Verticalized AI: A Primer on Learnings from Verticalized SaaS

Target Global
6 min readOct 18, 2023

By Joshua Stein

Artificial Intelligence (AI) has captivated the world by promising something that until recently seemed like a faraway dream — that a machine can achieve as much, if not more, than a human. While there’s unquestionably some hype and excitement around the topic, what’s clear is that from platform plays, to data pipelines and foundational models, we are witnessing a complete paradigm shift in software. That’s why we believe that verticalized AI applications will be some of the biggest value drivers to emerge from the AI boom.

Back to Basics: Verticalized Applications

To get a better idea of the path we need to forge ahead, we need to look at where we have been. When Software as a Service (SaaS) applications first entered the market, they aimed to serve a wide range of customers with broad functionality at the expense of creating a relatively thin layer of value. The reasoning was relatively simple: offer a product that can be useful for a wide range of customers and your Total Addressable Market (TAM) naturally grows in proportion to said user base. But as buying personas became more savvy about what was possible, horizontal applications were replaced with specialized vendors providing layered offerings. As the saying goes: “there are riches in niches”.

The parallels between what SaaS applications used to be and what AI applications are today are clear. Much like their SaaS counterparts, if you want to build AI products, they too can be split into either horizontal or vertical. However, unlike how SaaS came to the market, we believe that the order in which outstanding AI business will be built has shifted so that verticalized applications will take center stage before horizontal ones do.

Deterministic vs Non-Deterministic

How does an AI compute an answer? Particularly in Large Language Models (LLMs), outputs are predicted based on a probability distribution over the input set of words. As such, LLMs are, by nature, non-deterministic meaning in most cases, identical inputs will have similar, but not identical outputs. This inherently implies that end customers need to be comfortable with an AI’s ability to provide the answers to the problems they are having. This comfort is the divide that businesses that are building AI tools need to bridge.

To minimize approximation and achieve results that are as deterministic as possible, companies that want to succeed in the space need to take a more verticalized AI approach with models that are trained in domain based data sets. This is exactly where the divide between horizontal and vertical applications is most prominent; mission critical functions simply cannot tolerate the deviations a horizontal application may introduce.

Source: https://www.askpython.com/python/examples/deterministic-vs-stochastic-machine-learning

Building a GTM Wedge with Low Hanging Fruits

Much like verticalized SaaS applications, AI tools with specific industry-focused knowledge will have an easier time finding traction with customers. Just as with the merits of the deterministic vs. non-deterministic approach, it becomes clear that in a world where software spend is largely retreating, building out a very specialized product can be what makes or breaks the go-to-market motion of a business.

The most inherent pockets of value that we have observed where verticalized applications have found significant traction and user adoption include:

  • Co-pilots: These AI assistants leverage vast datasets to aid professionals including doctors, accountants, and lawyers in their work. While deep data sets are essential for these tools to be effective, they act as a step in the right direction towards improving work efficiency flows. A low hanging fruit in the co-pilot space is the healthcare industry. Healthcare professionals spend countless hours triaging patients and while doctors and nurses are highly trained professionals, they too can make mistakes. In fact, in the UK, more than half of the diagnostic errors made occur during a GP consultation, creating not only a huge financial burden on the healthcare system, but a potentially profound impact on the patient. Having a deeply trained AI co-pilot has the potential to alleviate this critical challenge.
  • Workflow Optimization: Much like the concept of co-pilots, AI tools that can automate and optimize complex workflows are gaining traction. These tools have extensive knowledge which gives them the ability to act like an extremely smart EA or additional co-worker, taking the burden off of human workers. A legal worker, for example, might be asked to review hundreds of pages of a shareholder agreement to understand the implications of a few changes made to a specific clause. Clients tend to be billed on hours spent reviewing documentation just as they are billed on actually receiving legal counsel. By having a tool work together with the AI to summarize changes made to a document, the client is able to enjoy a faster turnaround on their requests.
  • Industry-Specific Applications: Aside from healthcare, all spaces characterized by complex processes are prime real estate for disruption. Verticalized AI applications are emerging in the fields of accounting, finance, law, and many other fields where complex sets of rules are in place to sort information efficiently and reach an output.

The underlying thread that connects the above trends is the divide between internally and externally facing tools. Despite the significant value that can be derived from a verticalized application, buyers are still hesitant to unleash an off the shelf model to their externally facing end customers. Much like verticalized SaaS applications have done in the past, modules need to be layered on top of one another once an initial go-to-market wedge has been established. While there may be some hesitancy among buying personas today, we think that the fear will phase out as models become more adept at producing deterministic outcomes.

Source: https://assets.amuniversal.com/1dc58170065f013932bc005056a9545d

Overcoming Significant Hurdles: Data Moats & the Trust Factor i.e. developing a Tech First Attitude

It goes without saying that despite the recent boom of AI tools in the market, a general sense of unease is prevalent amongst the buying personas of modern day enterprises. While SaaS tools had to primarily contend with the notion of replacing existing, potentially analogue, capabilities within the buying organization, AI tools need to overcome one additional hurdle — gaining trust.

Buying personas need to be able to rely on the vendor’s ability to, at the very least, adhere to SLAs and not start hallucinating in front of the end customer. That’s how we see data quickly becoming the bedrock upon which a verticalized tool is built. The primary challenge for newcomers in this arena will be the sourcing of quality data and accumulating a substantial amount of it. Unfortunately, there is no “easy fix”, nor a way around it. Founders on a mission to build the next big verticalized AI platform must be prepared to dedicate significant time and effort towards conducting thorough research and analyzing data to be able to then gain the trust of the buyers they want to sell to.

Surmounting the barriers of acceptance represents another pivotal challenge for aspiring verticalized AI companies. First, in terms of technological adoption, it is essential to assess whether the target organization you are trying to penetrate as a business is ready to implement this type of technology. Many businesses lack the innovation hubs and general agility needed to adapt new methods of working.

There is also the fundamental issue of cultural acceptance. Employers must contemplate whether they are prepared to entrust AI tools with tasks mimicking their employees’ work and decision-making. On non-critical issues, these challenges can be overlooked and AI tools have thus far been readily accepted by governments, companies, employees, and even consumers. However, skepticism prevails when addressing critical matters. For instance, in fields like medicine, professionals demand vertical solutions tailored to their specific needs, eschewing horizontal solutions reliant on data from various sources.

All of the above can be distilled into a single northstar that founders need to be aware of: tech comes first. The best businesses in this space continue to live by this north star and have a team of technologists and researchers to back it up.

The Path Forward

The future of verticalized AI is a journey into uncharted territory. As businesses increasingly recognize the value of tailored, niche solutions, the user adoption of verticalized AI applications in critical fields will soar. While horizontal applications have produced the bedrock upon which the AI push currently sits, we believe that the time is now for verticalized solutions to make their mark by diving deep into customer segments, building trust, and delivering consistent results.

--

--

Target Global

Leading European tech VC with €3B+ AUM. Known for backing fast-growing startups & capitalizing on overlooked opportunities. 15+ unicorns, 21 exits, 7 IPOs.