.ai