In the field of AI applications, Harvey is undoubtedly one of the brightest stars.

In July this year, Harvey just completed a new round of financing, with a valuation of up to $1.5 billion. This means that it took Harvey less than two years to join the ranks of unicorns since completing its $5 million angel round of funding at the end of 2022.

Harvey's soaring valuation is inseparable from the support of star investors. Harvey's shareholder list includes not only AI giants like OpenAI but also renowned investors such as Sequoia Capital and Elad Gil, the largest independent venture capitalist in the United States.

One of the main reasons why everyone is so optimistic about Harvey is that its operating data is performing exceptionally well.

In just two years, Harvey's ARR (Annual Recurring Revenue) has approached $30 million. According to Lenny's statistics, the average time for the best B2B SaaS products over the past decade to reach $1 million in revenue was two years, whereas Harvey achieved $30 million in just two years.

Why has Harvey been able to achieve such rapid growth? And how should we view the opportunities in AI legal technology?

/ 01 / A Legal AI Assistant That Wins Praise and Popularity

Many people may not realize that the legal industry has always been one of the most proactive in adopting emerging technologies.

Previously, AI technology (primarily NLP) has already been applied in areas such as contract management, litigation prediction, and legal research. However, most of these efforts focused primarily on information retrieval, making it difficult to conduct in-depth processing and analysis of information.

Unlike the past, large models have improved legal AI software capabilities in two ways:

First, LLMs enable conversational search and can summarize data content, providing detailed answers to questions. Second, the generative capability, which goes beyond mere extraction, can be applied to drafting legal contracts and providing judgment suggestions to judges.

These capabilities are also reflected in Harvey's services.

As a leader in the legal AI field, Harvey primarily offers two service models: one involves fine-tuning legal expert large models to assist lawyers in contract analysis, regulatory compliance, claims management, due diligence, and broader legal consulting services; the other involves building customized models for clients to achieve better results.

The former service scenario mainly encompasses three areas: workflow automation (such as document drafting, review, analysis, and summarization), legal research, and AI assistance.

For example, during the document drafting process, when legal professionals need to draft a contract or a complaint, Harvey can generate a preliminary draft based on the provided Q&A content, which lawyers can further revise and refine.

Similarly, in the legal contract review process, when lawyers need to ensure compliance with the latest regulations, Harvey can help sort through relevant regulatory requirements and preliminarily analyze potential compliance risks in existing policies.

In addition, Harvey can also assist in legal research. For instance, when a lawyer is handling an intellectual property case, Harvey can help list relevant legal provisions, precedents, and academic articles, accelerating the case preparation process.

Apart from providing legal-related services, Harvey can also help large law firms build their own customized AI models. For example, PricewaterhouseCoopers (PwC) has partnered with Harvey to train its proprietary AI models, creating customized products that include tailored services for PwC clients and streamlining internal legal processes.

Currently, Harvey's products have achieved remarkable commercial success. According to Brian Burns, an employee at Harvey, the company's ARR (Annual Recurring Revenue) has approached $30 million.

According to Lenny's statistics, the average time for the best B2B SaaS products over the past decade to reach $1 million in revenue was two years, whereas Harvey achieved $30 million in just two years.

In the past year (August 2023 to August 2024), Harvey's user adoption rate surged from 33% to 69%. Meanwhile, retention rates have also been impressive. Data shows that Harvey's user retention rate remains around 70% after one year. In comparison, the monthly churn rate for SaaS products typically ranges from 3% to 8%, with an annual churn rate between 32% and 50%.

In terms of clientele, Harvey has revealed that over 15,000 law firms are waiting in line for its AI services, with over 100 firms already paying for Harvey's products.

Among its strategic partners showcased on the official website, Harvey boasts an impressive list of international top-tier law firms and institutions, including Allen & Overy (A&O), PricewaterhouseCoopers (PwC), O'Melveny & Myers, Reed Smith, Macfarlanes, and CMS.

/ 02 / Cost Reduction Drives the AI Legal Boom on the Verge of Explosion

The smooth application of AI in the legal field is primarily attributed to two factors: the high cost of lawyers and the strong demand for cost reduction, as well as the high compatibility between large model technology and legal business logic.

As a knowledge-intensive industry reliant on human capital, the hiring cost of lawyers is substantial. Taking Beijing as an example, the average annual revenue generated by lawyers in Beijing from 2019 to 2023 reached a staggering RMB 840,000. Even in Inner Mongolia, where costs are relatively lower, the average annual revenue generated by lawyers still amounts to RMB 160,000. In developed countries like the United States, these figures are even higher.

The exorbitant human costs have motivated law firms to adopt AI technology to reduce costs. Furthermore, the business model of the legal industry, which revolves heavily around knowledge and text, aligns perfectly with the characteristics of large model technology.

Firstly, a significant portion of legal work involves text processing, which happens to be an area where AI excels. Secondly, legal work is highly knowledge-intensive, and most of this knowledge is available in existing textual materials, reducing the cost of large model learning.

Thanks to these two factors, the AI legal industry is booming rapidly. In 2023 alone, the AI legal industry grew from $940 million to $2.39 billion, representing a growth rate of over 250%.

Looking ahead to the development of the AI legal track, two trends are worth noting:

Firstly, the value of high-quality data is continuously increasing. Given the legal industry's stringent requirements for information accuracy, there is an even higher demand for data quality. However, due to data privacy concerns, large model enterprises have limited access to industry data.

The issue of data is a challenge faced by all AI legal companies. Currently, there are two primary methods to address the data source problem:

One is to collaborate with large law firms to obtain data while meeting clients' compliance requirements. For instance, Harvey has already partnered with large law firms like Allen & Overy.

The other is to acquire more industry data through mergers and acquisitions. Last year, Thomson Reuters acquired the AI legal company Casetext for $650 million, recognizing its data advantage built over years of serving clients. Prior to the acquisition, Casetext had over 10,000 law firms and corporate legal departments as clients.

This year, Harvey had plans to acquire vLex, a global legal intelligence platform that provides legal and regulatory information from over 130 countries, boasting the world's largest legal and regulatory database. However, the acquisition ultimately did not materialize.

Secondly, integrating AI into the legal field requires a deep understanding of industry needs. In a previous interview, Harvey CEO Winston Weinberg disclosed that around 30% of the company's employees have a legal background.

These legal experts play a crucial role in constructing customized workflows and processes tailored to specific legal tasks. Weinberg explained that only experienced lawyers possess deep expertise in decomposing and recomposing complex legal work at the task level, and this type of 'process data is not publicly available anywhere.'

/ 03 / Conclusion

In the context of widespread exploration of large model applications, Harvey's popularity among investors stems from another crucial factor: due to data privacy concerns, legal AI applications are difficult to replace with general-purpose large models.

Most AI application entrepreneurs face a common challenge: how to ensure that as models become increasingly powerful, they do not encroach on the service scenarios of application products.

The legal industry offers a valuable insight into this challenge. Given the legal industry's emphasis on data privacy, valuable data is difficult to obtain publicly, yet output quality heavily relies on this specialized data. At least for now, general-purpose large models struggle to impact specialized models in the legal field.

Written by Lin Bai