Building Vertical AI: An early stage playbook for founders

Essential insights and best practices for AI startups ambitious enough to reimagine technology across vertical industries.

_{Want something even shorter? Get the quick read from the vault.}

Large language models (LLMs) have unlocked something traditional vertical SaaS never could: the ability to automate high-cost, language-intensive tasks that represent the vast majority of work in professional services. Lawyers reviewing discovery documents. Doctors writing clinical notes. Accountants preparing tax filings. These workflows were always too complex and nuanced for traditional software to handle. But LLMs have radically expanded what's possible.

The numbers paint a clear picture. Business and professional services, which are predicated on exactly these kinds of repetitive language tasks, account for 13% of the US’s GDP. That's roughly 10x the size of the software market. Vertical AI isn't competing for IT budgets; it's competing for labor budgets. Unlike vertical SaaS, which typically captures a fraction of Fortune 500 IT spend, Vertical AI taps directly into the labor line of a P&L. That's why we believe Vertical AI represents a fundamentally larger opportunity than vertical SaaS ever did.

The momentum is undeniable. Growing giants such as Abridge (transforming clinical documentation), DroneDeploy (drones and robotics for construction sites), EliseAI (AI for housing and healthcare), EvenUp (automating personal injury law), and Fieldguide (reimagining audit workflows), Legora (AI for corporate law), and MaintainX (digitizes maintenance operations for manufacturing) are demonstrating what's possible when AI serves previously untouchable workflows.

In a previous report, we predicted that Vertical AI companies would hit $100M+ ARR within historic timeframes. We’re already seeing this level of momentum from many AI supernovas, and we anticipate the first vertical AI IPO to likely happen within the next few years.

This early stage playbook offers founders principles and frameworks for building defensible vertical AI products. We’ll also offer guidance for teams on how to select the right business model, create competitive moats, and prove ROI quickly in your early stage journey of transforming language-intensive workflows in underserved industries.

The “Good, Better, Best” framework

	Functional value	Economic value	Competitive dynamics	Defensibility
Good	Introduces new feature that demos well	Productivity boost	SaaS incumbent or weak AI upstarts	Execution speed as a moat
Better	Introduces new feature with hard ROI, in particular, for a core business workflow	OpEx cost reduction	Adjacent competitor or sleepy incumbent	Complex product with early/weak data moat
Best	Tackles an end-to-end workflow with "LLM magic" not previously possible in absence of LLMs	Revenue boost or productivity improving for high value labor	No legitimate modern incumbent	Complex product with true moat around data and/or best-in-class multimodality

Jump here to dive into our Vertical AI recommended reading for founders, including our four-part series and our related roadmaps.

Getting started on building your Vertical AI product

The best vertical AI products don't start with a technology thesis—they start by solving a specific workflow problem in an industry desperate for better tools. Some of the greatest SaaS companies followed this pattern. Shopify, among many others, all started as internal solutions before spinning out to become category leaders.

Many startups are built by industry insiders who have deep experience in their vertical, but we have also witnessed the power of the outsider, which includes product and technological experts that solve entrenched industry challenges with GenAI and a fresh approach.

For example, Shivdev Rao, M.D. started his career as a cardiologist before founding the healthcare AI company Abridge and Jin Chang was an auditor before founding Fieldguide, the advisory and audit automation platform. And then there’s Max Junestrand of Legora who isn’t a lawyer, and yet has built the AI solution helping law firms collaborate better. Minna Song of EliseAI follows a similar profile; she saw the potential to build AI agents to improve the efficiency of healthcare and housing, despite never working in property management.

A Vertical AI founder’s “edge” doesn’t always come from industry experience on the resume, rather it’s the systems thinking, imagination, and deep customer empathy to find a vertical workflow or task that was previously not automatable before GenAI.

Four frameworks to help you select your product idea

When honing in on a product idea in Vertical AI, make sure to think through the following four considerations.

1. Clear ROI and technical feasibility

Assess potential product ideas along two dimensions: business impact and technical feasibility. The sweet spot is high-impact workflows that AI can reliably execute.

Look for:

Precisely-defined, repeatable processes your target customers have mastered. Newer or ambiguous processes should remain human-driven until standardized.
Reasonable expectation AI can execute safely without degrading customer experience or creating compliance or security risks.
Even better: workflows where AI introduces superhuman capabilities, such as analyzing 1000x more data, operating 24/7, detecting patterns invisible to humans.

Avoid:

Processes requiring constant judgment calls that haven't been codified.
High-risk workflows where mistakes compromise customer relationships or regulatory compliance.
"Mundane but easy" tasks that don't create meaningful value.

2. Insider vs. outsider advantage

Are you solving a problem you've lived through, or addressing an industry as an informed outsider?

Insiders move faster initially—they understand workflows intimately, speak the customer's language, and have immediate credibility. The risk: conformist thinking that blinds them to disruptive approaches.

Outsiders face steep learning curves that burn months and capital understanding nuances insiders grasp intuitively. Yet outsiders can win by bringing specific expertise in applying emerging technologies—if they marry that knowledge with deep empathy for customer problems. The failure mode is predictable: showing up with technology and assuming it'll work without obsessive attention to workflow integration. Successful outsiders don't just solve the problem technically; they understand how their solution fits seamlessly into daily operations, regulatory requirements, and existing systems.

Vertical AI typically rewards insider expertise more than horizontal SaaS did—you're reimagining complex, nuanced workflows in regulated industries, not just digitizing generic processes. That said, the best approach often combines both outsider creativity with insider networks for rapid validation. As we shared above, breakout Vertical AI leaders come from all backgrounds.

Here’s what matters most: Intimate understanding of your industry's challenges, clarity on which workflows drive real ROI for customers, and a specific initial wedge you can dominate.

3. Find your "magical feature" and earn the right to expand

The best vertical AI products don't just automate existing workflows—they demonstrate a miracle-like advancement in how work gets done today. Your entry point matters, but what matters more is using that initial breakthrough to earn the right to expand into increasingly core workflows.

There are three strategic entry points, each with different risks and requirements:

a. Tertiary workflows (the safer wedge)

Starting with ancillary tasks—legal research, clinical documentation, construction estimates—offers clear advantages: less resistance to change, faster sales cycles, and obvious ROI that frees up capacity for higher-value work. Abridge wins because doctors desperately want administrative burden eliminated, creating an easy entry point.

But here's the critical insight: if you stay at the periphery too long, you become vulnerable to disintermediation. You must use that initial “magical experience” to build a right to expand into core workflows. Show such dramatic improvement in research or documentation that customers trust you to tackle brief writing or diagnostic support. The tertiary entry is a wedge, not a destination.

b. Adjacent-to-core features (the balanced approach)

If your breakthrough sits closer to core workflows—automating demand package generation for lawyers, underwriting analysis for insurers—you can still sell it as a standalone tool, but it needs sufficient value that users will tolerate a new solution in their critical path. Sixfold succeeds here by embedding directly into systems, making the integration seamless rather than disruptive.

The advantage is that you're already proximate to the core workflow, making expansion more natural. The challenge is that you need stronger proof of reliability since mistakes impact business-critical operations. Early customers require more convincing, but once proven, your moat is deeper.

c. Core workflows (the Toast playbook)

Going after the "heart/lung machine"—the absolute center of how customers operate—can work, but the miracle must be undeniable. Toast didn't improve restaurant payment processing incrementally; they reimagined the entire point-of-sale system when competitors were still using outdated hardware.

This approach demands missionary zeal from your team. Your first customers take a genuine leap of faith, and they'll only do that if you demonstrate you're on a mission to transform their lives, not just sell software. The Toast team sleeping at customer restaurants, committing code on laptops behind the counter—that level of commitment convinced early adopters to bet their operations on an unproven solution.

Realistically, most founders should start with tertiary or adjacent-to-core “miracles” and expand from there. But regardless of entry point, the principle holds: demonstrate something miraculous, then use that credibility to expand before competitors catch up.

Keep in mind, speed matters more than ever compared to SaaS.

Traditional startup wisdom says "stay focused on one product until fully established." That made sense when each new feature required months of development. AI changes the equation. Once your initial miracle proves reliable, immediately begin planning adjacent expansions—your "Second Act."

This layer-cake approach (à la Procore, Toast, ServiceTitan) builds switching costs and deepens integration before competitors respond. Model capabilities evolve weekly, and the window to establish category leadership is measured in quarters, not years. Use your initial product to earn the right to expand, then move fast.

4. Progressive delegation vs. complete replacement

AI doesn't need to fully replace a workflow to be valuable. In fact, making complete automation the goal can be detrimental—it creates unrealistic expectations that kill momentum when reality falls short.

Instead, aim for "progressive delegation." Start by automating a manageable chunk of the workflow (the most time-intensive, lowest-value slice.) Let humans handle the rest initially. As you learn from production usage, gradually automate additional steps.

This approach has multiple benefits:

Ships faster with tighter feedback loops
Reduces risk by keeping humans in the loop for high-stakes decisions
Allows customers to standardize remaining manual steps, making them automation-ready later
Creates expansion revenue opportunities as you automate more of the workflow

Here’s an example: An AI audit tool might start by automating evidence gathering and preliminary analysis, while auditors handle final judgment calls and client communication. Over time, as the system proves reliable, it can take on more of the review process—but the initial product delivers immediate value without requiring perfect automation.

Selecting high-impact use cases: The three-part test

Before committing to a product idea, ensure it passes three tests:

1. Does your product idea provide enablement ROI?

Does your automation unlock entirely new capabilities customers couldn't do before? This is more valuable than pure productivity gains.

Enablement: Custom demos without engineering time, analysis of datasets too large for humans, coaching based on 1000x more conversations
Cost savings: Offset future hiring, prevent expensive turnover (e.g. Abridge saves millions in physician retention costs)
Productivity gains: Save time on tasks (only valuable if reinvested into strategic work)

Enablement ROI often matters most for enterprise sales—it's a capability unlock, not just an efficiency play.

2. Does your product target processes that are standardized and repeatable?

Look for processes your target customers execute regularly, the same way every time, with consistently positive results. If the process requires constant improvisation or judgment calls that haven't been codified, it's not ready for automation.

Test this by documenting the workflow in extreme detail. If you find yourself writing "it depends on..." frequently, the process needs more standardization before AI can handle it reliably.

3. Does the use case have narrow scope and low risk (initially)?

Start with automations where mistakes are recoverable and don't compromise customer relationships, privacy, security, fairness, or compliance. Even if you believe risks are low, wait until you have production experience before tackling high-stakes workflows.

Once you've proven reliability with lower-risk use cases, expand to higher-value (and higher-risk) workflows with confidence and customer trust already established.

10 principles for Vertical AI founders to follow

We see business defensibility comes from understanding workflows deeply enough to automate them reliably, integrating tightly into existing systems, and pricing for the value you create. (Revisit Part IV of our series for a deeper dive into each principle.)

Customer-centric automation: Build solutions only where automation aligns with customer needs and context, not just possibility.
Avoid commoditized features: Focus on differentiated, integrated workflows rather than features that competitors can easily replicate.
Leverage AI for superhuman tasks: Identify and implement AI in areas where it can operate at scales or speeds unattainable by humans.
Quantifiable ROI unlocks value: Demonstrate clear revenue gains or cost reductions to drive adoption and loyalty.
Innovate on business models: Embrace new delivery methods and pricing enabled by AI automation to access broader market segments and expand margins.
Target niche and underserved markets: Initial competitive advantage often lies in overlooked, high-ROI areas.
Customize for nuanced requirements: Serve complex buyer needs such as compliance or security to erect defensible barriers.
Technical moat comes from multimodality: Competitive edge increasingly depends on combining data types and workflow integrations, not proprietary models alone.
Build modular and adaptable systems: Ensure tech infrastructure can flexibly incorporate the best models as AI evolves.
Prioritize data quality over quantity: Early success depends on high-quality, relevant data, which compounds in value as the business scales

A call for the next generation of Vertical AI founders

Tomorrow's titans are being built today—but countless others are still ideas waiting to be imagined. If you’re an aspiring founder working in an industry in desperate need of revitalization, start by automating your own workflows in your everyday work. If you can make solutions work for yourself, you've validated your first product.

If you’re already working on a Vertical AI application, we would love to hear from you. Email our team at VerticalAI@bvp.com.

Our growing Vertical AI roadmap library

Beyond our foundational Vertical AI perspectives, we’ve released investment strategies across developer platforms, IT services, construction and real estate, and beyond. Explore all the emerging investment roadmaps across our Vertical AI library.

Research and roadmap	Vertical / Service	Summary
Built World AI	Construction and real estate	The built world drives nearly a quarter of U.S. GDP and yet remains one of the least digitized sectors. Multimodal AI is enabling the next wave of innovation in how we design, build, and operate it.
Voice AI	Voice application across all industries	How advancements across the voice technology stack enable human-like conversations, with applications in transcription, inbound sales, customer support, and outbound calling across multiple verticals.
Reinventing India's IT sector with AI services	IT services	How AI is disrupting India's $264 billion IT services industry through three new business models: AI-enabled services, services built for AI (data infrastructure and LLMOps), and pure software platforms.
Public safety AI	Emergency response	The ways AI is transforming emergency services and public safety through automated non-emergency call handling, dispatcher training, mental health support for first responders, and real-time incident detection.

This playbook was created in collaboration with Kent Bennett, Sameer Dholakia, and Brian Feinstein. The research was primarily derived from the Vertical AI series which was co-developed and written by Sameer Dholakia, Caty Rea, Alex Yuditski, Brian Feinstein, Byron Deeter, Kent Bennett, Mike Droesch, Maha Malik, Sam Bondy, and Aia Sarycheva.