Unpacking AI Startups: Metrics, Playbooks, and the Future

Julia Klein is a partner at March Capital, a growth-stage VC firm. Prior to this, she was the cofounder and CEO of CareerPeer. She has an MBA from Harvard.

Julia's favorite books:
- Brandon Sanderson's books
- The Nightingale (Author: Kristin Hannah)
- Crime and Punishment (Author: Fyodor Dostoevsky)
- The Will of the Many (Author: James Islington)

(00:00) Introduction
(00:07) What Sets an AI Startup Apart
(05:14) The Rise of Generative AI
(08:00) AI and Digital Biology
(08:38) Opportunities in Computer Vision
(12:02) The Potential of Synthetic Data
(13:55) Metrics and Challenges for AI Startups
(19:27) Important Metrics for Growth Stage AI Startups
(25:12) Exciting Technological Breakthroughs in AI
(27:41) Future Opportunities for AI Founders
(29:28) Rapid Fire Round

--------
Where to find Prateek Joshi:

Newsletter: https://prateekjoshi.substack.com 
Website: https://prateekj.com 
LinkedIn: https://www.linkedin.com/in/prateek-joshi-91047b19 
Twitter: https://twitter.com/prateekvjoshi 

Transcript

Prateek Joshi (00:01.796)
Julia, thank you so much for joining me today.

Julia Klein (00:04.59)
Absolutely happy to be here.

Prateek Joshi (00:07.004)
Now you invest in AI among other things. When you look at a company, what sets an AI startup apart in your mind? Like what makes you go, this is a company I want to invest in.

Julia Klein (01:01.71)
Yeah, so we at March Capital, we've been investing in AI for many years. You know, obviously, recently, it's all the generative AI hype that's been, you know, the past kind of two years now. But really, we've been thinking about how enterprises use AI for the past decade. And so I think, you know, it's really changed from thinking about it in the sense of.

you know, at first, similar to how, you know, the 2010s were characterized by everyone was cloud native. And that was sort of the, the exciting technology right now. It feels like everyone's claiming that they're AI native, but at the end of the day, right, it's a, it's a technology and it's really, we're looking at how is somebody utilizing that technology to better solve a problem for their customers. So while we definitely want to dig into.

you know, what is their advantage? Really that competitive advantage is framed from how we think about first principles investing, right? So that, is it a good company? Is it a good deal for us? Are they addressing something that's mission critical for us? It's mission critical within the enterprise because we're enterprise investors. And then, you know, are they addressing that mission critical problem in a way that makes sense and is easy to use for customers?

And so it's a lot of really how we would frame any investment when we're thinking specifically about AI investments and then we're layering on, right? You know, they need a sustainable competitive advantage. And so how do they achieve that? And in a lot of cases, they're achieving that through utilizing AI in a unique way.

Prateek Joshi (02:45.916)
And when you think about the AI tech stack, right? So, so many parts of the stack being disrupted and there's been debate about what parts will be ruled by the incumbents versus what parts are more likely to be disrupted by startups. So what's your view on the stack itself and also what parts are open to disruption?

Julia Klein (03:11.694)
Yeah, so this is an oft debated question. I think it has been on the minds of everyone who's investing in the space. And so I think part of where we have been looking is and where I think startups will really have an advantage is when either they're addressing net new capabilities. So something that AI has made possible, right, that historically wasn't possible. And so I think that can be,

that can manifest in areas where there's vertical applications of AI where historically maybe things you just weren't capable of doing things. And an example of that is within something like the insurance space. It's really AI and generative AI has enabled software to tackle that space in a way that historically it couldn't. And so that happens, I think, across different verticals.

And sometimes horizontal opportunities as well. So anytime there's like that net new capability that's introduced, I think that there's always an opportunity for somebody to come in at a startup level and address it. I think where incumbents have an advantage is where they can lean on their existing distribution channels. So anytime it is a matter of how quickly can you get distribution, and I think a lot of cases, right, with his.

historically with horizontal applications, that has been the case, you're going to have incumbents that have that advantage, right? So if Microsoft can bundle something, right, and offer it to you free, it's really difficult to beat that, even if your product is superior as a startup. So, you know, I think thinking through that, I think is probably one of the things we think about is we're thinking through, you know, where are there opportunities for startups,

versus kind of where do the incumbents hold maybe the bigger sway. Yeah.

Prateek Joshi (05:14.812)
Right. You wrote a post back in June, 2023 titled, The Rise of Generative AI, a tale of two venture markets. That was very interesting. So maybe, can you start with what the post was about? Like what was the viewpoint back when you wrote it? And also how has it changed since then?

Julia Klein (05:37.134)
Yeah, I would say so. So the post was written and sort of inspired by an observation that while 20, you know, the kind of period of 2022 and 2023, I would say from and probably most people in the venture community would agree was a pretty slow time for venture. And I think, you know, valuations were moderating from their peak. And in general, there just wasn't as much activity in the market.

But contrasting that with the absolute explosion of interest in generative AI when chat GPT first hit the market, and I think everyone realized that there was this whole capability out there. And there has since been a frenzy of how is it going to be impactful across different spaces, both for consumers and for enterprises. And so.

that post was opining a bit on the stark difference between what most companies were seeing and what the generative AI focused companies were seeing within the market, which was crazy valuations and incredible interest and just insane levels of growth. And so it was wanting to capture a little bit of that because it was such an interesting juxtaposition. And so wrote a bit about that and about the differences. And then,

A bit also about how we thought about it, right? And I think in general, we're pretty disciplined investors and we want to, as I mentioned, stick to those first principles when we're thinking about deals. And so laid out just a few different observations around sort of some of the ways that we were looking at these companies and things that we were thinking about as we were evaluating them as investment opportunities.

And then, yeah, it's been crazy since, right? I think it hasn't slowed down at all. In fact, I think it was close to 30 billion was invested last year just into generative AI. So absolutely frenetic pace and in general still very, very hyped. But actually there are a few companies, not all of them, but a few now that really are starting to see.

Julia Klein (07:56.398)
that incredible traction uplift that you can get. It seems unique to sort of the generative AI space. So yeah, I think it has continued to be, yeah, yeah, it continued to be crazy.

Prateek Joshi (08:00.732)
Let's talk about digital biology. Let's see of Nvidia, Jensen Huang.

Oh, right. Let's talk about digital biology. The CEO of NVIDIA, Jensen Huang, across many of his talks, keeps talking about how the next frontier for AI is digital biology. There's so much being done. And obviously, you also invest in the sector. So in general, what's your view on...

intersection of AI and biology. And because we've been here before, if you look at the history of venture life sciences, every decade or so, there's a new, newfound wave of excitement. So it has gone through ups and downs. So what do you think about this wave of AI and biology?

Julia Klein (08:55.566)
Yep. Yeah. Um, and so, you know, I think that, as you mentioned, we invest sort of across the ecosystem of where AI is impacting the enterprise, right? And I think a slightly, you know, different, you can think about it in a different way, but pharma companies and biotech companies are still enterprises. And I think there's tremendous potential to impact kind of both their day -to -day operations, but also.

the way that they actually do R &D and the way that they bring these drugs to market. So, you know, I think that for us, right, we've kind of invested within a few different spaces of that. One would be on kind of the drug discovery side of things. And that, you know, we have two companies in our portfolio, Generate and Tessera that are the first generators is using generative AI. They've actually built their own model called Chroma.

that designs novel proteins and that treats infectious diseases. It can treat autoimmune diseases. They do things in the oncology space. So they're actually, it's incredible how quickly they've been able to really get traction. They're partners with Amgen and doing really, really interesting things and have already filed a couple of INDs. So it's excited to see how that will kind of progress over the next couple of years.

And then the second, Tessera, is actually using AI to identify RNA gene writers. And so instead of just being able to go in and cut out pieces of DNA, they can actually insert, delete, they can rewrite genes. And so that allows them to treat things like sickle cell disease and other genetic conditions that kind of previously were not able to be treated. So that's another.

super, super interesting and compelling application on more of the actual therapeutics side of things. And we've also looked at stuff, really, really interesting stuff on the clinical trials side of things. Less than 10 % of really trials end up making it into actual therapeutics to get released into the market. And there are so many factors that go into that. And so questions around how can AI impact that, either through trial design or

Julia Klein (11:18.614)
or other things, super, super interesting space. And for us, I think, you know, we really love the potential impact there that it can have. And, you know, it's obviously also just a massive market.

Prateek Joshi (11:35.644)
And if you look at all the data that's needed to train all the AI models, synthetic data is being increasingly used across a number of sectors. Now, what's your view on the synthetic data opportunity? Obviously, there are a number of well -funded startups, and there isn't enough customer money to go around. So how do you think this market will play out?

Julia Klein (12:02.253)
Yeah, I mean, I think that I am, uh, I'm biased because we are invested in a synthetic data company, um, parallel domain. I think that the potential there is, um, very large on both kind of structured and unstructured synthetic data. So, you know, unstructured would be the data that's used right for computer vision models or.

Um, you know, while, whereas structured data is the more tabular data, right? That's going into, you know, financial decisions or, um, you know, other, other sort of types of data. And so I think within both of those realms, there is a huge opportunity because training all of these, you know, L even training LLMs take so much data, right? So, so, so much data. Um,

And I think we'll move towards smaller models in the future, but it's still going to be critical to have the right data to train them. I was, I was, um, noted at our, at our event the other day that Reddit, right, filed for IPO and 10 % of its revenue is actually coming from sale of its data to LLMs to be used as training data. So it's, it was so interesting to me. It just shows that there is absolutely a need in the market there. And I think.

synthetic data is one way to fill that need and a way that maybe is much more efficient than some of the historical collection of data that has to be done manually. So we're really interested in the space. I think it remains to be seen if all of that can be utilized in a way that's effective.

I think a lot of companies are still in the early days of their journeys there in terms of really figuring out like, what are we going to be using on the data front? But I think there is tremendous potential.

Prateek Joshi (13:55.964)
Yeah, it's actually, there is a lot of potential and it remains to be seen how it's going to play out. So it's very interesting. Also, when you look at computer vision, the sector has been going around for a while. It's a very important part of AI. And now, obviously, with more and more...

models, hardware, algorithms to see, that basically kind of enabling a machine to see. I think the sector is seeing another renaissance. So what's your view on the computer vision sector and also what opportunities are you seeing in the market?

Julia Klein (14:41.614)
Yeah, I think it is incredibly interesting. I would say there's a ton of, again, a ton of potential. I think also, though, it has been historically difficult really to get stuff. So let's use self -driving cars for an example, right? I think that the... I think that that...

is used as the poster child for different types of computer vision. And it probably is a good poster child for highlighting just how difficult some of these problems are to solve. I think, though, that really the space is and goes way beyond that. I think stuff we're seeing today that's interesting in the space and potential customers for, for example, the synthetic data company that we have invested in.

There's folks that are using computer vision for drone delivery, for supply chain, for manufacturing, for even smart homes. It's becoming also more and more on like a personal level. And generative AI has really accelerated the pace of being able to do things like 3D reconstructions, right? Where you're able to sort of go in and see a world and then recreate that and all of the applications that then can be that.

then be utilized for. So I think, you know, self -driving cars are sort of the first and the most commonly known use case, but I think there really are so many other broader aperture of computer vision applications. And that space is really, really being accelerated by generative AI.

Prateek Joshi (16:28.988)
If you look at the entire AI startup landscape, what's the biggest challenge that they're facing today? Let's say they've raised their series A. In fact, they've raised their series B. Let's say they have a lot of capital sitting on that. What's the biggest challenge? And also across your portfolio and also maybe outside your portfolio, how do you advise them to navigate through that?

Julia Klein (16:56.43)
Yeah, I think, you know, it's, it's, I think for a lot of the companies that raised on generative AI hype, the challenge now is that you have to build a business. And so beyond just having that, you know, that moniker, you have to really be effective in solving a problem, right. And for a customer. And I think some of the things that, you know, we,

talk to our companies about and really try and make sure that they have honed is first and foremost, right, having the talent, right, on board, which I think is even more important for a AI startup, a generative AI startup than it is for a regular startup, I think, regular. I think that, you know, that is, the talent wars are tremendous. And so having incredibly talented people on board, people who have a background in the space,

I think is sort of number one. So make sure that you have the right team. I think then understand, right, what problem are you solving for the customer and how acute is that need? And ideally you're addressing a problem with, you know, a very acute pain point. And then when you are going about solving that, how are you really having a quick time to value for that?

customer, right? I think that one of the roadblocks for a lot of AI startups that I have encountered is that there's these lengthy proof of concepts and it takes a while and then they really, customers are interested, but then there's a big setup time and then there's a lot of working and back and forth and there's a lot of customization and what ends up happening is it really...

becomes kind of a slog for customers to be able to see the value from what they're paying for. And so having that quick time to value that demonstrable ROI and having it be something that's easy to use even for teams that are using it that maybe aren't machine learning engineers, I think that is really, really critical, especially as you're thinking about penetrating different areas of the enterprise.

Julia Klein (19:12.558)
So we talk to all of our companies about these things and I think they have those and others that they keep in mind as they think about sort of KPIs and the metrics that matter.

Prateek Joshi (19:27.74)
And if you look at your previous investments, right? So obviously some of them worked out better than others. We don't want to admit it, but we always have the two groups. So if you look at what the characteristics among the companies of founders that worked out well, like what are the, obviously it's hindsight now, but what are the characteristics of the companies that worked out well versus maybe the ones that didn't.

work out that well.

Julia Klein (19:59.406)
Yeah, I think for those that worked out well, one, I think it's incredible teams and not just incredible founders. So I think that goes beyond one individual and to the ability of that individual who founded the company to really build out a talent moat around themselves. And I think that...

is huge and really can't be replaced when you have the right people in there. Usually you can figure out most of the rest of it. And so I think that is super, super critical whenever you're looking at any company, right? Even whether it's an AI company or it's not an AI native company, truly that I think that is piece number one. And you can't, there really isn't much you can do if you don't have the right team.

I think secondly, I think to what I was speaking about earlier, it's understanding and being able to have a solution to an acute pain point that is priority number one or two. For example, we were investors in CrowdStrike. I think the reason that company did so well is because it was priority number one or two within the security space and point protection.

was incredibly critical. And so, people were looking for solutions and really, it was a need. And so I think finding that and kind of serving that mission critical priority number one or two, within whatever space the company is operating in, I think that's really important. And then finally, just having something that allows customers and I mentioned it, but to be able to see...

and really experience the value quickly. I think the quicker that you're able to see value, the more likely you are to expand, less likely you are to churn, the more likely you are to really want to engage. And I think that's how products that get used a lot are usually the ones that end up being successful within the enterprise. And so that sort of ease of use and quick value to a customer, I think, can't really be understated.

Julia Klein (22:18.478)
And I think those are things that sort of go across AI and traditional, however you want to think about traditional software or SaaS companies. And yeah, so I would say that. And then for AI in particular, if you do have a specific data mode, so for example, for Generate for us, the data was the data mode is that set that they're using to develop.

right? And it's all of those, all of that data that they have on these proteins that they're able to feed in. And it is, if you have that type of unique data, that can also be a competitive advantage. And I think can give you a leg up, especially within some of the vertical sectors that are now being serviced by AI.

Prateek Joshi (23:07.196)
Right. And when you look at a growth stage, AI startup, right, different people look at different metrics and you mentioned metrics earlier. So I want to dive in a little bit. What metrics matter the most to you when you look at a growth stage AI startup?

Julia Klein (23:26.83)
Yeah, I think one thing that doesn't get mentioned all that much that I think will probably be coming to bear in the next year or two as companies sort of come out of this crazy hype cycle is efficiency. You know, I think we're growth investors, right? So we are thinking about how, you know, how do these companies become standalone public companies in a lot of cases, right? And so a big portion of that equation is efficiency.

It's sales and marketing efficiency. It's really cash burn and how efficient they are there. And it's a big, I think it's a big question for a lot of companies that are raising these massive rounds of capital. And so I think that, you know, that's definitely one that we pay attention to. As I mentioned, right, definitely the time to the value and the demonstrable ROI for customers.

I think also something that we look at specifically for AI companies, and this manifests in a lot of ways, but is really how much custom work is being put in versus how much is kind of rinse and repeat. And so what does that, whether you look at it as a kind of software services split or whether you look at margins or gross margins, they're all sort of pointing to the same thing, which is...

how much of this is repeatable and scalable versus how much is, you know, we're doing a project for a customer. And that can be a trap that AI companies I think can fall into. So we'll look at that as well. And then all of the normal metrics, right? Revenue, growth, you know, quality of revenue, retention, all of that stuff as well.

Prateek Joshi (25:12.156)
What technological breakthroughs in AI are the most exciting to you as an investor because they're going to hopefully unlock net new opportunities for you at the growth stage?

Julia Klein (25:25.742)
Yeah, well, it's it. I think everyone has been talking about the the first AI engineer, right? Like first, the first developer, I guess. And so I think it'll be interesting to see what comes of that and what the next personas are of that, because I would I would.

I would absolutely be thrilled if they came out with an AI investor. I would love to be the human in the loop on that and have somebody who's like helping and then kind of feeding great, great ideas. I think it's, yeah, I think that one's further down the road just because of the amount of judgment that's used. But I do think that there will be more opportunities like that. Like I think there will be a lot of, and I don't think they're that far off down the road. I think.

to the point where they're actually replacing people's jobs, I think that's quite a ways, but to the point where they're helping and accelerating the work that people can do by kind of a lot of the stuff that we would do manually, being able to do in a split second, I think that there will be a lot of that. And so, we, for example, like we're invested in a company, a couple of companies actually that,

are focused on the customer experience space, right? And so what they're doing is they're augmenting customer support agents and they're already giving them that ability to like not have to go in and look up all of this information. It's right at your fingertips and here's suggested remedies for customer problems. And I think we'll start to see a lot more manifestations of that across different areas. So I'm excited for those and it will be interesting to see what areas get disrupted next.

Prateek Joshi (27:15.196)
Amazing. And when it comes to AI startups, obviously we talked about trends you're excited by. What do you think is coming next from founders in this area? Again, you mentioned AI developer, which is fantastic, but within the next 12 months, what do you think is coming next that's going to lead to growth stage opportunities for you?

Julia Klein (27:41.934)
Yeah, I think on the technology side, I think there's a ton of room to run in enterprise, right? I think enterprise spent 15 billion in 2023 on Gen .ai. And while that may seem like a massive number, I think it's only about 2 % of total software spend. And so I think there is still such room to run. And

I think for me, that's very exciting because I'm investing on the enterprise side of things. And only about a third of enterprises today are using AI across more than one function. And so to think about how many different functions within the enterprise from how they're developing product to how they're serving their customers to the sales motion and all of that, I don't.

It's not quite as sexy as the biotech stuff, but I think it's such a massive, massive market. And so it'll be interesting. I'm really excited to see how over the next couple of years, we start to move from sort of these POCs to actually demonstrable ROI and really be able to see the start of these ecosystems that we'll build, right? So, you know, and...

For example, in the 2010s, it was Snowflake and it was Databricks and the ecosystems they built around themselves. I think we're going to start to see some of that ecosystem growth. And you've already seen it begin with folks like OpenAI. But I think we'll start to see more and more of that. And so it's going to be an exciting next couple of years to be investing in this space. I personally am very, very jazzed about it, despite the high valuations.

Prateek Joshi (29:28.22)
Right. And with that, we're at the rapid fire round. I'll ask a series of questions and would love to hear your answers in 15 seconds or less. You ready?

Julia Klein (29:38.222)
Yes, I am, I think so.

Prateek Joshi (29:40.456)
Alright, question number one. What's your favorite book?

Julia Klein (29:45.44)
Anything fantasy. I'm a big fantasy fan. So any of Brandon Sanderson's books, amazing. But other than that, the Nightingale, Crime and Punishment, and the best read lately was The Will of the Many. Sorry, that's more than 15 seconds, but I could talk about books all day.

Prateek Joshi (29:59.868)
Amazing.

Prateek Joshi (30:03.708)
No, I love it. I'm a huge fan of fantasy too. Next question. What has been an important but overlooked AI trend in the last 12 months?

Julia Klein (30:14.374)
Somebody told me, was talking to me about this the other day. I think it's going to be something adjacent, which is planning for all of the energy consumption around AI and the need to build all types of new data centers to support it. So that will be interesting. And I don't think it's talked about much because it's not typically venture investing.

Prateek Joshi (30:37.98)
Right. What's the one thing about AI investing that most people don't get?

Julia Klein (30:44.518)
Something I alluded to earlier, which is that I don't think AI is replacing jobs anytime soon. It might eventually, but I think it's changing them. And I think it's changing the way people do them. And so I think the more important thing is getting people up to speed on using AI, right? In their kind of day -to -day of different areas.

Prateek Joshi (31:09.212)
What separates great AI products from the merely good ones?

Julia Klein (31:14.766)
simplicity and ease of use.

Prateek Joshi (31:19.9)
What have you changed your mind on recently?

Julia Klein (31:22.766)
Um, I think that I was originally thinking, kind of invested in data infrastructure for a while. And I was unsure of the degree to which it would really change with all of this AI, um, it, the gen AI wave. But I actually think that we're going to see some more massive companies.

spring up within that space and maybe an interesting opportunity for another platform player. So I've gone from, I'm not sure we will, to I think we're going to. We'll see. We'll see how it plays out.

Prateek Joshi (32:00.156)
What's your wildest AI prediction for the next two months?

Julia Klein (32:04.974)
Uh, I predict that we will have some degree of AI talk around the presidential election. I will bet that neither one candidate will claim that, uh, that AI has influenced the election in a certain way, or like it's going to become, I think it will become a headline, like news in the headlines around it.

Prateek Joshi (32:14.396)
And just.

Prateek Joshi (32:29.148)
Right, all right, final question. What's your number one advice to founders starting out today?

Julia Klein (32:36.942)
Um, I think it would be, don't forget the fundamentals of what makes a company successful taglines, like technology taglines come and they go. And at the end of the day, it's still about being able to solve a problem for a customer and, um, you know, how you're solving it will change from kind of decade to decade, but it's still about that at the end of the day. Um, so don't forget the fundamentals.

Prateek Joshi (33:06.076)
I love that. I think throughout his day, I think that's fantastic. Like problem solving, that's what different shapes and forms, different businesses, different tools, but that's what people do. That's what successful companies have always done to find a way to solve people's problems. Julia, this has been such a wonderful discussion. Loved your views on this. And also I think growth stage is not something that gets a lot of coverage, at least on this podcast. So I think I loved exploring what happens.

in that area because once you get past the early honeymoon phase, it's just raw business. You got to run it as a business. You got to generate revenue. You got to be efficient. You cannot burn a mountain of cash. So love the discussion. Thank you so much for coming on to the show and sharing your insights.

Julia Klein (33:49.868)
Absolutely. Thank you for having me. I appreciate it and really enjoyed talking all things AI.

Prateek Joshi (33:57.284)
Perfect. Give me one.