Software is regarded as one of the most important scenarios for the implementation of AI. Sequoia Capital has mentioned that AI has the potential to replace services with software, creating a market opportunity of tens of trillions of dollars.
Despite the huge opportunities, there is still no clear path for how AI software can be truly implemented. On this issue, Bessemer recently made a valuable point:
Vertical AI software will be the future.
Speaking of Bessemer, people who are familiar with the SaaS industry may be familiar with it. It is one of the most professional investment institutions in the SaaS field in the United States, having invested in more than 200 SaaS companies in the past 10 years.
Although vertical AI is still in its infancy, we can still see the emergence of a number of excellent companies in the vertical AI field after the rise of generative AI, such as AI legal unicorn EvenUp (founded in 2019), AI medical company Subtle Medical (founded in 2017), AI medical company Abridge (founded in 2018), and automated collaboration software platform Fieldguide (founded in 2020).
Combined with the business cases of these vertical AI companies, Bessemer has developed 10 roadmaps for vertical AI implementation, covering the functional value, economic value, competitive position, and defensibility of vertical AI.
01
The implementation of vertical AI should start from the actual needs of customers
The core workflows of different industries have different needs for automation. However, whether or not workflows have the foundation to automate is not the only factor that vertical AI companies need to consider when building their businesses.
Customers’ interest in automation and their different requirements for automation will also have a great impact on the implementation of vertical AI.
Sometimes, these preferences or requirements can be addressed in the product design. For example, if the order is below a certain cost, a dental practice may want to set up the purchase of medical supplies for automated interviews, but interviews with large amounts will still require manual review.
In other words, the AI procurement solution needs to be flexible, not only to automate the procurement of some orders, but also to allow human labor to participate in other orders.
As another example, a law firm may be willing to fully automate the process of paying clients. However, when it comes to core workflows such as writing legal briefs, they need human feedback to create the final output (e.g., creating a first draft) because they want to control what comes out of the end.
The implementation of vertical AI requires sufficient research on the market and user needs of vertical scenarios.
For example, in the healthcare sector, AI solutions offered by AI companies such as Abridge to manage workflows are widely adopted as clinicians want to automate administrative tasks such as record-keeping.
While there is also interest in the use of multimodal AI in diagnostics, penetration remains low because the payment model for healthcare lags behind the technological innovation of its industry.
Therefore, the implementation of AI in vertical scenarios not only needs to consider whether it has the conditions for automation, but also needs to pay attention to the actual needs of customers and their expectations for AI.
02
Seamless integration into existing scenarios is the only way to build a product moat
Vertical AI solutions need to not only perform tasks well, but also build real moats.
AI solutions that can be easily replicated will face significant competitive pressure.
For example, in the financial services sector, there are more and more use cases for accounts receivable and accounts payable (AR/AP) automation solutions, where AI capabilities for data matching and invoice reconciliation may provide some value, but these subtle capabilities can easily be integrated into a workflow tool and replaced by industry-specific workflow vertical AI solutions.
In order to reduce the risk of commercialization of large models, the best vertical AI applications not only need to fully cover the entire business process, but also need to achieve seamless integration with existing systems through APIs/plug-ins.
Many B2B AI startups achieve the latter by partnering with established platforms, especially large existing ones, to create value through seamless integration.
For example, Sixfold, an AI insurer, is embedded in the existing policy management system (PAS) in the form of APIs or plug-ins, so that insurers do not need to completely rebuild the old system or rebuild the workbench. This “plug-and-play” integration allows underwriters to effortlessly integrate Sixfold’s AI capabilities directly into their daily workflows.
03
Look for landing opportunities where productivity is limited
AI is reshaping the division of labor in the workplace: it not only replaces repetitive work to free up manpower, but also empowers enterprises with breakthrough operational capabilities. Vertical AI products with truly transformative value often have two core advantages: full-process automation and massive data processing capabilities, which are areas that are difficult for humans to reach.
For example, Rilla, an AI company in the housekeeping space, records and analyzes sales reps’ face-to-face interactions with customers, and is able to provide customized feedback and recommendations to sales to help salespeople improve performance. Without Rilla, the sales manager would have to personally accompany the sales rep on the site visit, but would still end up being limited by personal energy.
On the other hand, Rilla can also audit large amounts of conversation data from sales reps across the company, which means that the amount of data it provides to sales reps is based on a much larger amount of data than any sales manager has.
That’s why certain industries, such as sales and marketing, services, and legal, are particularly well-suited to the implementation of AI:
Success in these areas is based on the knowledge generated by a large number of written texts and practical records. In the past, this was a time-consuming task, but now AI is able to do it better, or even take over completely.
04
Efficiency improvement, the key point of vertical AI products
Using data to visually demonstrate to customers the efficiency gains brought by AI solutions can dramatically accelerate sales cycles and improve customer retention.
This efficiency gain usually comes from two things: controlling costs and generating more revenue.
For example, Abridge can automatically record conversations between doctors and patients, reducing physicians’ workload, increasing physician satisfaction, and improving physician retention.
By improving retention, Abridge has dramatically reduced the cost of recruiting and training physicians – often in the millions or even tens of millions of dollars per year.
In addition to controlling costs, Abridge also increases revenue by saving each doctor one to two hours per day.
This extra time allows physicians to see more patients, directly improving the operational efficiency of the hospital and generating more operating revenue. Abridge’s detailed records and summaries of each patient visit also prevent revenue leakage by ensuring comprehensive coding and billing.
The case of EvenUp illustrates this point as well.
EvenUp leverages AI technology to generate demand packages for personal injury law firms, whereas in the past paralegals would have spent days collecting data from clients, collating through hundreds of documents, extracting data from medical and police reports, and more.
Because EvenUp’s legal operations team reviews every letter, law firms can maintain high quality standards while drastically reducing (or eliminating) the amount of time their teams spend on on-demand packages. This extra time allows the company to take on more cases, which increases revenue.
05
AI is reshaping service delivery and pricing, bringing new business opportunities
New delivery and pricing methods enabled by vertical AI solutions are opening up new opportunities.
Previously, many vertical scenarios did not have enough TAM (Total Addressable Market) to build a traditional software business. Now, this part of the market gap is expected to be filled by lower, lower-cost, and more standardized AI.
Historically, service businesses have been difficult to make a profit because of the high cost of specialized workers. And AI is set to revolutionize that. As of 2024, Bessemer’s vertical AI portfolio of service-oriented companies has an average gross margin of about 56% and an average capital burn rate of 1.6x, which means that only $1.6 of working capital needs to be invested for every $1 earned.
Some AI service products have shown better delivery results with the support of human QA, and other AI products have also performed well with AI products as their core service products.
06
Build for overlooked categories and workflows
In the field of sales and marketing, there are already large and well-resourced competitors, such as Salesforce or ADP. In this case, AI vertical companies should look for areas with relatively less competitive pressure.
While gaining a first-mover advantage in a vast market is ideal, most verticals already have at least one incumbency.
But it’s not without chance. When incumbents are stretched thin or slow to integrate AI, fast-moving startups can gain a competitive advantage by building superior, high-ROI AI products and services that can optimize some valuable but non-obvious workflows with automated AI solutions.
07
Serving customers with specific needs
Vertical AI companies differentiate by targeting customers in neglected categories, often with complex requirements that cannot be easily met with AI solutions.
For example, an AI startup that provides services to a bank or government contractor needs to build industry-specific security and compliance tools to sell to customers. This complexity, based on the needs of specific industries, creates a moat for AI companies’ products.
In order to reduce the risk of LLM commoditization, we may start to see foundational model players (such as OpenAI and Anthropic) also start to build corresponding vertical models for customers in these industries.
08
The model is not a reliable moat, but multi-mode can
As model infrastructure costs continue to decline, models will no longer be moats. Vertical AI founders need to ask themselves, “Why are the products we build with AI better than those built with public models and data?” ”
Building new technology architectures to solve specific problems can be one way to do it. For example, fine-tuning LLMs to better reflect the client’s writing style, or using Retrieval Enhancement Generation (RAG) to better perform information retrieval.
Bessemer believes that using RAG technology for industry-specific datasets is also a way to build barriers to business.
New business barriers will be discovered in solutions that can handle more complex (especially multimodal) workflows.
For example, Bessemer’s portfolio company, Jasper, is a good example. Jasper’s AI solution, which is ultimately used by marketers to create long-form blog posts with text-based GenAI capabilities.
Generally speaking, once a post is AI-generated and edited by a marketer, it’s time to look for the right image. As a result, Jasper acquired Clickdrop to strengthen its Jasper Art offering, using multi-modal features (text and images) to meet all the needs of marketers.
09
Focus on modularity and extensibility of the model stack
Traditional SaaS relies on the permutation and combination of standard technology stacks, while vertical AI companies must build a customized infrastructure system: integrate open source models and business solutions through self-developed capabilities, and flexibly fine-tune large language models to achieve the best results for customers.
This approach allows AI companies to get a head start in the rapid iteration of large models. At the same time, it reduces the cost of trial and error, and when the open-source model can achieve 90% of the effect of the commercial model after tuning, there is no need to risk self-development.
What’s more, this approach also allows businesses to devote resources to what matters most: providing customers with quality products.
Jasper is a great example of a product built for flexibility in this regard. The platform sits at the heart of the marketing tech stack and acts as an “AI brain” that helps users develop, design, and execute plans for all marketing disciplines.
The Jasper team designed a modular platform that uses multiple LLMs to run marketing inputs through multiple LLMs based on customer needs, model performance, and cost. For example, if Claude 3.5 outperforms GPT-4 in one case, Jaspe can support interchangeable model infrastructure.
10
Don’t be overly focused on the quantity of data, data quality is more important
The ability of proprietary datasets to build moats has been widely recognized.
But for a lot of early-stage startups, they don’t get the amount of data they want. That’s where data quality comes in, because high-quality data, regardless of quantity, can have a compounding effect that will benefit companies over time.
For example, in the early days of EvenUp, the team was heavily and consciously involved in legal operations, having all claims letters reviewed by humans; In this case, data size is not as important as data quality, and over time, the model will be further refined to improve the product through a large amount of high-quality data feedback.
In the early stage of entrepreneurship, it is more important to create a product with high ROI that meets the pain points of core customers and sells quickly. In the future, as the scale of use expands, proprietary data will follow, and these high-quality data can also lead to product upgrades.
Author：林白
Source：Anthropic投资人最新分享：对垂直AI落地的十个判断
The copyright belongs to the author. For commercial reprints, please contact the author for authorization. For non-commercial reprints, please indicate the source.
