We launched the Vertical Software Collective (VSC) with the sole purpose of connecting the best vertical operators. An invite-only organization, it serves ~100 members today, from exited CEOs to growth-stage VPs to pre-idea founders. Our goal with VSC remains quality over quantity, with a high ROI-to-time-investment for all participants. In response to demand from earlier-stage founders for a collaborative live environment for soundboarding, we began hosting periodic Vertical AI Roundtables last year.
While we believe trusted, closed-door conversations provide the most value to founders, in our first few roundtables, some poignant and consistent learnings emerged. So we thought it could be valuable to begin sharing those anonymized insights with our broader community. If these challenges or conclusions resonate with what you’re seeing/feeling in the market, we’d love your input!
Today’s “Founder Insights Report” summarizes our last Roundtable in April, in which we welcomed ten VSC members and friends in NYC to discuss general early-stage challenges in Vertical AI. All startup founders, our attendees represented diverse verticals including healthcare, supply chain, retail, construction, government, and insurance. The group reflected various stages of maturity from stealth to Series A, with a public AI operator rounding out the perspective.
Roundtable Takeaways
1. Challenge: Integration with Incumbent Systems of Record
Participants highlighted significant challenges in integrating AI solutions with legacy systems, especially in the healthcare and government sectors. Many legacy systems lack open APIs, and incumbents either lack technical capability or strategic willingness to integrate. Some participants suggested workarounds such as automation and "Computer Use" tools from providers like OpenAI. Some such strategies risk regulatory, compliance, or legal hurdles (e.g., HIPAA, PII, security, SaaS account usage stipulations). One participant noted, “It’s a painful road to integration, and sometimes brute-forcing solutions is necessary.” For the time being, several founders are simply exchanging regular flat files to avoid the complexity: “They dump an Excel file somewhere, and we pick that up, and then we dump a file back out.” Participants concluded that integration is often unavoidable to demonstrate clear customer value, but emphasized strategic prioritization—waiting for technical advancements in third-party solutions or platforms that handle integration could sometimes be optimal.
Note: We are writing an upcoming essay touching on integrations and data problems in Vertical AI now. If you have any input or insights to add, let us know!
2. Debate: Is It Necessary to Replace / Own the System of Record?
Participants debated whether startups ultimately need to become systems of record themselves. A consensus emerged that the necessity largely depends on industry specifics and customer size. One founder remarked, “Customers have explicitly said they’d switch if we built the full system.” Smaller businesses or niche markets might readily shift to new systems offering clear improvements, while enterprise or regulated sectors (like healthcare, finance, or construction) present significant switching costs and inertia. One founder lamented, “You will do it eventually, but you're gonna just [find a] painful road there.” Thus, participants emphasized carefully evaluating whether the system of record is essential strategically, particularly considering resource-intensive integration efforts required to replace incumbents.
In addition, the group discussed vertical expansion towards systems of record: "We're shifting from control point of data information to control point of transaction, to control point of capital. And then all of a sudden we can actually [execute] better than any other tool because we'll have all the data and all the interactions." Some founders questioned the future meaning of the system of record altogether: “We're selling outcomes and ultimately, the system of record is just the data that allows you to always have the moat to provide better outcomes than anybody else who's evolved.”
Learn More: We share more on our vision for Vertical AI system-of-record replacement in our recent essay, “Emerging Playbooks in Vertical AI.”
3. Insight: Benefit of Service Add-Ons to "Eat the Complement"
Several participants shared experiences using a services-oriented approach to penetrate markets. Offering high-touch services—even if initially manual—enabled startups to capture and own critical data, eventually using it to automate processes and become integral to customer workflows. As one founder stated, “We just do whatever the customer needs—human or automated—to eventually automate at scale.” Participants advocated targeting repetitive, error-prone, or financially significant services: “tapping into their existing workflows and helping make it easier, faster, without them necessarily knowing there's a lot going on in the backend… that’s the key.” Over time, these manual services serve as rich training grounds for automation, improving service efficiency and scalability while establishing trust and locking in customers.
Learn More: We discuss the concept of “eating the complement” (and other product tactics for Vertical AI founders) in our recent essay, “A Guide to Disrupting Legacy Incumbents.”
4. Insight: Importance of Vertical-Specific Customer Discovery
Founders emphasized the importance of deep industry knowledge, customer intimacy through extended discovery/design partnerships, and workflow-specific add-on features (to reduce implementation friction) as critical differentiators from generic, horizontal AI applications. One participant remarked, “If you’re not 10x better than existing solutions, customers won’t switch.” Participants highlighted iterative customer discovery processes to identify and solve specific industry challenges, underscoring the competitive edge gained from specialized, often tribal knowledge embedded in a particular vertical workflow.
Learn More: Get our framework for early-stage Vertical PMF navigation—including tips on customer discovery staging—in our recent essay, “Nobody Wants Your Product!.”
5. Other Discussion Points
Selling outcomes rather than positioning as “technology” significantly improves customer buy-in, particularly in traditional industries. “Customers don’t care if it’s AI—they just want results.”
Human-in-the-loop processes are critical interim solutions to ensure accuracy and customer trust while building data for future automation.
Participants recommended using data-driven decision-making to hyper-transparently demonstrate financial value, directly linking AI solutions to customer revenue (or, in some cases, cost savings).
Founders noted a practical approach to automation: first, solving simple but tedious processes to build user trust and familiarity before gradually addressing more complex tasks.
Future Considerations
How can startups effectively prioritize integration efforts versus waiting for advancements from third-party providers?
What metrics or frameworks best assess the need and timing to become a system of record?
Strategies for efficiently scaling high-touch manual services while maintaining customer satisfaction and trust.
Deep dive into specific industry regulations and compliance challenges when leveraging advanced AI tools like Computer Use or local agents.
Changing nature of investor receptivity to hybrid software-services business models or current vs. future service automation levels.
Thanks for reading Euclid Insights! If you know a founder who is thinking through Vertical AI product strategy, we’d love to help. Just reach out via LinkedIn, email, or here on Substack (via comments or the DM button below).