Venture Deep Dives: Vertical SaaS in the AI Era

Frequently Asked Questions

FAQ

• Where is the transcript for this webinar?

  Hi everyone. Welcome to another deep dive from Alumni Ventures. Deep dives are our chance to share with our community our latest thinking about emerging technologies, trends, and startup businesses. My name is Catherine Liu. I am a managing partner at Alumni Ventures at our seed fund based in Menlo Park, and I help Alumni Ventures source and sponsor investments into the very earliest of startup companies. I’m very excited to talk today about vertical SaaS in this new AI era, and that’s in large part due to a genuine interest I have in both of these spaces. You see, I studied computer science at Stanford. I got my bachelor’s and master’s, and for my master’s degree, I actually focused on artificial intelligence. I went to school at a time when neural networks and deep learning were just coming to the forefront of application, and I’ve been following this space very closely since.

  After graduation, I started a company in the retail AI space called Fancy. We sold that company to Palantir, where I then worked on similar problems on the commercial side of their business.

  More recently, post-Palantir and before joining Alumni Ventures, I was Director of Product at a company called DataVisor that uses the latest in AI technologies to fight fraud, spam, and abuse on very large internet properties. Vertical SaaS in today’s AI era is a topic that I’ve been thinking about a lot recently, and so we’re going to go through quite a bit in a short amount of time. We’re going to start with defining the space, talk about the history of vertical SaaS, and then apply its relevance to today. We’re going to dive into that a little bit more deeply on a specific investment thesis that we have in the space, talk about key success factors and misses we want to avoid, as well as companies of interest. So let’s go ahead and get started.

  So first, let’s start with defining the space. What is vertical SaaS? Vertical SaaS, or software as a service, is a model where the software serves a very particular industry or vertical market. In other words, it’s a software solution that’s designed specifically for the needs and requirements of a particular industry. A lot of traditional SaaS offerings actually tend to be horizontal in nature, meaning that they serve a broad range of industries or markets. For example, a horizontal SaaS product might be something like Dropbox, which can be used by a wide range of businesses and individuals regardless of their industry or niche. Vertical SaaS, on the other hand, is tailored to the unique needs of a particular industry. This means that it’s often more specialized and provides features and capabilities that are specific to that industry. The go-to-market, of course, or the marketing and sales efforts, are also going to be different. And as well, vertical SaaS solutions can be found in a very wide variety of industries like healthcare, finance, real estate, and manufacturing. For example, a vertical SaaS product for the healthcare industry would be electronic health record systems, while a vertical SaaS product for the real estate industry might be property management software.

  This industry focus generally leads to better solutions. But is it appealing to VCs? The reality is that vertical SaaS was mostly ignored by VC investors up until quite recently—maybe in the last 10 years. Venture capital relies on investing in a large amount of companies and having a very small fraction of them actually produce outsized returns, like multi-billion dollar outcomes. This is the power law that is very prevalent in venture capital investing. So is the TAM large enough in a lot of vertical SaaS solutions to support VC dollars and that sort of growth? The answer now is yes. In the past, the cost of developing software was onerous, which made it difficult to create specialized solutions for specific industries. However, this has really changed as software development has become cheaper, easier, and faster. Now, with careful selection of the vertical market, VCs and founders can get the best of all worlds.

  This includes getting to strong product-market fit with higher certainty. You can think about it like if you have a narrower scope of focus, of course it’s going to be easier to make a solution that really knocks it out of the park. There are also going to be a lot of weak incumbents that you can outcompete head-on, and if you of course select the right market, these TAMs are actually large enough to support venture-scale outcomes. Now, why is that? How do you get to that large enough TAM? Well, what a lot of founders and VCs have realized is that if you build this kind of core operating system or platform for an industry and you own the critical workflows and that data, there are so many opportunities for revenue expansion and growth. You can add new products and features, you can add on marketplaces, you can add embedded FinTech like lending or capturing a small percentage of very large transaction volume. You can even expand into complementary industries. And to put this all into perspective on market sizing, most industries today spend a tiny fraction—0.3%—of their revenue on software that powers their business. So considering that software is one of the greatest productivity drivers in human history, this number can and should be much higher over time. And now think about what the possibilities are of growth when you add AI into vertical SaaS.

  There are already many examples of very large successful vertical SaaS companies. Some examples include Toast for the restaurant industry, Procore for the construction industry, and Veeva for the life sciences and pharmaceutical industry. At Alumni Ventures, we’ve also invested in a number of vertical SaaS startups. Examples are Sense for laundromat businesses, Prada for food delivery companies, and Operative Intelligence for call centers.

  So how do these companies go from tiny startups to very large public companies? This part by TCV does a good job of explaining the steps to get there. And obviously this is aspirational, but steps one, two, and three are really about becoming a leader in the category and eventually becoming that system of record for that vertical. So it’s very difficult to, for example, use the system and then turn it off of it. And when you are able to have many workflows in just one single product, you can actually deliver a lot more value. Then four, five, and six here are all about how do you expand into new segments or maybe other verticals from that initial point of winning that initial vertical. If you’re interested in learning more, I have the source linked here, but this is a really great way of thinking about how these companies go up and achieve their full potential.

  Now we’re going to take a step away from vertical SaaS and talk about this modern AI stack. You can see in this stack we have four categories from the bottom layer—Infrastructure—all the way to the top—App layer. We can think of a lot of AI companies as being in one of these four categories. At the bottom, again, for Infrastructure, it’s really the companies that develop the hardware, the chips on which a lot of these large models are trained. These chips are created so that they can do the type of math very quickly that is required for training these large models. Nvidia is an example of a company that has really benefited from this AI boom.

  One layer above that is this category of Foundation Models. These are very large models, generally trained on unstructured data, and the resulting model can actually be applied to a wide variety of tasks through additional fine-tuning or perhaps just as general use.

  So one great example of this is GPT. A lot of you have probably heard of GPT, which is created by OpenAI. One other interesting foundation model is a large language model like GPT, but specifically trained to be able to be used for healthcare use cases. So Hippocratic AI is a company that is creating their own foundation model, but specifically for healthcare use cases. And they recently raised a very large $50 million seed round from top investors.

  The layer above Foundation Models is AI Tooling. An example company is Hugging Face. These are companies that are creating the software to make it easier for data scientists and machine learning engineers to potentially use foundation models and other types of models and apply them for actual applications, which leads us to the final layer at the top of the pyramid.

  Applications are generally used by the typical consumer or maybe business user. They don’t have to be technical; they don’t have to be a data scientist in order to use them. And within this App category, there are again horizontal and vertical plays, which we’ve touched on previously.

  So of course we invest in all of these categories, but what we have found is generally the quantity of companies tends to be kind of this pyramid but inverted. So there are a lot of companies we see that are creating AI applications, fewer in AI Tooling, even fewer in Foundation Models, and very few in Infrastructure. Why is that? It primarily comes down to the capital required to build very large venture-scale companies at these lower layers—and expertise as well. It can be harder to apply the latest in cutting-edge research to those lower layers, but primarily capital.

  So, for example, GPT-4—Sam Altman has mentioned that it’s taken over $100 million in order to train the latest GPT model, GPT-4. And Hippocratic AI again raised $50 million just in their seed round. For comparison, a lot of seed rounds we see are maybe two, three, or four million dollars. That’s why we see most of our investment opportunities at the App layer, which is at the top of the pyramid.

  Again, we will look at both horizontal and vertical AI applications, but why is it that we get sometimes more excited about these vertical SaaS applications with AI as opposed to horizontal?

  Here’s our investment thesis: We find that for early-stage horizontal AI apps, we really have to think critically about their differentiation. There is generally a lot more competition from many other startups as well as the looming threat of very large incumbents in the space. Of course, this can be true for vertical AI apps as well, but generally we find that, for example, there are more AI companies now that are trying to make sales teams or marketing teams more efficient across verticals, as opposed to companies that are really trying to focus on an initial more niche area.

  And so we find that vertical SaaS plus AI applications tend to have fewer competitors. They also might have a unique data advantage because they might just be able to be integrated more deeply with the workflow. And of course, they have a more focused go-to-market, which is generally very advantageous for a young and resource-strapped company. So the result is that for these vertical SaaS AI applications, there could be a higher chance of a very large winner-take-most outcome, which is of course what we’re looking for when we are making our venture capital investments.

  So to summarize some key success factors and checkpoints we look for in these companies that we’re interested in is having real defensibility around the product. This can be true again for both vertical SaaS and horizontal SaaS—just being a vertical SaaS product doesn’t necessarily mean you do have real defensibility or that you have real product differentiation.

  So we’re looking for that. And for instance, not just a thin layer over GPT APIs. Prompt engineering has been a very hot topic or there’ve been a lot of people that kind of talk about, oh, what’s the best prompt engineering? But really it’s kind of like saying what’s the best way to Google something? Sure, there is maybe a little bit you can have in sort of the secret sauce for prompt engineering, but there’s not really a moat that’s defensible. And so we’re really looking for maybe—are there real additional workflows that support this sort of vertical SaaS product or real data that is being collected that is a differentiator, and of course real value created through AI.

  So looking forward, we’re really just putting everything together. We want to invest in companies that solve a real pain point for a large vertical market. This could be a large market that hasn’t really been a great beacon for talent, for example. We also are looking for companies that use AI for actual efficiency gains and improved quality in terms of what they’re able to achieve for their end user. Of course, we want deep moats around the product, the go-to-market. There should be great trade secrets involved as well, and eventually being viewed as the leading software provider in the respective industry.

  So misses are if it doesn’t solve a big enough pain point—if you kind of reimagine what the best-case scenario is, that the company just can’t be a multi-billion dollar outcome, that’s unfortunately not really investable for most VC investors—or if the AI does not create a step-function improvement in the product.

  This point is actually really interesting. There’s a blog post by Elad Gil, who’s a serial entrepreneur—we’ll link it at the very end of the slide deck. And he talks about formulating hypotheses as to why the initial AI wave didn’t really create many large AI startups who eventually became successful. The kind of prior AI wave 10, 15 years ago where we had neural networks and deep learning really being applied very well and a huge step-function in computer vision, for example—a lot of that value actually accrued to the very large technology companies that already existed. So Google, Facebook, Amazon with Alexa. And a lot of startups tried to apply AI, but most of them fell short. They just didn’t become large companies. And one hypothesis as to why is that the first wave of AI just didn’t create a good enough step-function improvement for most products that were offered by startup companies.

  With this latest leap forward with large language models, for instance, and generative AI, we think that there are new opportunities that really are unlocked with these step-function improvements.

  And finally, moving on—it’s a miss if the marketing or the branding is not on point. Obviously, go-to-market is extremely important. It goes hand in hand with a great product. And so if that company is just not highly perceived within its respective vertical, then it’s not going to become that category leader.

  Here are some venture companies of interest. We have Minify, which helps automate software documentation, replacing or augmenting technical writers. There’s also Booth AI, which can take a single product photo on a white background and transform it into images of lifestyle and high-end product photos. So for example, that person you see is created by Booth AI. That person actually does not exist, and that photo of the jacket was fed into Booth AI to create such lifestyle photos. Pretty crazy what generative AI can do right now.

  I only provided two examples here because we are constantly seeing companies every week that we talk to that can be categorized within this vertical SaaS AI application category. And a lot of them are stealth—they’re not ready to announce what they’re doing yet. One that we’re talking to very broadly is in the real estate vertical—a little bit more niche than that—but don’t want to give too much away. And another is, very broadly, legal.

  So we think that this field is obviously very rapidly evolving. We’re extremely excited to dig more into this area and make some investments here.

  Finally, I want to leave you with just some additional resources that you can listen to or read that can help inform your thinking about this. No one obviously knows the future here, and so I welcome any comments or feedback as well. That’s the fun adventure—that we get to kind of think about what the future could be. And the really interesting thing about startups is that if enough people believe in a direction, then oftentimes it can be a way of predicting the future. And with all of the money and talent flowing into this kind of new space, we’re very excited for our investment opportunities for the next year or so.

  To summarize, vertical SaaS has more recently but already proven itself to be a very profitable sector for VC investment. The latest AI breakthroughs really unlock additional product differentiation and work that can be disrupted. A lot of competition in horizontal SaaS plays and AI infrastructure from other startups and incumbents abound. But for vertical SaaS, we tend to find, again, with the right mixture of ingredients, there can be really explosive tailwinds behind them. And so we are definitely keeping a lookout for these differentiated opportunities with AI superpowers.

  Thanks so much for listening to this deep dive from Alumni Ventures on vertical SaaS in today’s AI era. If you are interested in learning about any other topics, we have a bunch of additional deep dives—for example, one on generative AI done by my colleague Tom Meyer. Again, a great primer, especially for those who are interested.

  Thanks so much for joining us, and have a great day.