Part III: Business model invention in the AI era

Much like cloud computing transformed vertical industries over the past two decades, the human-like capabilities of LLMs — processing text, images, videos, voice, and code — are making it possible for Vertical AI to achieve what was previously unimaginable for sectors of the economy that only benefitted marginally from previous waves of software innovation. Vertical AI companies aren’t just streamlining workflows; they’re taking on vast, high-cost, language-heavy tasks that dominate industries like legal, healthcare, and professional services — sectors representing a 10x larger TAM than the software market itself. 

The scale of opportunity is unmatched, but the efficacy and future success of Vertical AI hinges on the strength of its business models. Delivering your product in the right format for your specific vertical alongside a pricing model that captures the value you’re creating for customers is critical for the short and long-term viability of your business. In this installment of our Vertical AI Roadmap, we dive into three new business models defining the AI era — Copilots, Agents, and AI-enabled Services — and explore the innovative use cases and functions vertical AI companies are addressing today.

Copilots

The first incarnation of AI-native business models that we saw emerge were copilots. Copilots are AI applications that sit side-by-side with users as they go through workflows. In this way, AI copilots supercharge employee productivity while keeping the human user in the center of the workflow. Because copilots aim to increase individual employees’ efficiency and efficacy, copilots are generally priced like traditional cloud software on a per-seat basis tied to the company’s headcount. 

This pricing model has already shown real momentum with public companies like Microsoft, Google, and Salesforce which have all been able to benefit from healthy price increases of their core products by offering copilot add-ons. For example, Microsoft’s Office365 generally costs $15 to $30 per license, but getting access to their copilot product costs an additional $30, doubling or even tripling the price per seat.

Copilot use cases by modality 

Code: Github Copilot was one of the first widely adopted copilot applications in the market. Developers use Copilot to complete code, generate sample code, and ask questions, leading to productivity increases (reportedly as much as 55%) and higher developer job satisfaction. Our portfolio company Supermaven, which recently joined forces with Cursor, is another example. Its one million token context window understands a developer's code base and offers the best contextual code completion at lower latency, turning users into 10x more effective developers.  

Text: Many text-based copilots assist or complete tasks for users within document-intensive workflows. For example, the AI copilot Harvey makes it easy for lawyers to quickly summarize contracts and get immediate answers to questions about specific content. Similarly, our portfolio company Sixfold AI helps insurance underwriters collate and synthesize information across sources to understand a prospective client’s risk profile.

Voice: As we discuss in part two of this series, significant, ongoing progress is being made on core components of the conversational voice stack. Advancements in speech-to-text models (automatic voice recognition), in particular, have supported a new generation of voice-based copilots addressing transcription use cases. These copilots “listen” to conversations, transcribe conversations, and often offer additional features, such as analytics or preparing and even helping users complete the next steps in a workflow.

Abridge is a company at the forefront of innovation in voice AI for clinical use cases. Its copilot product for healthcare providers transcribes, structures, and summarizes doctor-patient conversations, and writes clinical notes, which doctors can review and edit later. Abridge’s speech recognition technology can automatically detect the doctor’s specialty and the language spoken by the patient (across 28 languages), and translate in real-time to draft the clinical documentation in English.

Image: There are also AI copilots that generate images to streamline design processes in a variety of industries but are particularly becoming more prevalent in the architecture, engineering, and construction (AEC) industry. Workpack AI and Togal AI are visual copilots for estimators in the preconstruction process that automatically perceive, measure, and label project specs.

Agents

While copilots help employees do their work, AI agents fully automate workflows for specific functions with minimal human intervention required. In this way, agents are decoupling software and productivity from human headcount and transforming what businesses can achieve. Agent solutions are not only being built by vertical AI startups; incumbents are diving in headfirst, with Salesforce launching Agentforce, their employee and customer-facing support platform. 

Pricing strategy for agents is still developing. Given that agents can substitute future incremental headcount and allow existing employees to work on higher value tasks, many are priced based on the solution’s output relative to human workers, and agent ROI is framed in terms of the money saved on expanding headcount. 

Initial use cases for AI agents

Software sales: Relevance AI’s sales development representative (SDR) agent, Bosh, automates the process of identifying, researching, and contacting leads, and scheduling meetings.

Recruiting: LinkedIn recently announced the launch of its first AI agent Hiring Assistant, which takes on certain workflows typically done by recruiters, including sourcing candidates, turning notes into drafts of job descriptions, and more.  

Customer support: Slang’s voice AI fields phone calls for restaurants, answering simple questions, making reservations, and connecting customers to employees as needed. Similarly, Assort Health’s AI agent call center for healthcare schedules patients appointments, reducing wait times and dropped calls.

Back office functions: Tennr automates document and referral processing, data entry, and other manual workflows related to healthcare administration. 

AI-enabled services

By using software to automate work, AI-enabled services companies have the potential to deliver cheaper, faster, and more consistent services to the market and take share from incumbent services companies. Pricing for these services generally anchors to existing legacy service provider pricing, but, in many cases, automation allows AI companies to undercut existing providers and still retain higher margins, thanks to AI’s lower cost structure. For example, EvenUp charges per demand letters generated, which is less than the hourly pay required for an in-house paralegal to complete this work (and as a result,  frees up paralegals to complete higher-value tasks.

Initial use cases for AI-enabled services

Legal services: EvenUp automatically builds demand packages for injury lawyers, allowing them to settle cases faster and more economically, and to sometimes win higher settlements than when all aspects of the workflow are done manually.

Medical billing: SmarterDx automates clinical documentation integrity (CDI) specialist work in hospital inpatient departments, analyzing 100% of the data contained in every patient chart to capture revenue that might otherwise be lost.

Third-party insurance administration (TPAs): Reserv automates the insurance claim process and delivers better data and insights to carriers, MGAs, and other partners. Reserve is able to compete with and often replace legacy TPAs like Sedgwick and Crawford by delivering superior service through the use of generative AI in addition to talent. 

Early pricing model examples of emergent Vertical AI leaders 

Among many breakout Vertical AI businesses, we’re seeing AI founders fully embrace usage and output-based pricing to align value capture with value creation. This approach of tying pricing to specific outcomes delivered means an easily quantifiable ROI for customers that can be benchmarked to their existing spend. In addition to this output-based pricing, vertical AI companies are often either putting this output into tiers or adding a base subscription fee to ensure a predictable baseline spend from each customer. This means that vertical AI companies can benefit from the predictability of subscription fees or tiered pricing while also capturing the upside of expanded use—a hybrid pricing represents an attractive model of value capture for vertical AI. 

Up next: Our investment framework and founder advice

Vertical AI is a new frontier. While frameworks and strategies developed for horizontal AI and vertical software companies still have some utility for Vertical AI, we’re also seeing a significant (and growing) number of unique opportunities, challenges, and trends that founders will need to understand and contend with in order to build innovative and valuable vertical AI products and services, successfully bring them to market, and maintain defensibly in the face of competition from new AI upstarts as well as AI features from vertical incumbents who aren’t asleep at the wheel. In the fourth and final installment of our Vertical AI Roadmap, we’ll share our initial investment framework and founder advice focused on four core pillars of vertical AI products and businesses: functional value, economic value, competitive dynamics, and defensibility.

Bessemer Venture Partners has been fortunate to partner with legendary vertical software companies like Shopify, Procore, Toast, ServiceTitan, and MindBody — we think the next generation of vertical AI presents an even bigger opportunity. If you are working on a Vertical AI application, we would love to hear from you! Please reach out to our team at VerticalAI@bvp.com.