VoiceOps founder Daria Evdokimova: “Our mission is to be moneyball for sales” Edit this post

VoiceOps founder Daria Evdokimova:  “Our mission is to be moneyball for sales”

VoiceOps (formerly Clover Intelligence) is an AI startup for voice enterprise optimization that just presented at Y Combinator’s Demo Day and raised seed funding from Accel, Founders Fund, Lowercase capital, Y Combinator, and AngelList, through Edelweiss’ syndicate. In this interview with VoiceOps co-founder and CEO Daria Evdokimova, we learn:

  • Daria thinks voice optimization could be a US$36B opportunity
  • Enterprise voice data is a treasure trove nobody is looking at
  • AI is less about magical robots, and more about building proprietary data sets

Julie Ruvolo: How was YC Demo Day for you?

Daria Evdokimova: It was actually fun, because we raised our venture round before YC. We had a lot less pressure than most companies that presented, so for us, it was mostly about lead generation and building credibility for recruiting.

When we got into YC, we were already wrapping up the round, but we decided to still do it, because the feedback from our other founder friends that went through YC has been only positive.

Another awesome thing that we got out of Demo Day is that we've been getting hundreds of really great candidates from AngelList applying to work with us.

VoiceOps, as I understand it, is an artificial intelligence platform for analyzing enterprise voice. Please explain.

There’s an intersection of AI, voice, and enterprise that’s impending—and it’s going to change how sales teams work. We’re seeing it in consumer; Google, Apple, and Amazon are collecting and mining voice data through their own products, because they see enormously unfulfilled potential value.

To address the lack of tools designed to take advantage of call data, our team has developed technology that can transcribe calls, parse transcripts at 99.5% accuracy, then deliver insights on an individual’s core skills, like describing benefits vs. features, upselling attempts, and asking for the close.

Right now, the first tool that we built is for sales managers. It generates insights on phrases that maximize successful outcomes. For example: “Our reps don't close enough. Are they asking all of the right qualified questions at the right moments?”

With customers like Advent, Livestream, and Weebly, we’ve seen significant increases in close rates and growth in revenue, and we’re saving managers an average of 10-15 hours every week in call shadowing and coaching. We’re now preparing to come out of beta.

The longer term vision is to own all of enterprise voice data. Voice technology is finally getting to the point where it's possible to do analytics at large scale.

This is the kind of product that seems to make sense if you're talking about a certain number of salespeople, or a certain repetition to the kind of sales. Is there a target size or type of sales team for VoiceOps?

The bigger the team the better, to be honest. Right now, the average team size is around 50 to 150. There's less of a pain point when you’re talking about a five person sales team, because the manager probably sits right next to the reps and already knows what they're saying.

In terms of type of sales team, we've actually found that inside sales isn't the only the type of sales where the product works really well. Some of our clients are field sales teams, but most of the conversations still happen over phone, and our product is working pretty well for them.

Another interesting niche that we kind of stumbled upon is “customer success.” A lot of companies have success teams that actually carry a quota these days. They're basically sales teams that are there to service customers, but also try to upsell and cross sell when that's appropriate.

Even with support teams that don't carry a quota, they still have their own success metrics that we use and help people optimize. For pure support, it is customer satisfaction, so there are still metrics to work with there.

You mentioned in your interview with Huffington Post that 70% of sales conversations happen over the phone, versus about 10-15% which happen asynchronously over email. There’s already a huge market for optimizing email for the business space. How big is the opportunity for voice data?

The way we're thinking about this is less in terms of the addressable market, and more about the potential revenue that we can realize at our current price point. The way we pitched it when we were fundraising is that in the U.S. alone, there are two million salespeople. At our even current price point, with our current product, that's already $3.6B in potential annual revenue.

If you take into account service and customer success reps, that's 5 million people, or about $9B in potential annual revenue. If you take into account an even larger pool of all outbound professionals, where the core of the business happens over the phone—be it marketing, partnerships, fundraising, recruiting, everyone who's externally facing in the company—that would be 20 million people or so in the U.S. alone, or about $36B.

Then you wind up with just a ton of data, right?

Exactly. That's the idea.

So is VoiceOps a data play or an AI play?

We don't really want to self identify as an AI company, because there's a lot out there that over-promises to be this magical artificial intelligence robot that does everything.

We don't really want to self identify as an AI company, because there's a lot out there that over-promises to be this magical artificial intelligence robot that does everything.

We want to think about it in terms of the value that we're providing to users. Machine learning happens to be a part of it, but it's not the biggest value proposition.

The AI landscape at large is a lot less about actual algorithms at this point. It's a lot more about the proprietary data set that you're building that nobody else has access to. It's very specific to your use case. That's our thesis: Salespeople make a ton of phone calls every day, and so do support and service people, and that's just a treasure trove of data that nobody's quite looking at.

To give you a comparison, the Google Voice engine—their transcription engine—was built with just voicemail data. How interesting is a typical voice message? It's usually like 30 seconds or a minute, and very basic language. If you want to build a data set that is larger in terms of the pool of potential words and the language that people are using, sales and business conversations at large are a lot more interesting, because there are a lot more things that people talk about, and the conversations are just a lot longer. That's why we think that this is the treasure trove of data that nobody's looking at.

You're talking about voice data specifically, but I imagine that you will have access to other dimensions of data as well, like time of day, who you're calling…. Is that part of the plan? Because that gets quite interesting.

Absolutely. In sales, historically, there are a lot of opinions and a lot of things that people think work, but there hasn't really been a lot data around it. Our mission is to be basically moneyball for sales. It's to actually have data on what actually works and how to optimize that process.

How did you get interested in this fascinating area of enterprise voice data?

My background is actually not in sales. I studied computer science at Harvard and dropped out about three years ago. Initially I decided to join Google and did some work with machine learning, specifically with voice data, so it's interesting how it all came together later.

Then I went over to Gusto, which is a payroll and benefit provider and built tools for their support team and their sales team a little bit and then ended up at this startup called Coinbase, working on their data engineering.

One day when I was at Coinbase, my friend from Harvard, Ethan Barhydt, called me up and told me he was at this company called General Assembly, and was building tools for their sales team. Their sales team had doubled, and they were looking for a tool that would analyze their conversations and essentially tell them how to improve. Ethan asked if I wanted to try and take a stab at it and just build a prototype.

He moved out from New York, and both of us quit our jobs and just started working on that. He had built the first very rough prototype that was literally Excel spreadsheets with some voice analysis in terms of some keywords that people are saying, and some talking ratio and intonation, and some basic analysis on sales calls. It was kind of the moment when we realized that if two engineers who have never sold anything before in our lives can sell Excel spreadsheets for ten of thousands of dollars to real customers, we really are solving a pain point for them.

That's how we started it. We're both very, very, scrappy. Ethan lived out of a kitchen for a few months when he first moved out here. I lived in a van for a while. I think we still, to be honest, keep that in mind as we're growing a company. We're still trying to keep scrappiness a part of our culture.

That's how we started it. We're both very, very, scrappy. Ethan lived out of a kitchen for a few months when he first moved out here. I lived in a van for a while.

Nate Becker, our third co-founder, joined us a couple of months later as he was shutting down his previous company, which was also a machine learning startup. He was previously a data scientist at LinkedIn.

You just closed your seed round. How did you get in touch with your various investors, and how did your the round come together?

I'm actually good friends with Elaine Wherry and Todd Masonis at Edelweiss from when I first moved out to San Francisco. Elaine and Todd are also successful tech founders (they founded Plaxo and Meebo), and now they run Dandelion Chocolate.

When we started fundraising, they introduced me to their partner, Lee Jacobs. Lee has actually been crucial to the whole process. The round definitely would not have come together without his help, and I'm not just saying that. From the moment when we started to when we finished fundraising, he's put in a lot of work in terms of helping us with the pitch and introducing us to the right people. He's been super helpful.

Matt Mazzeo at Lowercase was an introduction from a friend. Accel was also an introduction from a friend, Ruchi Sanghvi, who runs South Park Commons, a co-working space that we used to work out of. And Lee introduced us to Cyan Banister at Founders Fund. This was one of her first investments after joining Founders last year.

Our partner at Accel, Steve Loughlin, also had just joined Accel. He used to run Salesforce Einstein, which is Salesforce's machine learning, artificial intelligence arm, after selling his company, RelateIQ, to Salesforce in 2014. We're his first and only investment so far.

How did Elaine and Todd’s AngelList allocation come into the conversation?

I was familiar with AngelList, but didn’t really know how startups raise money through private AngelList syndicates. I wanted to find out who was able to see my information and what exactly was shared. Lee was transparent about it: “Here's how much we're coming in with directly, and here's how much we probably will be able to put in from a private SPV from AngelList.” It was pretty straightforward.

VoiceOps is in the sales optimization space, but you're also figuring out how you're actually going to sell your product. It's a kind of meta situation. What is your sales strategy?

The first thing is we are dogfooding our product—a lot. All of our sales conversations are recorded and analyzed. We've gotten a lot of value that way. In terms of go-to-market strategy, it is very much looking at our initial customer list and finding more companies like that.

We aren't doing a lot of paid ads; so far it's been organic in terms of referrals. We've also started to go directly after the companies that we think are a great fit, doing the whole cold call or cold email process, showing them the demo, setting them up with a free trial, and then proving out the value that way.

Where is this headed for you?

VoiceOps is on a mission to own all of enterprise voice data. We started with sales and already expanded into customer success and support. In the future we'll expand into other enterprise verticals where calls are core to the workflow— recruiting, fundraising, political campaigning, banking...

Check out VoiceOps on ProductHunt.