Vertical SaaS vs. Vertical AI: a distinction with a (key) difference

AI is not the death of Vertical SaaS (quick rehash from Jan ‘24) Almost 2 years ago, in January 2024, I published a post on why AI is not the death of Vertical SaaS. Since then, AI capabilities have advanced at a rapid pace (movie trailer in Jan 2024 vs. move trailer today), funding for privately-held AI companies has exploded (2025 is likely to end with a 50%+ dollar increase in AI investment vs. 2024), and NVIDIA — a good proxy for the AI market overall — has seen its market cap roughly triple. Notwithstanding this incredible and indisputable progress, I believe the aforementioned post has aged well.

For those who have not read the post, let me summarize: a prediction was made on the All-In Podcast (minute 58:40) that vertical software companies are doomed since their customers can now leverage AI tools to build custom internal tools as replacement (this sentiment is still prevalent today). I strongly took the other side of that argument: my perspective was (and remains) that companies will largely continue to procure vertical-specific software from 3rd-party vendors vs. leveraging AI to build those products in-house (Jeff Horing seems to have taken a similar position in a recent Invest Like The Best episode, 1:22:30 in). My rationale was (and remains) that:

(1) For many verticals (e.g. construction, hospitality, legal, etc.), there is a lack of in-house technical talent and capability required to build functioning internal tools using AI;

(2) Even for those companies who have the technical capability to build these tools in-house, maintaining custom-built tools and managing all the corresponding data, privacy, and security issues would prove too burdensome and risky. For instance, this would still require expensive headcount responsible for maintaining the internally-built software and, if that headcount cost is anywhere near parity to purchasing from 3rd-party, it would almost certainly not make sense to build vs. buy. Additionally, the champion on any internal build (as well as his/her superiors) would likely be fired for any major issues that arise (e.g. a hack or data leak) vs. having the ability to shift blame/accountability onto a third-party vendor who would likely have much stronger security in the first place; and

(3) The same efficiencies that AI enables for in-house tool building also accrue to third-party vendors (who are generally more technical than their end-customers in these types of markets), enabling those vendors to build even better products and/or offer their products more cheaply (making the value proposition of an internally-built tool less enticing).

Overall, I remain skeptical that companies “vibecoding” their own internal software will lead to the death of Vertical SaaS. However, the more emergent topic of conversation within the startup/tech/investing world today is not whether internally-built apps will kill vertical software, but whether “Vertical AI” will kill vertical software. This is a question that is surfaced in every meeting with an LP or GP (and one I‘ll cover in a future post, though I think Tidemark and Insight, among others, have put out great content related to this topic), but there is a more fundamental question that needs to be answered first - what does “Vertical AI” actually mean?

Vertical SaaS + AI ≠ Vertical AI (and that’s ok) In today’s market, nearly every startup positioning itself at the intersection of an industry and AI claims to be a Vertical AI company. But in reality, most are still Vertical SaaS businesses, just ones that have embedded LLMs into their products. They’re leveraging AI as a feature layer, not as the fundamental architecture or value creation engine. These companies may be “AI native” (created in the post-ChatGPT era) or “AI fluent” (created pre-ChatGPT, though now implementing LLMs into their product), but not necessarily Vertical AI (to be defined shortly).

To be clear, that’s not a criticism; Vertical SaaS, with or without LLMs, can still be incredibly valuable (we are absolutely still investing in these businesses at Reformation). What ultimately matters is whether you are solving a real job to be done for your customer, and using the best available method to do so. In some cases, that will involve SaaS companies leveraging LLMs within their product and in others cases it will involve no LLMs. In either scenario, these companies are still fundamentally Vertical SaaS businesses as we have historically known them and can therefore behave in a largely similar manner to prior Vertical SaaS companies with respect to building, selling, pricing, and monitoring the efficacy of their software for their customers.

However, a new paradigm is emerging for which “Vertical AI” is indeed an appropriate title to distinguish this category from traditional Vertical SaaS, as it is fundamentally different in meaningful ways. Whereas, over the past decade, Vertical SaaS transformed industries by digitizing records and enabling workflows for sector-specific problems, Vertical AI represents a class of products built not just to enable workflows, but to make probabilistic judgments within them.

At first glance, the difference between Vertical SaaS and Vertical AI may seem subtle: both are domain-specific software systems built to solve real problems. But underneath, these two categories operate on entirely different principles: one is deterministic and observable (Vertical SaaS); the other probabilistic and non-observable (Vertical AI). Understanding the difference isn’t just semantics: it determines how you build, sell, price, and measure the efficacy of your product.

Vertical SaaS: Deterministic and Observable Vertical SaaS is built on deterministic logic, meaning its outputs are fixed and predictable given the same inputs. When you perform an action in a Vertical SaaS product (e.g. creating an invoice, logging a patient note, or scheduling a delivery), the result will be the same every time. This determinism makes the value of traditional Vertical SaaS observable: you can see the results immediately. The user can immediately see the impact of their action: the invoice is created, the note is appended to a patient’s profile, the delivery is scheduled tomorrow. Cause and effect are tightly coupled. In short:

Deterministic: Same inputs → same outputs

Observable: You can see and measure the result right away

Vertical SaaS digitizes and streamlines workflows. It provides structure and visibility, but it still relies on human decision-making to choose what to do next.

Vertical AI: Probabilistic and Non-Observable Vertical AI flips this paradigm. Instead of encoding deterministic workflows, it uses models trained on data to make probabilistic recommendations: forecasts of what’s most likely to produce a good outcome. The system no longer says, “Here’s the field you need to fill out.” Instead, it says, “Based on what we’ve seen, this is the next best action.” That suggestion is never 100% certain, and may differ slightly even with nearly identical inputs. The model is learning and adapting in real time, not executing static logic.

This shift makes Vertical AI non-observable in the short term. You can’t log in tomorrow and see the full results, especially on one specific action (though that action has a higher probability of being the correct one, it is not a certainty). Vertical AI’s full impact emerges over time across many decisions, many actions, and many feedback loops. To understand whether the AI is actually improving outcomes, you must compare aggregate results against your pre-AI baseline. The payoff curve is longer, but potentially steeper. In short:

Probabilistic: Same inputs → likely similar outputs, but never guaranteed

Non-observable: Results only emerge over a significant time horizon (and sample size)

The Implications of This Shift This shift from deterministic and observable to probabilistic and non-observable represents more than a technical distinction: it changes how Vertical AI companies should think about building, buying, and measuring efficacy of their software. It means embracing feedback loops and performance metrics that take longer to mature: you’re no longer optimizing for “time to value,” but longer-term “lift versus baseline”, changing the way success is measured. Instead of usage or engagement metrics, the signal lies in marginal gains over time (e.g. better sales conversions, lower readmission rates, higher yield), which requires longitudinal data to confirm.

It also changes how products are sold. Because results are non-observable in the short term, Vertical AI companies will need to ensure that customers trust the system’s recommendations even before proof is visible. Therefore, Vertical AI companies must: (1) find early customers who are willing to take a risk or leap of faith: buyers who understand the potential upside and are open to betting on probabilistic improvement; and (2) have a core competency — especially early on — in change management, as selling Vertical AI is much more about managing belief/expectations and closely guiding adoption than it is about demonstrating immediate ROI.

From Efficiency to Judgment / Workflow to Action Vertical SaaS improves efficiency: it makes workflows faster and more reliable. Vertical AI aims to improve judgment: it helps decide what to do next. That’s a fundamental leap. The former digitizes workflow; the latter optimizes cognition and action.

As industries continue to evolve, the most valuable vertical-specific companies will likely blend both paradigms, combining deterministic infrastructure (SaaS) with probabilistic intelligence (AI). The former provides organization and structure, while the latter provides adaptive insight. Both are not required, but combining the two has significant upside and defensibility. And, importantly, a company can masterfully combine the two by starting from either side: Vertical SaaS that layers in Vertical AI, or vice versa.

Vertical SaaS vs. Vertical AI examples To crystalize the difference between the two categories, below are some examples across a handful of industries of what Vertical SaaS does today and what emerging Vertical AI does or will do tomorrow.

Read the full article here: https://medium.com/reformation-partners/vertical-saas-vs-vertical-ai-a-distinction-with-a-key-difference-8816c6868bf4