Navigating the AI Wave Through the Lens of Verticalization

With the ceremonious arrival of large language models, never before has such a transformational technical advancement been democratized at scale overnight. The resulting wave of AI first startups is unprecedented, and the utility created across the AI landscape is undeniable, but it’s also created a complex backdrop for VCs wanting to capitalize on this wave.

We are witnessing the AI-fication of the vast majority of early stage B2B investment opportunities, which is perhaps an effective strategy to get investors excited, but also creates a crowded market filled with many redundancies. Many AI first companies have experienced remarkable growth over the last year, and by all historical measurements it would appear product market fit has been achieved, but given the lack of barriers to entry and technical differentiation, many of these companies could struggle to achieve venture outcomes. Frameworks for evaluation at the pre-seed and seed stage, in which we spend most of our time at Forum, present clear challenges for horizontal AI solutions and we have therefore been increasingly bullish on AI powered verticalization.

Before jumping into the “why” let’s set the stage with the “what” and define horizontal and vertical AI powered solutions.

Horizontal AI solutions are designed to serve a range of industries and use cases, rather than focusing on a specific vertical or niche. The product is commonly applicable to business functions performed by a multitude of sectors that can be powered by non-specialized capabilities including natural language processing, automation, analytics, image recognition, and machine learning algorithms, to optimize operations and improve decision making processes. A few examples of these are virtual assistants, sales outreach tools, chatbots, data/ predictive analytics, and content generating platforms.

Unlike the broad capabilities offered by horizontal products, vertical AI solutions are designed with a specialized industry or use case in mind. They address nuanced pain points and processes specific to a domain and require deep expertise in the field to develop. Their algorithms, integrations, and data sets used to train models are also industry specific. Examples of vertical capabilities are industry specific document review/ analysis for insurance claims, medical imaging diagnosis, and quality control for manufacturing.

Given those two definitions, there are a few inherent challenges to the horizontal AI model that are particularly critical at the earliest stages of venture investing given their impact on an investor’s ability to evaluate and underwrite risk:

1. Low barriers to entry mean early market saturation: Our team receives weekly inbound of horizontal solutions such as sales, business intelligence, HR, and marketing tools, where synthesis of internal materials is combined with OpenAI to power those functions. Given this volume, it becomes incredibly difficult to diligence true differentiation and defensibility, and while none of these are necessarily “one-winner-take-all” landscapes, even identifying the handful of winners is challenging. Notably, one of the stronger early use cases for these solutions is code generation, and with Meta’s recent release of code generating, Code Llama, as an open source solution, we would not be surprised if more horizontal applications will get disrupted similarly.
2. Inferior data sets and lack of underlying knowledge of end-user workflows limit the ability to truly deliver transformational results for customers. Much of the initial wave of horizontal application layer solutions is ideal for less mission critical job functions and still requires a fair amount of human in the loop oversight. For AI solutions to truly be ready for the bright lights and disrupt mission critical job functions across the enterprise they need data supremacy and deep industry expertise. Horizontal offerings lack the access to valuable industry specific training data and unique understanding of complex workflows.
3. Likelihood of hallucinations and undesirable outputs: Training a model on a large, horizontal data set may result in a greater frequency of AI hallucinations, less likely to occur with a verticalized, niche dataset that is continuously refining a model’s output. Hyper verticalized models are positioned to deliver demonstrably more effective, safe, and efficient model outputs that translate into measurable business outcomes when compared to any horizontal offering.
4. A crowded GTM: We, as investors, are not the only ones flooded with the newly available tools, so are potential customers. Considering that horizontal solutions provide a “blanket” or a “one-size-fits-all” product, the customer landscape is not limited to any one industry. In a world where everyone is the customer, identifying the right segment and establishing meaningful market penetration is a complicated task. How does one solution stand out from the next if the sales pitch is seemingly the same? This presents a distribution hurdle and without an unfair advantage through a loyal community or network, is likely to slow down and/or limit growth. An argument could be made for customer segmentation through the lens of business size, i.e. catering to enterprise clients or SMEs, but the reality is that even this distinction is not enough to build a targeted distribution plan.

The above challenges affect horizontal solutions more drastically than vertical ones, however there is an ongoing battle that vertical tools remain vulnerable to: incumbents VS. entrants, and therefore an investment thesis centered around making vertical bets must take this reality into consideration. We are already seeing big players develop custom tools for employees, including McKinsey’s Lilli and BloombergGPT as well as other incumbents entering external territories, such as Amazon’s AI tool aimed to transcribe doctor visits. The question remains — could a young startup beat Shopify’s Sidekick, WayFair’s Decorify, or Adobe’s Generative Expand? As always, the most frustrating yet true answer is — it depends. We are of the opinion that each side has clear advantages and disadvantages, which can be opportunistically exploited.

We started putting together a market map of startups across a few verticals that we have been actively exploring and are looking to continue building out this landscape analysis.

If you are building a vertical AI solution, let’s chat!

To be included in our market map fill out this form.