Part I: The future of AI is vertical

After building one of the largest vertical SaaS portfolios in venture, we learned that great vertical software companies can unseat incumbents, transform industries and the way people work, and become highly profitable, generational businesses. Now that the top 20 public vertical SaaS companies in the US have a combined market capitalization of ~$300 billion, this perspective feels obvious, but when Bessemer started investing in Mindbody, Shopify, Procore, and others back in the early 2010s, vertical software startups were seen as “sleepy” and their potential was uncertain.

The meteoric rise of these businesses over the past 15+ years and the advancements in AI during that same period has set the stage for an exciting new development in the vertical software landscape: Vertical AI. This all came to a head in 2023, when we saw a new class of LLM-native applications harnessing novel business models and AI capabilities in order to serve functions and even entire industries that didn’t meaningfully benefit from the previous wave of vertical software.

Unlike their predecessors, these vertical AI applications are able to target the high cost repetitive language-based tasks that dominate numerous verticals and large sectors of the economy — such as legal, healthcare, and finance — that were largely out of bounds for legacy vertical software. Given Vertical AI’s ability to both capture new markets and tap into more sizable TAMs within those markets, we predict that Vertical AI represents an even larger market opportunity than that of legacy vertical SaaS.

There are already AI-first teams beginning to solve industry-specific problems using LLMs and generative AI. And then there are vertical SaaS leaders continuing to serve businesses with software solutions. (For the latter, it’s time to consider incorporating AI into your product, if it’s not been incorporated already. Look to Intercom, Zapier, and Canva for inspiration.)

In this first installment of our Vertical AI series, we focus on the dynamics driving this promising, fast-moving, and highly competitive category.

Vertical AI is changing startup physics in the enterprise software landscape; we’re seeing this emergent category address an even more massive TAM, offer new capabilities, serve new sectors of the economy, and grow at unprecedented rates. In Part I, we distill why this moment is so significant in software history, the key opportunities driving this category’s massive value creation, and the core and supporting workflows underpinning these emerging businesses, as well as highlight a few promising startups already gaining speed within their industries.

Why build now — the potential of Vertical AI

You might be thinking, “A sea change is certainly coming, but it’s not here yet and we still have time.” While it’s true that horizontal SaaS preceded vertical SaaS by a significant margin in the previous SaaS wave (with Salesforce going public an entire decade before Veeva Systems), we also know that many of our frameworks and timelines for SaaS businesses don’t always map perfectly onto AI.

We’re excited by the early data we’re seeing on Vertical AI business models. We’ll note that for the most part Vertical AI players are leading with functionality that isn’t competing with legacy SaaS. The utility of these applications is typically complementary to a legacy SaaS product (if one exists at all) and thus doesn’t have to replicate and displace an incumbent.

Analyses of our Vertical AI portfolio hint at the strength of this new class of applications. LLM-native companies in this cohort (with founding dates of 2019 to present) have quickly reached 80% of the average contract value (ACV) of the traditional core vertical SaaS systems, and are growing ~400% year-over-year, while still maintaining a healthy ~65% gross margin.

Based on the growth rates of more mature Vertical AI startups in this category, we predict we’ll see at least five Vertical AI companies with $100M+ ARR within the next two to three years. We also anticipate the first Vertical AI IPO in the next three years.

We’re already seeing exit activity among vertical AI companies through M&As. In 2023, Thomson Reuters acquired CaseText for $650M and a year later, DocuSign acquired Lexion for $165M. Incumbents are building as well as buying. In a recent survey, a majority of Bessemer investors who collectively oversee Bessemer’s vertical SaaS portfolio reported that companies across regions and sectors are rapidly incorporating AI features into their products.

Key opportunities in Vertical AI

There are numerous reasons why we see so much potential in Vertical AI and believe that current conditions are ripe for success similar to (and likely exceeding) what we saw in the previous waves of vertical software. Three opportunities stand out:

Expand total addressable markets (TAMs)

Historically, software builders have focused on the largest TAMs, ignoring niche categories where it would be difficult to build a big business. Vertical AI is able to unlock markets that previously would have been perceived as too small to build a sustainable SaaS business because they are significantly increasing the scope of value delivered through AI.

Take EvenUp, which automates demand letter generation for personal injury attorneys. By allowing firms to take on more customers at a lower cost — and thereby increasing margins — EvenUp has exceeded the traditional SaaS TAM that you’d expect of a solution that improves workflows for demand letter management.

The US Bureau of Labor Statistics cites software spend as 1% of the US GDP, and the Business and Professional Services industry — dominated by repetitive language tasks — at 13%. Based on this telling statistic as well as our own surveys, research, and observations, we predict that Vertical AI’s market capitalization will be at least 10x the size of legacy Vertical SaaS as Vertical AI takes on the services economy and unleashes new business models uniquely capable to serve this category.(We’ll share more on those new Vertical AI business types later on in our series.)

Unlock new functions and verticals

Vertical software replaced outdated and cumbersome systems and brought many industries — from construction to hospitality — online for the first time. But these developments left large swaths of the economy behind. In many cases, the ROI of a software solution alone wasn’t high enough to convince technophobic decision-makers and justify the upfront costs of setting up the required infrastructure, implementing new processes, and training employees.

Vertical AI startups have been gaining traction in markets that vertical software couldn’t access by providing solutions that can radically improve workflows and often take over tasks entirely. As a result, we’re seeing large incumbents within these industries become receptive to this technology, and sometimes even actively seek out AI-enabled tools out of concern that the competition will overtake them by adopting these tools first.

For example, in healthcare, an industry with notoriously long deal cycles for SaaS, providers are adopting solutions such as Abridge — which turns patient-doctor conversations into clinical notes — and ClinicalKey AI — an AI-powered medical search platform —- to take over busywork and support clinical decision-making. Law firms, which rarely even use CRMs, have also already begun adopting co-pilot based solutions for contracting, demand summary generation, case intake, and other time-intensive tasks.

Provide unprecedented value

There’s a future where, depending on capabilities, AI applications could be integrated into every industry in the economy, from home services through accounting. That said, the potential penetration of AI will vary by industry.

The most attractive markets to build Vertical AI companies will likely be in contexts where AI facilitates or completes work that was previously impossible or too expensive to achieve with human labor alone, and so the work was not being done or being done poorly. A common, successful use case for AI is automating or streamlining workflows by reviewing significantly more data than humans would previously have been able to audit.

For example, Axion Ray helps manufacturers by analyzing large volumes of product data across IoT & telematics, field failures, production, and supplier data. Similarly, JusticeText automatically reviews hundreds of hours of camera footage to help public defenders build their cases — something that’s extremely time-intensive for lawyers to do during discovery and which also takes away focus from building cases. Later on in this series, we’ll explain how the rise of multimodal models is allowing Vertical AI businesses such as JusticeText to go beyond purely text-and-data-workflows and also leverage voice, video, images, and other sources of input.

Vertical AI for core vs. supporting workflows

Winners in the previous wave of vertical SaaS created cloud platforms that were purpose-built for underserved markets, with many adding more and more integrated products and services over time to eventually produce a “layer cake” that provides an all-in-one solution for a given vertical and drives continuous growth (as Procore did for construction and Toast did for restaurants, for example).

As we’ve discussed in this article, vertical AI businesses can access larger TAMs within a given market by offering high-ROI solutions and therefore don’t always need to have as large of a product scope as their predecessors did in order to build successful businesses. In fact, some promising vertical AI startups are able to break into industries and drive returns by addressing just one or two of the target customer’s workflows.

We divide these workflows into two categories: core and supporting. Core workflows are those that are a primary function of a job; for example, financial modeling for an investment banker or contract drafting for a lawyer. Supporting workflows are those incidental to a job or business but still necessary; for example, marketing and patient relationship management for a dentist (e.i. Weave) or freight procurement for a shipper (i.e. GoodShip).

As companies of all types digitize in every industry, they need a software stack that helps them not only do the job, but also run and operate the business. Now, in the AI era, there are opportunities for companies in every industry to leverage AI in their trade or service and their business operations.

Core workflows

Today, text-and quantitative-heavy work have the highest propensity for automation. That’s why we tend to see more vertical AI solutions addressing the core workflows of industries that are dominated by traditional office work —- such as in legal and professional services — rather than those that require significant manual labor — such as home services and manufacturing. For example, portfolio company Fieldguide is revolutionizing the core workflows of auditors involved in diverse projects such as SOC 2 engagements, financial audits, PCI DSS assessments, and internal audits. By leveraging automation and generative AI, Fieldguide enhances auditor efficiency, leading to significant productivity gains in tasks where professionals spend most of their time.

However, just because a core workflow has a high propensity for automation doesn’t necessarily mean it’s a good use case for AI. A target customer’s desire to automate a given workflow matters too, and that will vary significantly across sectors. For instance, investment bankers may use AI to automate the tedious process of slide creation, but be unlikely to use a voice AI that gives presentations to clients because of the importance of relationships in the space.

Supporting workflows

Supporting workflows may be better targets for Vertical AI specifically because they are ancillary to the job of the target customer, and therefore are typically the type of work that can be delegated to and completed satisfactorily by AI. For example, a doctor has both expertise and interest in treating patients but less in, say, notetaking and paperwork (increasing doctors’ “pajama time”) or even ordering medical supplies. That’s likely why we see a high market demand for AI solutions addressing supporting workflows across back office operations, sales, procurement, finance, and other functions.

However, addressing supporting workflows with AI is not without challenges. For one, many tech-forward horizontal incumbents in these sectors have already begun incorporating AI into their platforms, and vertical AI startups will need to deliver meaningfully better solutions in order to compete.

On the upside, vertical-specific AI startups may be better poised to understand sector-specific nuances and integrate with underlying systems (such as CRMs) and therefore be able to create an experience that’s harder for horizontal competitors to replicate with a general LLM. For example, an AI solution built specifically for home services can identify a customer’s problem and route a technician to fix a solar panel faster and more effectively than a horizontal solution that can only make an appointment at the customer’s request. Still, founders will need to pay close attention to the potential TAM of any solution given that these forms of defensibility may come at the expense of market size and therefore require the layer cake approach mentioned above.

Whether building AI for core or supporting workflows, founders need to have good judgment, a deep understanding of customer needs, effective feedback channels, and a clear grasp of the regulatory landscape in order to identify the specific sectors and tasks that are well-suited for an AI solution. Remember: just because something can be automated, doesn’t mean it should.

A note on Vertical AI moats

One of the biggest and most interesting debates our team has on AI applications is about defensibility. A common criticism of AI applications is that they are just mere “wrappers” around third-party AI models, and that they don’t add enough value and are easily copied. While this is certainly true in some cases, we’re also seeing many compelling vertical AI applications that are far more robust and have built real moats (related to data, product depth, economic value, etc.). Given the speed that the industry is moving, it’s critical to understand and achieve defensibility, but what makes an AI application defensible is nuanced and evolving.

In Part IV, we’ll dive deeper into what Vertical AI founders need to know about building a moat, including key differentiators and initial paths to defensibility that we think can lead to market leadership.

Up next: Multimodal innovation

Vertical AI excites investors and entrepreneurs, but it also terrifies them. With jaw dropping products, early breakout growth, SaaS margins, and healthy TAMs come an unprecedented pace of development, intense competition, elevated valuations, and risks from incumbents who are not asleep at the wheel — and that’s just in the first inning. Given these competitive dynamics, we’re excited about the future of multimodal AI. In the next article in this series, we’ll cover developments in multimodal model architecture, exciting multimodal voice and vision applications, and the promise of AI agents.

If you are working on a Vertical AI application, we would love to hear from you! Please reach out to our team at VerticalAI@bvp.com.