In the realm of enterprise services, Beisen has consistently distinguished itself as an outlier.

Crowned as "China's inaugural HR SaaS unicorn," the company has long held sway over the HCM (Human Capital Management) market. Yet, beneath this illustrious guanghuan (Chinese term meaning 'halo' or 'aura'), Beisen has grappled with persistent losses. It wasn't until the 2026 fiscal year (spanning March 2025 to March 2026) that Beisen managed to eke out a meager 5.0% adjusted net profit margin by implementing stringent cost-cutting measures.

Just as it teetered on the brink of profitability, Beisen's CEO, Ji Weiguo, made a bold and unexpected move that left shareholders and industry peers alike in shock.

On June 24, 2026, Beisen officially launched Mavens, its all-encompassing AI HR expert platform, and declared its full-fledged transformation from a "traditional HR SaaS software vendor" to an "AI application company." To facilitate AI implementation at client sites, Beisen plans to deploy an extensive team of over 300 FDE (Front-End Deployment Engineers). Even more daringly, Ji announced a staggering 1 billion yuan investment in AI R&D over the next two years.

With 1.6 billion yuan in cash reserves and having just turned a profit, Beisen is wagering more than half of its financial war chest on "AI Agents"—a technology that, in China's enterprise services market, still lacks a proven business model.

What's driving this urgency?

For over a decade, Chinese HR SaaS vendors have concentrated on a singular objective: digitizing workflows. They have transitioned cumbersome HR processes, such as resume screening, interview scheduling, offer issuance, attendance tracking, and payroll calculations, to the cloud. Customers have been paying for "management tools" through per-seat licenses or annual subscriptions.

However, this model has reached a dead end in China's current enterprise services market.

Firstly, severe product homogenization has forced vendors into ruthless price wars. Secondly, customer demands have shifted. During economic downturns, enterprises are no longer interested in software that merely "beautifies HR approval workflows." Instead, they demand bluntly: "Can your software help me hire the right people cheaper and faster? Can it do the work for me?"

Traditional SaaS systems fall short in this regard, but AI Agents can deliver.

Following the LLM (Large Language Model) boom, Beisen experimented with "AI skins" in late 2023—adding AI chatbots or resume summary generators to existing HR systems. However, Ji soon discovered that clients were unwilling to pay for these piecemeal AI features.

After reevaluating, Beisen abandoned the "feature stacking" approach in 2024 and pivoted to standalone AI Agents designed to solve end-to-end scenario-specific problems. Take its flagship product, the "AI Interviewer," as an example: it doesn't merely assist human HR but directly conducts initial interviews. It generates interview outlines based on enterprise-specific job models, employs a three-layer probing technique ("Results-Behaviors-Motivations"), and outputs standardized evaluation reports.

Ji calculates that deploying 100 human interviewers for campus recruitment costs 250-300 yuan per interview. With Beisen's AI Interviewer, costs plummet to around 20 yuan.

When AI evolves from a mere tool into a "cheap yet efficient digital worker," the business model undergoes a fundamental transformation. Beisen has shifted from traditional "software subscription" revenue to "token-based consumption" via platforms like Feishu.

Financial data substantiates this logic. In FY2026, Beisen's new AI product contracts surpassed 87 million yuan, marking a tenfold increase year-over-year. The AI Interviewer alone accounted for over 21 million yuan in new contracts, boasting a 120% renewal rate.

This 87 million yuan figure explains Beisen's audacity in abandoning traditional SaaS for AI.

However, the question remains: If implementing a basic AI Interviewer merely requires integrating LLMs, then Baidu's ERNIE, Alibaba's Tongyi, or even startups could accomplish the same. Why does Beisen believe its 1 billion yuan investment will create a competitive moat?

Beisen's answer lies in its robust team of 300 FDE (Front-End Deployment Engineers).

The essence of SaaS is "standardization, asset-light operations, and declining marginal costs." Yet, Beisen is defying this trend by deploying a massive human workforce. Why?

Because enterprise AI is not a "plug-and-play" solution.

General-purpose LLMs may excel at writing poems, but they lack an understanding of how enterprises evaluate "regional sales managers" differently. They are unable to adjust communication tactics when dealing with leaders of state-owned enterprises.

Beisen recognizes that the biggest bottleneck to AI adoption in enterprises is not computing power or models but the "know-how" of business scenarios.

This is Beisen's core competency, honed over 24 years. Before venturing into software, it was China's largest psychological and talent assessment agency, boasting over 300 job competency models, 100 million+ real assessment samples, and a team of 200 psychology experts. This exclusive "People Science" corpus is invaluable for training vertical AI.

But that's not sufficient. To transform these models into usable "AI employees," the FDE team must intervene. These 300 "hybrid talents"—who possess a deep understanding of HR operations, Beisen's PaaS system, and LLM prompt engineering—must work on-site to input enterprise-specific job frameworks, organizational language, and compensation rules into the AI.

As Ji puts it, psychology consultants who felt marginalized during the era of standardized software now find themselves "Beisen's most valuable assets" in the AI age.

By deploying a substantial human workforce for the grueling "last-mile" delivery, Beisen aims to make its AI "client-exclusive." This ultra-heavy delivery system will erode Beisen's newly improved profit margins in the short term. However, in the long run, once these customized experiences crystallize into reusable industry templates, they will create an impenetrable moat against general LLM vendors and ordinary SaaS players.

Despite the grand narrative, Beisen's AI transformation is not without risks.

In FY2026, Beisen reported 1.105 billion yuan in revenue, with AI applications contributing less than 10%. The company only achieved adjusted break-even by rigorously controlling sales and R&D expenses over the past two years.

Now, during this AI revenue transition phase, Beisen plans to increase AI R&D and industry content investment by 1 billion yuan over two years while maintaining a 300-person high-salary FDE team. This level of spending could easily drag Beisen back into losses. Can its 1.6 billion yuan cash reserves withstand this financial burn?

Then there's the issue of "non-standardization" regression caused by the FDE model.

SaaS commands high valuations because it's a standardized, high-margin business. The FDE model, however, resembles "on-site consulting + system customization." If the 300-person team fails to rapidly abstract client-site experiences into reusable product features, Beisen's AI services will devolve into labor-intensive "next-gen IT outsourcing."

Project-based work undermines scalability and gross margins—the antithesis of SaaS logic.

Most critically, enterprise clients are facing trust crises and payment hesitancy toward AI applications.

While the AI Interviewer has found success in high-frequency standardized scenarios like campus recruitment, Beisen's Mavens platform plans to introduce "AI Leadership Coach," "AI Shift Scheduler," and other applications that access core enterprise data such as compensation, performance, and organizational structures. With Chinese enterprises growing increasingly sensitive about AI data privacy, convincing clients to entrust their data to a third-party AI platform for training remains an enormous trust barrier.

Ji once joked, "I thought I was nearly ready to retire, but now I have to start over." In reality, he had no choice.

In 2026, as SaaS falls out of capital favor and traditional software growth peaks, clinging to the old "per-seat license" model would likely condemn Beisen to mediocrity. Actively smashing its old SaaS bowl to bet its 20-year-accumulated "assessment knowledge base + PaaS system" on Agents—to sell "digital workforce capabilities"—is the company's only viable new story.

This transformation is not just Beisen's but a survival imperative for the entire ToB (To Business) industry in the AI era.