The business of analytics is booming, as all manner of companies try to understand what they’re doing right and what they could do better. But analysis is only as good as the underlying data. For many companies, that data is limited to what appears on the screen, leaving out a whole world of face-to-face interactions.
“The best way for people to communicate isn’t over something like [instant messaging], it isn’t over email, but it’s talking to one another,” says Russell Levy, CTO and co-founder of Chorus.ai. “None of that information is captured—none of that information is analyzed in any way.”
Chorus.ai brings machine learning to bear on that missing data, helping its customers learn as much as possible about how they really do business.
Chorus provides a “conversation cloud” that records, transcribes, and analyzes sales and customer success calls. “We’re going to analyze who is talking when,” says Levy. “What are you talking about? How are you talking? What’s your intonation? Are you talking about the right things?”
That information is then mapped onto CRM data, enabling companies to make smarter decisions about how to communicate.
One Chorus customer noticed that whenever a sales representative mentioned a specific competitor, they were 10% more likely to convert the sale. But the company couldn’t understand why mentioning a competitor would drive up their own sales.
Using Chorus.ai’s analysis, they realized that all of those discussions had been in the context of the founding story of the business, since its CEO had been an engineer at the competitor being mentioned. It turned out that people loved hearing their origin story—a piece of actionable intelligence that was missed by ordinary observation.
“There’s a lot of information,” says Levy. “We try to pull as much metadata as we can so you can get the real insights that you need.”
That isn’t easy. Human conversation is a labyrinth of abstract rules and customs, and natural language processing was out of reach for many generations of AI. Conversational functions that we take for granted—identifying parts of speech, or giving meaning to words based on their context—can be extremely difficult for machines to master.
Chorus.ai has built all of their technology in-house. Their tools can recognize which speaker is talking, track what’s being displayed during screen sharing, and automatically figure out which topics are discussed.
Processing conversations for tech companies runs into an unusual problem: how should a program understand words that aren’t in any natural language, like newly-coined names of software products? At Chorus, the platform first takes a guess at how the idiosyncratic word is pronounced, translating human speech into text. Then, it turns that guess back into an audio file so the user can correct the program’s pronunciation. Gradually, Chorus.ai develops a unique language for each client.
It took several years for the Chorus team to develop these tools and figure out a market strategy. Along with CTO Russell Levy, the company was co-founded by chief scientist Micha Breakstone and CEO Roy Raanani.
Breakstone and Levy both hail from Israel, where Chorus has an office. In the four years since they founded the company, Levy says, the start-up scene in Tel Aviv has gone from hot to hotter, fueled by strong tech talent coming out of universities and the military.
Raanani, a former management consultant, conceived of Chorus after spending years working with some of the largest companies in the world.
“He realized that the natural way that people communicate with each other is through voice,” says Levy. “Everything we do, both in the business world and anything on computers, is all through text. That’s not natural—and we’re missing everything that’s natural about it.”
As natural language processing continues to improve, applications like Chorus.ai are opening the field of data analytics to the rich context of spoken interactions.
from AWS Startups Blog