Part 3: An Investor's Guide to Exciting and Less Exciting Frontiers in Generative AI
Since the start of the new year, we have evaluated hundreds of generative AI startups, allowing us to uncover interesting trends and develop informed opinions. In the final section of our AI deep dive, we discuss these findings to highlight categories within generative AI that we are least excited about and most excited about.
Not excited: Sales enablement and customer support
Why? This is often commoditized AI. We have evaluated half a dozen companies and seen a dozen more raise their pre-seed/seed rounds that are all doing some iteration of the same thing. These tools are essentially passively listening to and tracking different types of company data to surface pertinent information and provide actionable insights to sales representatives, account managers, or customer support agents.
For sales enablement companies, generative AI is used to streamline customer outreach and evaluate customer calls. That way, sales teams can receive immediate feedback and learn how to improve within the discussion.
For customer support companies, generative AI allows customer service agents to quickly find answers to questions based on stored institutional knowledge and can preempt well-written responses to customer inquiries.
In short, there is often little technological defensibility for sales enablement and customer support startups, so unless the founders have very unique customer acquisition channels and amazing backgrounds, it is likely not a compelling opportunity.
Excited: Industry-specific AI co-pilots
Why: AI-copilots built for specific industries and businesses become highly engrained in the day-to-day responsibilities of their users. That way, churn is minimized in times of financial hardship, and there is a competitive advantage built around customer usage. Moreover, from a data perspective, these use-case-specific LLMs are often trained on unique proprietary data, which should create a technology moat at least in the near term against new incumbents.
For example, we evaluated a business that created an AI clinical assistant to automate patient intake and clinical interviews. The company is focused on a specific industry, use case, and data strategy and is trying to solve burning pain points across healthcare staffing shortages and inefficient patient intake processes.
Another example is a company building an AI co-pilot for insurance brokers. Their co-pilot will help agents dig through extensive carrier documentation to receive answers immediately and will directly autofill everything inputted on any insurance carrier portal. It will also recall all interactions with a client and automatically contact carriers to get a quote, compare the insurance options, and reach out to the client with cost information.
Not excited: Horizontal and consumer-focused AI co-pilots
Why: Consumer interests are very fickle, so with any consumer startup, I am often concerned about churn. Given the high volume of first-generation consumer-focused AI co-pilots going to market, few of these products will likely be sticky enough to generate long-term usability.
For example, there are startups building AI co-pilots to become browser extensions. That way, when a user asks ChatGPT for the best restaurants or vacation ideas, these tools will actually go ahead and book the restaurants and flights for them. Other companies are building AI executive assistants to automate busy work by connecting to email inboxes, automating responses, and setting up calendar meetings.
These types of companies are first-movers and can definitely add value, but they do not present any substantial moat from a feature perspective. Just think: what is stopping OpenAI from building a browser extension co-pilot? Or Gmail or Outlook from creating a proprietary email executive assistant? They have the existing distribution to influence rapid adoption, which would quickly upend a new entrant.
Excited: Vertical generative AI marketplaces
Why: Vertical marketplaces are solving the deep-rooted information asymmetry that plagues countless industries. In the past, verticalized marketplaces often functioned as service businesses as they had to manually source and match supply with demand, especially in niche industries. Now, with generative AI, these marketplaces can more efficiently source opportunities by scraping large data sets, identifying leads that are a strong fit, and then providing tailored recommendations to a buyer or seller. Meaning generative AI can more accurately match supply and demand based on predicted outcomes.
Lastly, vertical marketplaces are often focused on collecting data for poorly mapped and fragmented industries. This data holds significant value as it can be used to create software products in the future and, at a minimum, make their sourcing and matching models even stronger.
For example, we spoke with a company building a marketplace for biotech firms looking to outsource their clinical trials to CROs (contract research organizations). For context, the costly factors behind drug development include R&D inefficiencies, lab space constraints, high failure rates, and talent acquisition challenges. These pain points are primarily derived from an efficient matching system between biotechs and CROs. Thankfully, AI is now allowing the platform to build a scalable, data-driven marketplace that can efficiently match based on biotech trial requirements and predictive models that assess potential outcomes.
______
There are obviously many more trends to be excited about, but these are a few we have uncovered from our due diligence and conversations. We are keen on seeing if they hold true over the next few years and how the industry landscape shifts. Until then, if you agree, disagree, or have any comments, questions, or concerns, feel free to reach out! We would love to hear from you.
