LevelAI’s founder on moving fast, betting on AI, and finding product-market fit
by Anthony Ha
We’re kicking off a series of founder interviews focused on product-market fit — what those founders did right, what they did wrong, and what they’ve learned along the way.
It’s a topic the Eniac team is passionate about, having worked with more than 200 founders as they try to find PMF. And as we wrote last week, while we could talk all day about definitions and tips, we believe there’s nothing more valuable or educational than getting the stories from the founders themselves.
We’re starting the series with Ashish Nagar, founder and CEO of LevelAI, a company that provides AI-powered tools to customer experience teams. We invested in LevelAI years before the current AI hype, and Ashish’s history in the industry goes back even further — he previously served as a product manager on the conversational AI team at Alexa.
Here are a few of the main takeaways from our conversation:
Focus on problems that customers will pay to solve. “Our general framework was, ‘Is this a big enough problem? Is this a top three problem for this person? Number one?’ Because a lot of times people say, ‘Yeah, this is a problem.’ But is it something big enough that you will write a check right now, if you had a magic solution? That’s very different.”
Finding PMF is an ongoing process. “I think PMF is a little bit of a continuum in enterprise, because you can get tighter and tighter and tighter and tighter product-market fit or product-market alignment. But you need to have a kernel, some small product that works for that [ideal customer profile].”
Be clear about what kind of customer you’re pursuing. “We do AI for call analysis, but a lot of people tell us, ‘Why wouldn’t you do it for Gong and Chorus for sales teams?’ The simple answer is, that’s a whole different buyer, that’s a whole different market, we just don’t sell to the salesperson. When you’re building a sales product, the go-to-market journey is so different than it is for a VP of a contact center or the director of a contact center. Their value proposition is different, their budgeting is different, their problems are different, the other technologies they integrate are different.”
Don’t forget the meat and potatoes. “We ask ourselves these three things: Where is my core differentiator where nobody else is playing? The second is, where do I need to catch up with other products that people like? And the third one is, what are the table stakes, meat and potatoes features you need to play in the space? For an enterprise SaaS company, you need all three. If you don’t have the meat and potatoes, even if it’s a boring form workflow or something, nobody takes you seriously.”
And here’s our full conversation, edited for length and clarity.
Anthony Ha: How would you describe LevelAI in three sentences or less?
Ashish Nagar: LevelAI has built AI for customer service. Our software is a contact center intelligence platform. We use natural language understanding to get insights from customer service data, augment customer service agents’ productivity, and as a result, improve customer experience and also manpower productivity.
Anthony: If you rewind and we think about when you were first getting ready to start a company, what was that initial idea?
Ashish: The initial idea — I actually pitched [Eniac co-founder Hadley Harris] with it and he was like, “Really?” But Hadley and I, we’ve gone through a lot together.
The initial idea was a voice assistant and data input entry analytics tool for frontline workers. This was pre-pandemic. The thought was frontline workers who work with their hands, have their gloves on, have grease on their hands, it’s not good for them to access computing on desktop machines or an iPad. And AirPods were getting big at that time, and so was voice computing.
Before I started LevelAI, my past job was as Alexa, so my thought was to build an efficiency and augmentation tool, using voice AI for frontline workers. And that’s what we did, with retail as our first market. That’s what I pitched Eniac and Hadley with.
We built our MVP [minimum viable product] and after pitching it for a couple of months, we realized that a few people were giving us this feedback that hey, the technology would be great for customer service.
That’s how we started. Some of that original stuff is still around, the naming and some of that code is still valid, but we’ve moved on to a completely different area. Yet the underlying idea remains the same: Orienting human human productivity with AI and analyzing this large enterprise data stream with AI.
Anthony: Let’s talk about your MVP. In consumer software, it’s pretty easy to whip something up — well, not easy, but you can do it relatively quickly. Given that AI was such a big component of what you were doing, was it a more intensive process to build the MVP?
Ashish: It is much more intensive. Our MVP involved building a product that was almost like Siri, but we were interfacing with AirPods. The other product we built was an Alexa-like speaker. I kid you not: We built a small speaker, literally a small LevelAI smart speaker which would be kept on the tables of banking institutions. Like if you go to a bank teller, they would have it.
So we really had to build a mobile app, hardware, software, everything. But we were very fast with all that. You can’t waste time at that stage, when you’re finding product-market fit. If I want to emphasize one thing, it would be the focus on fast experimentation.
Building was one thing, and then we had to show it to customers. If you have a consumer product, you can run an Instagram campaign, a Facebook campaign, send it to your partner, your parents, your friends. In enterprise, you need to find the ICP [ideal customer profile], the persona and all of that.
I’ve been advising a few seed-stage companies that are asking, “Do you first find a customer and then you start building with them as a design partner? Or do you first start building and then find a customer?” That’s a classic problem.
Anthony: So in your case, which one of those camps do you fall into?
Ashish: We first started building, because we always had a hypothesis of a problem we were looking to solve. (Also, the right customers don’t want to waste time with a company at the idea stage.)
But then we were okay with changing that hypothesis quickly. Because the thing is, the kind of customers whose feedback would mean something to you — unless they already know you, you have to make it worth their time. And to make it worth their time, you have to show them something working. “Hey, this will work, we’ll help you solve this problem, and we can expand from there.” But if you just show them PowerPoints and or a Figma prototype, it’s interesting, but you don’t get a lot of meetings with these people.
Anthony: After you had these meetings, you got pushed in the direction of call centers. Was that something that came up right away?
Ashish: I’ll say a couple of things. One is, in addition to looking at retail, we’ve looked at enabling wind farm workers. And then we looked at some health care workers, as well as finance. So those were three or four different things we evaluated, each of them is a different persona, a different buyer. We ran through that journey in about 3 months.
I think customer service was the fourth or fifth one. But I was always conscious of the need for speed in shifting between those personas and the mental framework for evaluating them.
Anthony: Talk me through how you evaluated each of them and moved quickly from one to the other.
I think it’s the same thing as B2B sales. You want a representative sample of your ICP — if I am wildly successful, who is going to buy this? So at hospitals, we looked at literally the CTO or CIO level at a large hospital in the Midwest. And also, who’s the end user? Who’s the check writer?
At least part of our general framework was, “Is this a big enough problem? Is this a top three problem for this person? Number one?” Because a lot of times people say, “Yeah, this is a problem.” But is it something big enough that you will write a check right now, if you had a magic solution? That’s very different.
So that was one thing. The second thing was: Is the market big enough? Which is an obvious thing.
And then thirdly, the big feedback from Hadley and all the investors was that retail is not just big enough and their margins do not support buying a lot of technology. Or rather, the industry is really big, but they don’t write big checks.
Anthony: How long did it take you to go from that idea of using AI to augment frontline workers to having a product for customer service?
Ashish: Nine months. The first six months was retail, healthcare, finance, then we started to build the customer service product.
Anthony: When you built that customer service product, did that feel like the moment when you found product market fit?
Ashish: It was still a journey. Our first product in the space was more like a real-time augmentation system, on the heels of what we were doing in those other markets. Hadley and our other VCs were saying, “Real time is the thing,” but we quickly realized that that’s not where the real money was. The real money was in other, slightly allied use cases. It was good to position the company as offering real-time augmentation, but the dollars were in aligned use cases around AI-driven analytics, and on real-life augmentation and quality improvement and things like that.
So I would say we launched our product in October 2020, and then mid-2021, we started seeing traction.
Anthony: You said that in those early conversions, the real test is: Is this a big problem that you’re willing to write me a check right now to fix? But once you actually have a product on the market, what are the things you’re looking at to determine if this is the right direction or if you need to continue to pivot?
Ashish: That’s a good question. Number one, what are the dollar signs you can extract from a customer? Number two, are there enough of those customers? And, number three, is it one of their top three pain points?
I am a big fan of the founder of Cohesity (though I have never met him), and his benchmark for product-market fit is when a median sale rep — not your best rep, not your worst rep — can sell it on their own, without any further involvement, for a $100,000 deal.
My bar would be a step lower: You can sell enough of it with some founder involvement, get $20,000 to $100,000 from somebody, and you can see that there are tens of thousands of such customers out there.
Anthony: Another framework that Hadley and other VCs have talked about is the distinction between the stage where you’re pushing a solution on the market, and you can sell it but it takes a lot of work, to the moment when there’s market pull, and suddenly you feel that people you’ve never done proactive outreach to have heard of it. At that point the challenges are less about selling and more about scaling because the demand is suddenly there.
Ashish: That is much later. That’s a much higher bar. And again, I’m just talking from our experience. But that moment where you hit a sales inflection point and you need to just give fuel to your go-to-market engine, I would be surprised if that’s less than a Series B or later, even for the best companies.
Anthony: As we talk, I’m realizing how I’ve conceptualized product-market fit as this on-off switch, night and day. Instead, it sounds like this multi-stage process where you get more and more PMF as you go along. Hopefully.
Ashish: I totally agree, especially in enterprise. If somebody’s writing you a $100,000 check, their finance team is involved, they have to have a team meeting about the whole thing, so it takes a lot of time. But when you get through that process, you can land in the organization with a use case, and then you can expand or pivot or change from there.
I think PMF is a little bit of a continuum in enterprise, because you can get tighter and tighter and tighter and tighter product-market fit or product-market alignment. But you need to have a kernel, some small product that works for that ICP.
The other thing I would say is, think about ICP and market definition, because we talked about the product part of it, but not as much about the market part of it. Let me give you an example. We do AI for call analysis, but a lot of people tell us, “Why wouldn’t you do it for Gong and Chorus for sales teams?” The simple answer is, that’s a whole different buyer, that’s a whole different market, we just don’t sell to the salesperson. When you’re building a sales product, the go-to-market journey is so different than it is for a VP of a contact center or the director of a contact center. Their value proposition is different, their budgeting is different, their problems are different, the other technologies they integrate are different.
So you can land with a product and a persona, then expand from there, or do microprobes from there. But if you start saying, “Oh, I will sell the software to customer service, to sales, to investment teams,” that is much, much harder.
Anthony: I want to rewind a little bit and talk about the space between when you realize there’s a need here and you can sell something, and then actually creating the best product in the market. Obviously, this is a very big topic, but can you talk about how you’ve continued iterating on the product?
Ashish: That’s such a great point, and a hard one. One, that’s where the founder-market fit comes in. One of the reasons we were able to figure out some of these things (and I hope we are working to figure out the others, because we haven’t figured out a lot) is because being an expert in the space (somewhat of an expert in the space), you know what problems to solve. We knew, let’s say, within AI, which problems to solve where there is novelty and customer value.
Like in anything else, you iterate through it, but we iterated through a small set of things on the technology side, which we thought were prioritized in our minds: “These are the four or five things we should try.” And then a couple of them became core architectural differentiators.
For example, we had a generative AI approach three years ago, and we had a large language model-based approach two and a half years ago, because we knew that this is where the world is going.
Maybe I’m giving us too much credit. Maybe we just knew that this seems to be the most logical way of doing it [back then], the state of the art logical way of doing it [back then]. And it just so happened that it’s also where the world went.
That is the AI part. The non-AI part is relentless. Speed is one of our virtues and values. We can drive people crazy with it, but we prioritize the fast launch to have multiple product experiences. We had, and I hope still have, a great sense of urgency.
And we ask ourselves these three things: Where is my core differentiator where nobody else is playing? The second is, where do I need to catch up with other products that people like? And the third one is, what are the table stakes, meat and potatoes features you need to play in the space? For an enterprise SaaS company, you need all three. If you don’t have the meat and potatoes, even if it’s a boring form workflow or something, nobody takes you seriously.
Anthony: What can you say about your current traction?
Ashish: We don’t share too many numbers, but we have aggressive growth goals. This year we want to 5x our team, we’ve been hiring consistently through Covid. And in terms of user traction, we probably are 5x to 10x where we were last year, albeit from a small number of users to start with.
Anthony: Lastly, I imagine there are a lot of founders who are excited to get into the generative AI space right now. Do you think they can still find fresh opportunities?
Ashish: For sure. It’s a great time.
This is a bit of a tangent, but back in 2018, 2019, I live in the Valley and I used to go to AI meetups, which were full of good people. That lasted for nine months, and then crypto came along. Then the same group of about 30, 40 founders, 80% of them were like, “Oh, now we’re interested in crypto.” And I was like, “Well, I’m still interested in AI. Do I need to find new friends? What do I do?” So there was a year and a half of crypto capturing all the attention, but now AI is back.
My point is, find your area of interest and go deep in it. You will find all kinds of opportunities, whether it’s AI or it’s crypto, or if it’s HR Tech, I don’t know — my advice would be, if you are deeply interested in something, first of all, don’t give up on it six months from now if people stop talking about it. Number two, look at it deeply.
