Part 2: How to Evaluate Generative AI Investment Opportunities
Beyond the technological background and context we reviewed in our first post, there are still many components to generative AI that are critical to understand when evaluating the technology. In this section, we will explore the key questions to think about when trying to reach conviction on a generative AI investment opportunity.
What type of model is the company building off of? Is it a hyperlocal model or open source?
This is often where experts most frequently diverge in the space. Some believe that access to property data and building a hyperlocal model will create a significant moat. Others are of the mindset that in the coming years, open source models will become so powerful, accessible, and cost-effective that they will outperform, or at least be at parity with, the majority of hyperlocal models.
Regardless, most almost unanimously agree that some sort of proprietary data that can be trained with an open-source model will create some sort of moat, at least in the near term. The question of how big the moat will be can only be answered with time, but it is hard to imagine that hyperlocal models will prove to be more effective than open source models with large dedicated teams and significant budgets. Because of these disagreements, many investors have honed in on identifying application layer solutions given the business model parallels they share with traditional SaaS solutions that make them easier to evaluate.
Does it matter if the application layer solution is horizontal or vertical?
In short, no. However, there are specific nuances to each type of solution worth considering when evaluating the opportunities.
Horizontal application layer solutions are tools that have wide-spread applicability to multiple industries and stakeholders. Horizontal consumer applications include AI powered assistants like digital secretaries, travel agents, and coaches, while B2B examples include customer discovery, employee sentiment tracking, and website design. Although horizontal application layer use cases add significant value, they can often be considered “commoditized AI” given that they are often built on top of open source models with indefensible fine tuning.
For that reason, when considering horizontal AI opportunities, non-technical questions need to be more heavily leaned upon, such as understanding the company’s speed to market, unique distribution channels, and competitive wedge. It is also important to be very mindful of product-market fit, especially for early-stage opportunities. There are many horizontal AI companies that have very impressive products from a tech-perspective but have yet to find the use cases where they will add significant value.
Verticalized AI opportunities are a bit different. They are built for specific industries, use cases, and workflows so that they become core to the day-to-day responsibilities of their users. Some early examples of vertical use cases are in the legal, insurance, healthcare, and construction industries. Data can prove to be a competitive moat as it can contribute to hyperlocal models built around one specific use case, whereas open source models may lack the industry-specific data inputs to perform at the same level.
Moreover, non-technical considerations like speed to market are not of much concern given that a verticalized approach can always upend a horizontal competitor offering a retrofitted, less applicable product. Product-market fit is also usually stronger with a verticalized approach, but it is still important to be mindful of whether the industry has unique workflows that necessitate industry-specific software. That way, you can ensure that the product being built is centered around a specific pain point.
One pitfall to consider, however, is the number of selling opportunities in the given industry. It is especially important to review market size for vertical opportunities, given that they may lead to smaller outcomes.
Where will value accrue?
This is the million dollar question, and no one truly knows the answer. It is up for debate whether proprietary data will prove to be a longstanding moat. For some use-cases it can definitely carry some defensibility, especially in the early years of generative AI, but time will tell if open-source models become the more efficient building blocks of a winning company. Moreover, there is no right answer as to whether verticalized or horizontal opportunities are stronger; there are just different factors to key in on when evaluating the opportunities.
Because of all of the outstanding questions around generative AI, it is important for investors to not forget about the fundamentals of evaluating any startup. Time and time again, customer usage proves to be a long-standing moat, so it is important to stick to the first principles of early stage investing to evaluate a business’ potential.
Team: Are these the best founders to execute on the given problem?
Problem statement: How acute is the pain point they are solving for?
Market size: Is this a large enough market for there to be a successful outcome?
Competition: Who is already playing in the space, and is there room for multiple winners?
Customer acquisition: What unique channels do the founders have to acquire customers?
Business model: How is this business monetizing, and what is the path to profitability?
These are only some of the factors that are crucial to evaluating a startup. However, if the company is able to provide strong answers to these questions, they will likely attract a strong user base, which, at the end of the day, is where value usually accrues regardless of industry.
