TL;DR

• The first wave of vertical SaaS was the system of record, leaving the “grey work” (i.e., judgment, interpretation, drafting, coordination) to humans
• Vertical AI is the next wave, where software directly executes much of the grey work (i.e., reading unstructured inputs, generating expert output, reasoning through decisions, executing workflows)
• The most durable vertical AI companies start with a high-stakes “AI wedge” at a critical point in the workflow, using automation to shift value from data entry to decision-making
• Defensibility comes from domain mastery, combining proprietary data and accurate workflow automation in mission-critical use cases
• AI shifts value from record-keeping to decision-making, with “systems of action” becoming the primary surface where work is executed, while systems of record increasingly serve as downstream data repositories
• Rising productivity expectations, labor shortages and newly usable unstructured data create a once-in-a-decade opportunity to create value in vertical AI, system of action applications
• Opportunities for an AI wedge are particularly powerful in previously “too messy for software” spaces like government, healthcare, legal, construction and manufacturing, or in previously difficult modalities like voice, video and image data

The First Wave of Vertical SaaS – Software Becomes the System of Record

Less than two decades ago, vertical SaaS was considered too narrow to support venture outcomes. The 2010s proved that assumption wrong. Category-focused companies became multi-billion-dollar businesses because they digitized workflows and became the central systems through which industries operated. They captured structured data, owned the workflow spine and layered on payments, communications and financial products that deepened customer engagement (and dependence).

Mindbody and Weave, both former Catalyst investments, illustrate this clearly. Mindbody replaced clipboards, calendars and cash drawers with a single operating platform for fitness and wellness businesses. Weave unified communications and customer engagement for dental practices and small healthcare providers. Each transformed messy offline workflows into a single pane of glass and captured the operational exhaust of millions of SMBs. These systems brought order, standardization and predictability to industries that had relied on manual coordination.

But even the best systems of record left a huge gap: the “grey work.” We define grey work as everything that happens between the visible steps of a workflow: interpretation, judgment, searching, drafting, reconciling and stitching that only humans could do. Legacy software digitized the steps, but it couldn’t execute them.

Vertical AI introduces new layers of functionality, enabling software to participate in the work that it previously just documented.

AI Raises the Ceiling for Vertical SaaS

LLMs and AI agents unlock new capabilities that were unavailable during the first wave of vertical SaaS. These capabilities align perfectly with the messy, nonlinear, unstructured grey work that traditional software couldn’t touch. Now, AI-native software can read unstructured inputs, generate expert-level content, reason across constraints and orchestrate multi-step workflows. What legacy systems could only store and track, software can now interpret, create, decide and act on.

As a result, verticals that historically resisted robust software adoption because their work was rooted in unstructured content, nonlinear decisions and / or field activity (e.g. government, healthcare, legal, construction, manufacturing) have become high-leverage verticals for AI. The very qualities that made them “too messy for software” now map directly to AI’s strengths.

Vertical AI will increasingly take on the grey work in these laggard sectors, including:

• Reading and responding to RFPs in government contracting (i.e., Procurement Sciences)
• Capturing clinical expertise from conversations in healthcare (i.e., Abridge)
• Automating revenue-critical communication, quoting and execution in roofing (i.e., ProLine)
• Interpreting siting and permitting data to help energy developers quickly assess project viability and approval risk (i.e., Paces)

These tasks previously required human judgment and now software can shoulder meaningful portions of them. When administrative and interpretive overhead shifts into software, industries unlock capacity, labor bottlenecks loosen and humans spend more time on relationships, strategy, sales and craft.

As AI absorbs more of the grey work, the question shifts from what AI can do to where AI should enter.

Choosing the Right AI Wedge

Once AI makes previously difficult workflows automatable, positioning becomes the strategic unlock. Not all starting points create the same leverage. The most durable vertical AI businesses begin at a critical point in the workflow where solving one painful, high-stakes problem generates gravitational pull for everything around it.

We see a few wedge strategies emerging as consistently powerful:

1. AI that makes a previously impossible digital workflow possible

This is category creation. Our most recent investment, Procurement Sciences (“PSci”), exemplifies it. Government RFPs often exceed hundreds of pages, with shifting requirements, compliance clauses, evaluation criteria and hidden dependencies. Software could store these documents, but it couldn’t take action. Humans handled every task, including reading, interpreting, tracking compliance, mapping requirements to past performance and drafting from scratch.

LLMs changed this because AI can now:

• • Read the entire proposal
  • Extract requirements
  • Identify compliance gaps
  • Surface risks
  • Generate responses based on historical proposals

PSci transforms proposal creation from a brute-force reading and writing exercise into an automated workflow that software can meaningfully participate in.

2. AI that generates net-new proprietary data

Voice becomes structured clinical documentation in healthcare. Video becomes structured incident and safety intelligence in manufacturing. Contracts become machine-interpretable obligation maps in legal.

This LLM-generated data has three structural advantages:

• • It differentiates because competitors cannot easily replicate it
  • It compounds since every action improves task-specific performance by generating additional proprietary data
  • It unlocks new products as the dataset becomes the substrate for future automation

For clarity, proprietary doesn’t have to be the initial wedge. Many winning vertical AI companies enter by automating a painful task using existing data. Over time, however, the strongest platforms generate net-new proprietary data as a byproduct of usage. That data compounds beyond accuracy alone; strengthening reasoning, encoding edge cases, improving judgment and unlocking new categories of insight and automation. In practice, proprietary data becomes the engine that turns an initial wedge into a durable platform.

3. AI that leverages low-frequency expertise for high-frequency automation

Many vertical markets rely on specialists whose knowledge is critical but unevenly distributed across the organization. Their work becomes the throughput bottleneck.

LLMs change this dynamic by transforming episodic, high-expertise work into continuous, software-driven capability. AI can review contracts every time they change, scan every opportunity in a pipeline, score every project plan for risk or validate every compliance clause. This wedge is powerful because it expands organizational capacity without expanding headcount, standardizes judgment that previously varied by individual and exposes patterns buyers could never see before. When AI makes scarce expertise always-on, it becomes difficult for customers to imagine operating without it.

In each wedge type, “10x better” is redefined. These wedges remove hours instead of clicks. They automate work humans dislike or cannot do consistently. And they position vertical AI companies to expand naturally once they own a mission-critical moment.

What Catalyst Is Looking For

Since the launch of ChatGPT there has been a lot of AI hype and it’s important to acknowledge the reality of today’s market. AI adoption is accelerating because boards and CEOs are mandating “AI readiness,” which drives strong early revenue and top-of-funnel activity. It also creates noisy cohorts because many deployments begin as experiments with unclear owners and unstable usage patterns. This makes underwriting more difficult, since early traction does not always reflect long-term revenue quality. While we’re still learning as we go, we wanted to provide some insight into our mental model for approaching vertical AI opportunities.

1. Clear ROI that supports durable monetization

ROI in vertical AI is measured by outcomes: reduced expert hours, faster cycle times, higher throughput per employee or lower execution and compliance risk. The strongest companies design their products, so customers realize this ROI through routine use inside core workflows. When outcomes are directly tied to day-to-day execution, ROI is legible to buyers, which supports expansion and long-term retention. Products that reliably increase how much high-stakes work an organization can complete, or how quickly and accurately it can do it, earn pricing power and are difficult to displace.

2. Defensibility comes from domain mastery, not simply code or data

Since model and UI differentiation erode quickly, early defensibility must come from domain expertise and customer truth rather than technical complexity. Winning teams live inside the buyer’s workflow, partner with subject-matter experts and handle edge cases that horizontal platforms will never capture. Customer Success becomes an even larger part of product development since every deployment surfaces patterns, constraints and failure modes that strengthen the system. In vertical AI, defensibility comes from knowing the job, absorbing its nuance and translating it into great product.

3. Accuracy is king for high-stakes buyers

In regulated, audited or mission-critical workflows, buyers optimize for correctness, not cost. In markets where customers don’t place a premium on accuracy, AI becomes commoditized as competitors can deliver “comparable” results at increasingly lower margins. The companies that win are the ones who target high-stakes buyers and whose outputs are consistently correct because they are driven by proprietary data and strengthened by a feedback loop that compounds via incremental context from each use. High-stakes buyers select vendors based on quality and are willing to pay for it.

4. Platform potential emerges when a company owns a critical wedge point

Vertical AI companies earn the right to expand when they control the point where the most important work happens. The workflow point tied to revenue, compliance or risk. When a company owns this point, three things follow naturally:

• • Data primacy: it sees the inputs, decisions and outputs that every other upstream or downstream system depends on
  • Workflow authority: the product becomes the primary place where complex work is executed, pulling activity out of legacy systems of record and redefining what those systems are used for
  • Decision leverage: if the output shapes what the business bids, builds or approves, the product earns permission to automate more of the surrounding process

In some ways this pattern mirrors the system of record era, but AI shifts the critical moment from data entry to decision-making, meaning the point of control now sits where judgment is exercised rather than where information is stored.

We’ve seen this with PSci. Proposal generation is the right wedge because it sits at the revenue-defining moment for government contractors. The wedge created platform gravity because every adjacent workflow (i.e., bid search, delivery etc.) ultimately needs the context, data and decisions generated at that critical moment.

Another example is Harvey. Harvey broke into large law firms by automating judgment-heavy legal research and drafting directly inside live matters. By sitting at the point where legal reasoning and work product are created, the platform generates proprietary context around how firms analyze issues, structure arguments and handle edge cases. That wedge has enabled expansion into drafting, matter workflows and advisory use cases, shifting leverage away from legacy legal systems that primarily store documents after decisions are made.

In our experience, platform companies are those that control the data, own the workflow and anchor the point where decisions get made. These companies pull the rest of the stack toward them because other tools increasingly rely on their outputs to function.

5. The end state is software as a trusted industry intermediary

Durable vertical AI companies use automation to make, or at least heavily influence, decisions. Most start by integrating with the system of record to reduce friction, but, over time, become the system of action where work is executed and decisions are made. This does not necessarily imply that systems of record are fully disintermediated, but their leverage diminishes as vertical AI platforms take on execution and judgment. The power relationship inverts where systems of record exist as downstream repositories, while systems of action become the primary surface where work is performed and value accrues.

Why This Vertical AI Moment Matters and Why Catalyst Is Leaning In

Multiple structural forces are converging:

• Every industry is reassessing its tech stack as AI unlocks new levels of productivity and resets expectations
• Labor shortages and expertise gaps increase the need for automation
• Vast quantities of unstructured content are becoming usable for the first time

These shifts create a once-per-decade window for new platform companies. AI also redistributes advantage toward vertical specialists because horizontal incumbents cannot encode domain workflow nuance fast enough. Vertical AI companies can move faster, learn deeper and embed more completely because they focus on one industry’s reality rather than every industry’s abstraction. The last decade rewarded companies that organized industry data. The next decade will reward companies that act on it. In moments of structural change, new champions emerge.

At the operational level, this transformation manifests as a redefinition of the human role inside the workflow. Vertical AI elevates human judgement, and these platforms allow teams to accomplish more with fewer handoffs, faster onboarding and higher output per employee by shifting repetitive interpretation and execution into software. Humans spend more time on strategic decision-making, relationship management and exception handling, while AI absorbs the grey work that previously constrained throughput. Over time, organizations that learn how to operate effectively with these systems develop structural advantages, as productivity and institutional knowledge become embedded in the workflow rather than dependent on individual experience.

We believe the next category-defining platforms will be vertical-first and built by founders who understand their industries more deeply than ever before. Whether an incumbent SaaS provider or an AI-native startup, we believe the winners will be those that combine deep domain workflow expertise with the right AI wedge to automate a sector’s grey work and finally fulfill the true promise of enterprise software.