Emerging Playbooks for the Vertical AI Era
Over the last few years, vertical SaaS has evolved into vertical AI and the approaches founders are taking towards building companies have shifted accordingly. As a firm, we have been backing vertical software companies for over a decade; however, 2025 is a particularly exciting time for the category as vertical founders experiment with bold new business models and pricing structures in an effort to best apply AI to their industry. Here are four of the business models we are seeing vertical AI startups gravitate towards as they contemplate what the next chapter of vertical software will look like. We are open to companies building under any of these frameworks, and the approach a company takes will largely depend on the type of customer they want to reach. This list isn’t meant to be exhaustive but think of it as a snapshot of what we are seeing as investors in today’s market.
1. The AI-Native Business-in-a-Box
In the cloud era (roughly 2013-2022), we saw many business-in-a-box vertical SaaS startups get funded. They typically ran the Bessemer Layer Cake playbook and offered a one-stop, browser-based interface for communicating with customers, managing employees, accepting payments, running basic marketing campaigns, and so on. While seemingly a great idea, many of these startups struggled to charge their SMB customers meaningful ACVs. And the unit economics associated with rapidly acquiring thousands of low ACV customers were another continual challenge. In retrospect, the math for most of them did not really support a venture outcome. Those that succeeded either went up market (e.g., ServiceTitan) or they found a way to monetize on something other than SaaS - usually this meant payments (e.g., Toast). Monetizing on something indirect like payments helped Toast and others bypass SMBs’ limited willingness to pay up for software subscription fees, when they could always opt out of software and simply run their business without SaaS (as they had been doing for years).
Today’s AI-native business-in-a-box startups have found success by offering the same basic toolkit but augmenting it with AI to take on more of the labor done by their customers. The goal here is to use labor augmentation as an alternative way to drive up ACVs for business-in-a-box offerings. One common early-stage vertical AI playbook we see looks like this:
Step 1: Build a basic CRUD app following the layer cake model. With the advent of AI coding assistants, rebuilding basic business-in-a-box apps can be done quickly and at little cost.
Step 2: Use AI to automate - and fully manage - some burdensome administrative task on behalf of your customer. Examples I’ve seen using this approach include submitting forms to private insurers or Medicaid to get reimbursed, managing the day-to-day scheduling of low-cost services (e.g., nail salons), or automating marketing and lead generation.
Step 3: Finally, instead of a SaaS fee, try to monetize by charging a percentage of topline under the argument that the ISV only wins when the customer wins. Founders often frame this fee as covering the software, as well as whatever managed service they are using AI to offer. From what I have seen, these startups are typically aiming for take rates on topline revenue in the 5-10% range.
Founders typically position their AI-native solutions as replacing the role of 1-2 FTEs at the customer and try to comp their effective pricing accordingly. One caveat - take rate models can be tricky when a customer’s topline grows to the point where the customer no longer feels like they are getting a fair deal in terms of fees paid (% of revenue) vs. services rendered. Customer comfort with this model is still being tested, but it appears to be a way to quietly back into higher effective ACVs. Time will tell what long-term take rates look like at scale for business-in-a-box players charging against topline; I would not be shocked if many eventually get pushed towards more traditional pricing over time; however, by adding more value to buyers they should see some ACV uplift vs. the v1.0 of these companies.
Examples: Alpaca Health is an early-stage company using this approach - they use AI to automate insurance submittals and the paperwork involved for ABA practitioners. Doorstead is another - they offer an AI tenant placement and property management suite and charge a percentage of the total rent for the units it manages.
2. Workflow Automation -> Enterprise Platform
Another vertical AI approach we are seeing is for startups to approach large enterprises and try to productize the ability to automate some critical but annoying task with AI - this is usually something in the back-office or middle-office. Over time, these companies will add on new modules and try to become the de facto platform for their given category. These companies are typically aiming to land with mid-market/enterprise buyers at ACVs of $50k-$200k while still a Pre-Seed/Seed stage company. This will usually require taking on a pilot-driven GTM approach. Under this company building motion, early customer relationships need to be highly consultative, and require the founder (who is also leading sales) to be actively reshaping the product based on customer feedback. Here, the goal is to drive adoption of the initial automation solution while leaving room to build complementary modules to grow ACV over time.
We are seeing lots of these workflow automation players cropping up in insurance, loan origination, and other process and paperwork-heavy industries which already run off existing checklists for tasks like underwriting and loan approvals. For vertical AI founders targeting enterprise customers, having one killer automation is a great entry point which wins you the right to build a much larger offering - something that’s easier to do once you have access to your customers’ internal datasets to expand and improve upon your product.
Examples: Sixfold (automating insurance underwriting); Anterior (automating prior approval submittals); Mandolin (specialty drug insurance verification); Casca (SMB lending platform).
3. Forward-Deployed Dev Shop -> Enterprise Platform
One of the more innovative approaches in software in the post-AI world is for companies to use the decreasing cost of code to build truly custom software. Over time, these startups try to generate some internal technical asset, which enables them to offer “custom” software quickly and at scale (e.g., Palantir’s Foundry product). At Pre-Seed/Seed, many of the initial customer relationships start off looking like a consulting engagement focused on automating the workflow challenges faced by a large corporation. The work these startups do for the F1000 can include spinning up customer-facing or employee-facing tools, enabling data capture and transfer across the organization, or creating internal risk-scoring and decision-making systems. The rise of these custom applications has been a boon to consulting companies who have charged huge fees to build basic GPT wrappers (e.g., Accenture). Now, startups are entering the picture and can offer far more performative solutions on faster timelines. This custom software model generally requires a “forward-deployed” sales approach to ensure you are building for that particular customer’s needs, and that you can access their internal data for model training and fine-tuning.
While many of these companies look more horizontal (e.g., Distyl), they are vertical in the sense that they are competing against vertical AI entrants for the same industry budgets and what they ultimately build often ends up effectively being vertical software. Instead of buying the industry leading vSaaS platform, these F500s may turn to companies like Distyl or OAI to build custom solutions for their specific needs. Today, Distyl supports customers in a number of industries including telecom, healthcare, manufacturing, insurance, retail, and CPG. These engagements compete with vertical software providers but are fully personalized instead of just being designed for a given sector. Some examples of vertical solutions Distyl lists on its website include building custom apps for a F100 healthcare payor to manage their prior authorization approval process, generating AI insights to support a major insurer with fraud investigations, and building a supply chain AI assistant for a top CPG brand. Palantir, the company where Distyl’s founders previously worked, also does similar automation work for its corporate clients although with a more rigid framework - they also tend to have a much heavier government focus.
My prediction is some percentage of revenues which would have previously gone to a vertical software vendor will be captured by these custom workflow shops - particularly in the F500. If you are a founder in this category, I am particularly interested in what you are doing as I think this is an underdeveloped space in terms of VC backing.
Examples: Distyl (raised $20MM Series A in Q4 2024); OpenAI is also reportedly flirting with this model. Watch out for more on this new approach to software startups in an upcoming edition of our blog.
4. Voice AI Wedge -> Modern System of Record
Advances in voice AI have created openings to get historically difficult user types to engage with software by making interacton dramatically easier and meeting users where they are. Voice AI is great for capturing customer interactions and eliminating burdensome record-keeping. One example of startups relying on voice as a wedge can be seen in the medical scribe companies expanding into creating supporting documentation for medical billing. Startups are also using voice AI to solve data entry challenges for on-the-go field sellers - this enables CRM data to be captured that would previously be totally lost or only partially entered hours or days after the customer conversation took place. The goal of these entrants is to get in with AI and use that as a wedge from which to build. For categories with savvy vertical incumbents, they will likely bolt on their own voice AI offering, but for sectors with dated or non-existent software providers, voice AI can be a very strong entry point.
Another approach we see these voice AI companies take is to start off as the System of Interaction while sitting on top of the existing tech stack; then, once trust is established, they will aim to become the System of Interaction and the System of Record, ultimately making a pitch to rip out the underlying incumbent as they can offer a better version of the same product. A few weeks ago, Toma announced a $17MM Series A led by a16z. They started off using voice agents to manage inbound calls for car dealerships; now they are building out a full suite of AI agents to manage all kinds of operational tasks that car dealerships face. This is a great example of using voice agents to solve labor challenges, gather data, and then expand from there to build a true operational platform for a sector.
Examples: Rilla (capturing customer interactions and acting like a Gong for field services sellers -> long-term, they have an opportunity to be the CRM for IRL); Abridge (using medical transcription as a wedge into the revenue cycle management category); Superdial (automates calls to insurers, longer term their vision is to be the AI-to-AI communication layer for medical billing and authorization).
Different Buyers, Different Models
As vertical AI matures, we are seeing adoption of different business models and go-to-market motions depending on the buyer type, their needs, and their willingness to pay. We are actively evaluating companies in each of the four categories outlined above. If you are building a vertical AI company - or even thinking about embarking on that journey - I’d love to hear from you. You can reach me at patrick.mcgovern@bowerycap.com or @pw_mcgovern on X.
If you liked “Emerging Playbooks for the Vertical AI Era” and want to read more content from the Bowery Capital Team, check out other relevant posts from the Bowery Capital Blog.
