The Long View

The Race to Become the System of Action

In vertical SaaS, the next battleground is the system of action—where AI doesn’t just help run the business, it does the work. The winner gets the user, the workflow, and the economics.

By Dave Yuan

Founder and Partner, Tidemark

“AI is eating software” is a narrative that moves $100Bs of private investment and huge swings in the public SaaS market.

We've been on a journey to understand the impact of AI on vertical SaaS for the past three years (2023, 2024). In particular, we wanted to know how control point incumbents would compete with native AI challengers. We now have an answer: the battle will be won by a race to become the system of action.

A system of action is the next step beyond the system of record, where humans, AI-assisted humans, and fully autonomous AI agents are able to act on data and trigger downstream workflows. In doing so, they subsume the control point and gain the high ground. 

Historically, software helped a business run. It automated front-and back-office functions that weren’t the core activity of the firm itself. Generative AI is opening a new frontier—doing the work itself. Legal tech, for example, moves from providing tools that help owners manage their legal practice to tools (and automation) that actually practice the law itself. This means that the employees who are crucial to delivering the product of the firm are suddenly capable of far more. Control points who embrace AI have an opportunity to do much more for their customers.

“Doing the work” also creates an opening for native AI challengers to gain a foothold and “integrate & surround.” In industries where there are “Hero users” with strong agency to choose their own tools, “doing the work” can shift the locus of control. Software companies are now able to sell directly to these Hero users, which has profound implications on product, go-to-market, pricing, and business models. Control point incumbents that are slow to adapt open the door for a native AI challenger to win large pools of the control point’s Hero users and the strategic high ground. With these Hero users, the native AI can force integration into the control point’s data and workflows. The native AI player then benefits from “All the Data”, triggers downstream actions and becomes the system of action. 

Today, we’ll explore how both an incumbent control point and a native AI challenger should run this race using a hypothetical case study in veterinary practice management software.

This is a long one, so buckle up!

The Theoretical Foundation

You can dig into our many preceding essays on vertical SaaS and AI, but the short and sweet version is this: control points, as a function of their workflow and data gravity, inherently enjoy an unfair right to sell multiple products to their merchant customers, even potentially becoming an "operating system" within their vertical. This remains true not only for traditional software products but also for new AI offerings. 

Native AI players may be more effective at pursuing full substitution strategies—eliminating the need for a human by automating tasks completely and, in doing so, bypassing the control point software. But full substitution in a small business environment is rare as each worker touches many tasks—some in explicit workflows, most in implicit ones—and merchants are generally resistant due to quality control and trust. 

If control points aren’t asleep at the wheel, they can leverage the $100Bs of venture capital poured into the infrastructure players and provide AI products that rival native AI players. 

Congrats! You just skipped having to read about 4 of our previous essays.

What’s new is a slowly building paradigm shift of AI allowing software companies to expand scope from running the business to doing the work itself.

Run the Business vs. Do the Work 

Forever ago, when dinosaurs roamed the earth, vertical software was sometimes called “business management software.” Today, it’s sometimes called “property management software” or “practice management software.” All of these names point to the same thing: vertical SaaS has historically focused on running and managing a business, not doing the actual work of the business itself. 

From our earlier example, legal practice management systems are focused on providing tools to run a law practice—onboarding clients, managing the workflow and tasks of the various employees, sending out the invoices to clients, and collecting payments. What they generally don’t do is help lawyers actually practice law.

The goal of vertical SaaS generally is to provide workflow software to owners and managers to operate the business consistently, efficiently, and profitably. And, because they sell to owners, they haven’t always needed delightful user experiences to be successful.

Former Benchmark partner Sarah Tavel’s advocacy around “selling the work, not software” recognizes the new capabilities of AI to narrow the gap between empowerment and outcomes. These new capabilities are being used to blur the line between running the business and doing the work itself. In the following example, transcription tools start by automating the administrative task of documenting and summarizing patient notes. But, thereafter, it’s not hard to imagine triggering administrative tasks like follow-up scheduling, invoicing, and even clinical recommendations.

In verticals where the practitioners are influential, this provides an opening for the native AI player. They can target Hero users with tools that actually do more of the work. Hero users are those who:

  • Are highly influential and valued because they are scarce or add unique value.
  • Are solo contributors or work independently. If they aren’t heavily matrixed or ensconced in multi-person workflows, they can quickly pick up new tools.
  • Have agency within the company to choose and buy tools independently.

If a merchant has employees like this, a native AI can offer them tools that makes their lives easier. The best Hero users to target do digitally native work (vs. physical labor) that involves lots of manual drudgery—data entry, reconciliation, summarization, tagging—that AI can eliminate. If the experience is “magical” enough, the native AI can quickly acquire these Hero users.

If you own the Hero user, you have a head start on becoming the system of action.

The Race to the System of Action

To understand what system of action even means, you have to start with the fundamentals. 

Software applications take in data—either via human or system input. They analyze that data, present it to humans to inform decisions (typically via dashboards or apps), and/or trigger actions—either by humans or other applications.

The goal isn’t to replace humans outright. It’s to initially empower them through agents. When done in a digital workflow with both learning loops—workflows where the AI can learn and improve over time and a human remains in the loop for quality control—more and more of these tasks can move toward higher levels of automation. Not just to reduce headcount (although that’s often a natural byproduct), but to make the experience more effective, more efficient, and more magical

If you’re not the system where this progression of work—from human action, to human + agent, to full AI agent—work is done, then you’ve already lost the race! The system where this work is done has the right to pull in all the data (usurping data gravity), and to drive downstream workflows and actions (stealing workflow gravity).

h/t Abraham Thomas “Data and Defensibility

Who owns the system of action? 

Well the system of record (i.e. the control point) is the logical incumbent. But if a native AI player is able to win the hearts and minds of the Hero users, they can flip the script and become the system where downstream actions are decided and triggered. They can usurp the system of record’s data gravity and become the system of action.

Let’s bring this to life with a case study.

The Death of a Control Point: A Hypothetical Case Study

Consider the veterinary clinic software market. The control point is typically the practice management system (PMS)—a historically powerful control point, due to its strong workflow and data gravity. The workflow gravity comes from communications with clients, scheduling, and billing. While the data gravity stems from the electronic health record (EHR) where notes and patient records are stored. A practice management suite can serve as both the front- and back-office control point. Done well, it is a true operating system that captures the vast majority of a merchant’s software spend.

Into this stable market emerge these seemingly flimsy AI transcription tools. Initially dismissed as wrappers built atop platforms like OpenAI—with none of the complex machine learning capabilities or sophisticated infrastructure traditionally respected by incumbents—these tools nonetheless address a painful gap: veterinarians' struggles with documentation.

Veterinarians, already scarce and overworked, usually have hours of notes to transcribe after dinner. Burnout is a real issue. A tool that can automatically transcribe session notes is transformative and gives them their lives back. These native AI solutions offer a magical, visceral experience: veterinarians previously burdened by hours of post-session documentation can now finish their notes during sessions, reclaiming precious personal time.

Despite clear market demand, the incumbents hesitate—choosing to "gold plate" their offerings by waiting until transcription can be flawlessly integrated into their EHR, calendaring and billing workflows. There also is a huge learning curve for the team, which has a great workflow for control and compliance, but not for a solo user experience. Then, it’s tough to get the sales team to pay attention. Sales reps don't know how to sell it, and low price points mean low commissions. Add to that a marketing team with little instinct for PLG—and fear that a PLG funnel will cannibalize their traditional demo-request motion—and the result is a perfect storm of friction. Nothing gets done. In prioritizing perfection, founders ignore the immediate needs of users who simply want relief from documentation burdens.

Meanwhile, the native AI startup capitalizes on the incumbent’s delay. Sure, their product is “just a wrapper”, but the user experience is great, so who cares? By wrapping the wrapper (heh!) with vertical-specific language, templates, a fast and easy UI, and great prompts, they're able to make the product feel tailor-made for veterinarians. Leveraging a PLG strategy, they offer an easy-to-try solution designed specifically for veterinary professionals. Veterinarians quickly adopt the solution, appreciating its immediate and tangible benefits, which prompts swift market penetration. With learning loops, the product improves quickly, attracting more users. Buoyed by rapid user adoption, the AI startup raises substantial venture funding, further accelerating its growth.

As its user base expands, the startup begins to force integrations into the existing PMS—initially via tactical methods like Chrome extensions and robotic process automation (RPA). They may even create a virtual user in the PMS that mimics other human users to control the system. Eventually, the sheer volume of user demand compels incumbent providers to formally integrate the AI transcription tool—effectively ceding ground.

With formal integrations secured, the native AI provider begins to expand its product capabilities, incrementally taking control of downstream workflows—scheduling, billing, and billing & payments—previously dominated by the incumbent. They leverage the data now accessed from the incumbent’s system of record to launch even more AI-driven features. Transcribed notes get linked to billing codes, significantly streamlining the billing process. They introduce their own scheduling system and begin extracting and managing calendar data from the PMS.

They may even provide an alternative billing & payments workflow which is 3x the PMS monetization. With a lower cost distribution model and higher ARPU (and a lot of venture money), they take it to the control point and start to offer big parts of PMS functionality for lower cost or free. 

Then? It’s game over. The incumbent is done. They just don’t know it yet. 

What can we learn from this hypothetical case study? What’s really going on here? 

Let’s look at it from both the native AI and incumbent SaaS perspective.

Native AI Strategy Abstracted

In the past, we talked about picking your market based on TAM and fragmentation. Now, you want to add another criterion to the list: a slow-acting incumbent. The good news? This is much of the world. Thoma Bravo recently shared that SaaS is only 25% of $5.9 trillion in software spend (outweighed by on-prem and hosted). That’s a lot of old companies for AI founders to devour.

Once you select your market, you focus on the Hero user and the work that they do. You want to solve their most compelling problem (which may not always align with what’s most compelling to the business owner). Help the Hero do the “Hero’s work” better and automate the “Administrative work” from their life. (We’ll publish a subsequent essay talking through scenarios where the Hero user doesn’t have sufficient agency.)

Ideally, this work sits at the start of an important workflow and is key to decisioning or actioning the next step. This gives the native AI a right to be the system of the action and an opportunity to expand its offering over time.

Three Types of Work

There are generally three types of work that a native AI should consider:

  • Hero’s work: What a merchant prides themselves on doing. It's why they exist.  Oftentimes the people who do it are the most celebrated. For an accountant or lawyers, it’s advising clients. For a medical professional, it’s delivering care.   
  • Administrative work: The work you have to do. Scheduling, reconciling, invoicing. You want to do it as quickly as possible. As a result, it’s probably not done that well or that efficiently.
  • Work not done: This is valuable work that doesn’t get done—either because the merchant lacks the time, skills, context, or resources. Often, this gets outsourced or ignored. Alternatively, this can be work that is valuable for the merchant but not exciting for the worker. Examples include:
    • Picking up the phone during lunch hour
    • Updating the website or social media account
    • Filing compliance statements
    • Running multiple bids on procurement to ensure the lowest cost (vs. calling your buddy vendor rep)

Then comes the hard part—making magic. You are past the science portion of the work. Now, it’s time for the art. You need to deliver a magical experience for the user—even if the tech behind it isn’t groundbreaking. Wrappers are totally ok! Ideally, the product has strong learning loops. It may be a wrapper over some other player’s infrastructure, but it gets better with usage. It learns from user input on correct and incorrect outputs. It may observe user actions or workflows, detecting informal but important patterns and context. The more the product is used, the more it learns and improves to stay ahead of incumbents and other entrants.

Your GTM playbook is more likely classic PLG than sales-driven. Your products should be easy to try, buy, and find:

  1. Tryable: Free trials or standalone offerings that have no need for integrations with a simple, intuitive UI.
  2. Buyable: Initial price point at the user requisition level such that they can just swipe their credit card without a second thought. Afterwards, you scale the deal size with usage and value. Think of it like Slack—you go bottoms up from practitioners versus top down from CEOs. After they buy it, there should be instant ROI.
  3. Findable: PLG sales motions help, but we’ve seen product wedges go viral via strong content marketing, referral flywheels, and SEO. 

Once you have acquired a meaningful community of these customers and they are finding daily delight in your wedge, you can “integrate and surround” the incumbent’s control point. Your goal is to integrate into the system of record. You’ll start with scrappy tactics: copy/paste, Chrome extensions, RPA. But when you reach scale, have your users demand integration for you. In our case study, that would be the synching of transcription with the EHR notes. From there, you start triggering downstream actions in workflows. 

This is the moment you start beating down the incumbent and getting aggressive. 

Pull in or replicate core functionality from the system of record into your own product. Then look for adjacent or downstream expansions to add ARPU fuel. Or, because customers already trust you for all things “magic AI,” you create alternate high-value/high-monetization workflows to increase your economics. Look for 5x + ARPU opportunities like claims management and insurance billing. (For the full range of product ideas, check out our Benchmark study.) You then use that revenue fuel to expand the product, field a sales team, and lower pricing if the control point tries to bundle in a competing offer. 

This is a good story and one that many investors are deploying capital for. Here is how incumbents can fight back. 

Incumbent Defensive Strategy

Native AIs are coming. While they may be hidden from view in incubators and Silicon Valley garages, you need to watch your back!

Protect your flanks to fight off the “integrate and surround” strategy. Who are the Hero users in your merchants? What key systems of record do they work with and rely on? In our case study, it was the electronic health record system and the calendar.  In other industries, look for systems where merchants bring in revenue and engage with customers—e.g online ordering, channel management, or e-commerce. (See our discussion on these key systems in our consumer extension essay.)

Once you’ve identified these systems, lock them down. Ensure all your customers are using your version. Consider making them free through bundling or freemium models. Do your best Jrue Holiday impression and play lockdown defense—making integration tough on all challengers. No API access, no MCP integration. Block all the little hacks and tricks, even if they seem initially harmless. If you’ve already opened these up, you’ll need to have some uncomfortable API and partnership discussions.

Think through the work your heroes are doing. You are looking for the “Hero’s work” you can empower, and the administrative work you can automate. Look to your user groups or survey your base for clues—many fast-scaling tools are addressing these pain points, even if they seem outside your traditional scope. You need to expand your scope beyond running the business, to understanding how your software can help its merchants do the work.

Build Products for users not owners. Learn from your native AI competitors. Hero users want tools that are easy to try, easy to buy. An empowered Hero should be able to download your product, test it, validate that it works, and purchase—all within five minutes. That means deprioritizing sales enablement bells and whistles like demos, onboarding calls, and SSO (at least for now).

The goal is to build products that delight Heroes, not win Noble Prizes (h/t my fellow board member Kent Bennett). Yes, we love technical maturity models. But your goal is to build a product that delivers value for your users, not win the local science fair. Wrappers are fine!  Concepts should be able to be prototyped quickly.

Great is the enemy of good. You have amazing data. You have powerful workflows. Great. Get over it! You need to ship something, NOW! It may be just a “wrapper” on public infrastructure, but as long as it adds value to your Heroes, you are capturing the high ground. You can show the roadmap of how this all comes together in incredible ways in the future, but launch something today.

Find Heroes where they are. Sell to them how they want to buy. You may be a sales-led company, but it’s now time to learn PLG.  

Engagement first, money later. Success = engagement, not monetization. We love monetization layer cakes, but you’re looking to win influential users and engagement as quickly as possible. Measure DAUs for daily work and WAUs for weekly work. And let your CFO know: you’re okay to try bundling or even going free (gasp) to ensure long-term victory. You can always add paid tiers with advanced functionality later.

If you’ve made it this far, congrats! You’ve likely headed off that pesky native AI at the pass.

Of course, while you’re doing all of this, lay the foundation for the agentic age:

  • Infrastructural build: Design your infrastructure for modularity and adaptability to enable rapid experimentation and easy integration of evolving AI capabilities.
  • Reconsider different pricing models: Transition pricing strategies from static seats to value-driven usage metrics, aligning your revenue with the tangible outcomes AI delivers.
  • AI skills and cultural enablement: Foster a culture of trust, transparency, and user control by embedding clear AI governance and explainability directly into your products.
  • GTM enablement: Implement product-led growth strategies that empower influential users ("Heroes") to discover, trial, and advocate for your AI-enhanced tools organically.

(More coming soon from Tidemark on the AI opportunity for vSaaS. Sign up here to be notified when it launches.)

Ultimately, success hinges on who secures the system of action first—embedding deeply into daily operations and critical workflows. Whether incumbent or native AI, controlling user engagement and workflow integration will determine who dominates the AI era.

Ending Note: The Brave New Agentic World

Whatever your origin, native AI or incumbent control point, if you own the system of action, you unlock a brave new world—transforming how you serve your customers, your TAM, and your overall economics.

You’re no longer just helping merchants run and manage their businesses. You’re helping them do the work itself.

There’s so much you can do:

  • Empowerment: Help Heroes solve high-ROI problems—like bringing in more revenue or finding key talent.
  • Assistant: Take tasks off their plate such as:
    • Answer the phones during the lunch rush
    • Schedule appointments
    • Call to remind customers about appointments
    • Pay bills, payroll, and taxes
    • Reconcile and report
    • Train staff
  • AI Agency: Handle work they can’t do on their own and typically outsource at a high-cost: 
    • Revenue managers help hotels determine their pricing based on capacity and channel (can sometimes take 5% of revenue)
    • Ad agencies will often take 10% of ad spend just for the privilege of spending your marketing dollars
    • Buy supplies at the best price (Group Purchasing Organizations will sometimes take up to 5% of spend for the privilege of spending your inventory budget)

We’re excited for this future and actively looking to invest in systems of action regardless of whether you started out as an incumbent or native AI.

Authors:
No items found.