Gabor graduated from UC Berkeley with a BS (with honors) in EECS. He then went on to pursue a Ph.D. at Stanford. During that time, he was the NLP Architect at Baarzo (acq by GOOG, 2014). He is also a core contributor to the popular Stanford CoreNLP toolkit. In 2016, he co-founded Eloquent Labs, a conversational AI company, with a fellow Stanford NLP researcher Keenon Werling. He Served as Eloquent Labs’ CTO until it was acquired by Square in 2019. He now leads the Conversations team at Square to bring cutting edge conversational AI to small businesses.
What did Eloquent labs bring to the marketplace, that wasn’t already prevalent? What is the unique selling point of Eloquent labs as compared to other B2B NLP startups?
One way to characterize our unique insight is that there are a bunch of ChatBots that either answer questions, like static question answering, or are otherwise integrated with a small set of APIs. From our experience, talking to customers and deploying our ChatBot, this was not how most query streams look. Take even something simple like a shipping company: tracking a package, everyone says, is the most common query that people have. But if you look through and figure out how a bot or human would solve all of these queries, it breaks down into 100 different smaller API endpoints or smaller things that you have to do. For example, questions such as “You’re stuck in customs, why do I have to pay duties?” “You delivered to the wrong address”, so on and so forth. They all show up in conversations that customer service categorizes as tracking the package.
Eloquent Labs’ big contribution was a way to quickly incorporate new intents into the ChatBot in a way that didn’t require manual effort to integrate with the associated APIs. The end result was a ChatBot that took less time to program for a new intent than it would have for an agent to perform the task themselves.
What was your motivation while building up Eloquent Labs? What was your drive that got you in the NLP space? What was pushing you forward?
What caused me to do a startup in the NLP space is straightforward. I did my PhD in NLP. That was the unique set of skills that I could offer to the world.
Why Eloquent labs and why ChatBots? I had just graduated from my PhD, and my co-founder had done research in the lab that I was in as well. What we were good at was building high performance, accurate NLP systems. We looked around in the market for a place where that would be an actual advantage, a place where the technology was hard enough that we have a competitive edge, but not so hard that it’s impossible. We created Eloquent out of that philosophy.
What made you transition from research to entrepreneurship? Did you have other entrepreneurial experiences before starting Eloquent labs, or was it the first time you really went into this space?
This was my first startup and first real experienced entrepreneurship. I worked as a fellow at XSeed capital, which was a wonderful experience and one that I’d recommend to anyone that has the time during their PhD. That gave me a bit of a sense of what the VC climate was like, what fundraising looks like, and how these people that have been involved in entrepreneurship and startups for decades look at the space and evaluate companies.
How did you assign roles to each co-founder? How did you distribute the work amongst yourselves?
We fought over who would get to be CTO, and I won. We’re both technical people. So in a sense, we’re both on the technical side. On the other hand, Keenon has much more of a talent for talking to people and communicating the vision for the company.
What is the most challenging thing you faced at Eloquent Labs?
There’s a bunch of little, medium, and large challenges that are very specific to us or businesses like ours, but I’ll answer broadly. The most useful answer I can give to someone thinking about starting a startup is the most challenging bit was operating under uncertainty. There’s a bunch of different types of uncertainty, but the one I’ll highlight is product uncertainty.
Everyone gives the advice that you should talk to a lot of people, hundreds of people. What they don’t tell you is that you can talk to as many people as you want, you’re still not going to get a clear picture of the world. You get little snippets of truth; you get little ideas of what might be, but it’s very hard to run even just a single interview in a way that people give their honest impression, and aggregating on top of it is even harder.
That leads to this perpetual challenge. In a startup, it’s never okay to sit still, because if you sit still, you’re just going to die. The default state, if you don’t do anything, you run out of money and collapse. So you have to go in some direction or another, and you just never know enough to be confident that that’s the right or the wrong decision.
What constitutes success for you, personally? What drives you in the startup sense?
Keenon has a lot of family friends that are in business and successful in business. He was asking for advice from some of them, and retelling the woes of Silicon Valley and all the weird perverse incentives of fundraising and hiring and so forth. He recounted advice he got from one of his family friends, ‘Look, businesses aren’t hard, you have one job, bring in more money than you spend.’’ That stuck with me throughout the remainder of the startup and even now, as very sensible criteria for a successful company. Success in the startup is you bring in more money than you spend.
There’s many other ways to have strange, perverse Silicon Valley success. One of them is getting acquired. You can go after users and go after mega growth. But these are all anomalies in a sense. The core truth remains that if you’re looking at what makes a successful long term company either now or sometime in the future, you should be bringing in more money than you spend.
What was the most valuable thing you learned from the Alchemist experience?
At a high level, the role that Alchemist played in our particular startup venture was to get us exposed to the business side of things. We had very little experience about what the components of running the actual sales and marketing and business development side of the company is. Alchemist actually focuses a fair amount, in both their classes and their mentorship, on precisely this. It was useful to hear a bunch of different perspectives from the meetings and presentations that they gave. It was especially useful to get one on one mentoring from the various Alchemist mentors that they paired us up with.
Do you have any advice for the next generation of Alchemist Accelerator founders?
Don’t start a startup. It’s very painful. Most people aren’t going to listen to that and they’re going to do it anyways. That’s good. That shows some amount of determination. If there’s doubt there, and I can dissuade you, then you shouldn’t be doing a startup. I got the same advice once about getting a PhD. They told me, don’t do a PhD, and tried to persuade me otherwise. The motivation is the same. If someone can be persuaded out of it, then it’s not going to go well. It’s a very painful experience and much more painful than its portrayed in the media and by VCs and in the general culture of Silicon Valley.
Do you have any plans of getting back into the startup space in the future? Or, would you like to continue developing your technology at Square?
No, I’m not likely to return to the startup scene.
That’s because of what you said; because it’s very painful? Or, is there some other reason?
Mostly that. There are more interesting places to do interesting work than at a startup. As a technologist, startups are — contrary to my initial impression — not the most impactful way to bring new technology into the world. It’s a wonderful way to bring existing technology to a larger group of people. But if the interest is to build something new, to build something creative, to start something from scratch: startups, by virtue have all of these extra pressures being put on them, are not actually a particularly effective way to do this.
If you had to do this entire startup journey once again, what would you do differently than you did the first time? What were the biggest mistakes you made while you were working on it?
A ton of mistakes were made. A few things I could have done differently. It’s difficult to do a startup that is both developing new technology and trying to bring in substantial revenue. We tried to both develop something that was, in a sense, new to the world: conversational AI. At the same time, we were trying to monetize it and get actual customers and fulfill this criteria of success, of bringing in more money than you spend. Doing both at the same time at a high level is very difficult and adds extra burden to the startup.
In practice, there are plenty of successful companies that develop new technology, and then wait to get absorbed into a big company to productize it. There are also plenty of successful companies that take the technology that’s new or underutilized or utilized in an adjacent field that can be applied to something else, productize it and become a self-sustaining company. Many of these actually go on to have research links or research and development engineering arms, that then develop new technology. However, to do both of these together was probably a high level strategic mistake in Eloquent.
About the Alchemist Accelerator
Alchemist is a venture-backed initiative focused on accelerating the development of seed-stage ventures that monetize from enterprises (not consumers). The accelerator’s primary screening criteria is on teams, with primacy placed on having distinctive technical co-founders. We give companies around $36K, and run them through a structured 6-month program heavily focused on sales, customer development, and fundraising. Our backers include many of the top corporate and VC funds in the Valley — including Khosla Ventures, DFJ, Cisco, and Salesforce, among others. CB Insights has rated Alchemist the top program based on median funding rates of its grads (YC was #2), and Alchemist is perennially in the top of various Accelerator rankings. The accelerator seeds around 75 enterprise-monetizing ventures / year. Learn more about applying today.