Tongdao Liepin(6100.HK): When AI starts to "understand" recruitment, the business logic changes

The human resources services industry is going through change. In the past few years, the most discussed factor was traffic: whoever has a larger resume database and more enterprise customers could secure a foothold in the recruitment market. But in 2025, as overall employment stabilizes and structural contradictions remain prominent, companies have become increasingly selective in hiring, and job seekers are finding it harder to find work. The gap in between is matching efficiency.

Whoever can reduce the cost of mismatches can secure the next “boarding pass” during the industry shakeout. Against this industry backdrop, looking again at the performance report for 2025 just disclosed by Tongdao Liepin (6100.HK) will make the context much clearer.

Data shows that during the company’s 2025 fiscal year, it achieved revenue of RMB 1.986 billion, gross profit of RMB 1.522 billion, attributable net profit to equity holders of RMB 103 million, and net operating cash flow of RMB 229 million.

Although the data does show ups and downs—especially given the combined impact of decreased collections in 2024 and the first half of 2025, as well as the overall recruitment market for mid-to-high-end positions not yet having fully stabilized—revenue from the enterprise side for the full year declined year over year. However, in the second half of 2025, by focusing on high-value customers and promoting its AI products, the company’s business cash collections have already turned from declining to increasing on a year-over-year basis.

More worth关注, though, is how this company turns efficiency into a business core using AI.

01

BHC’s three-sided ecosystem continues to expand, with structural change driven by AI

By the end of 2025, Liepin had verified 1.474 million enterprises, up 3.2% year over year; paid enterprise customers were 72,000, up 5.3%; verified headhunters were about 225,000, up 4.7%; and registered job seekers totaled 116 million, up 10%.

As can be seen, during the industry adjustment period, the BHC’s three-sided ecosystem is still expanding. This indicates that the platform’s appeal to all parties has not weakened due to fluctuations in the external environment; the stickiness of existing customers is increasing, and the platform’s ability to acquire incremental customers is also improving.

But more than the growth in user scale, what deserves closer attention is how these users are using Liepin. The most important change in 2025 is that the revenue contribution from continuously兑现 AI products. For the full year, total cash collection revenue from AI products exceeded RMB 100 million, accounting for 5% of total operating revenue—meaning AI is no longer a cost center; it is already a profit center.

This is a signal worth noting in the recruitment industry. Whether technology investment can be converted into commercial returns often determines whether companies can keep stepping up investment in the efficiency competition.

Breaking it down, the AI products that Liepin deployed on the BHC three sides each address efficiency bottlenecks at different stages of the recruitment process.

On the B side, the “Recruiting AI Agent” underwent a comprehensive upgrade to an end-to-end AI-driven mode. The traditional job-matching process that used to take days can now deliver in as fast as 4 minutes. By the end of 2025, 65% of job orders could be delivered within 2 hours with candidates who expressed communication intent, and the product’s monthly repeat initiation rate increased to 66%. A high repeat purchase rate indicates that it genuinely resolves the core pain point of long delivery cycles.

For the interview stage, the multi-interviewer system uses the digital interviewer “Doris” to complete the interview process instead of humans, with consistency in interview assessment results of 95% or more compared with evaluations by senior experts.

This means AI can already take on most of the initial screening work, freeing HR from large volumes of repetitive interviews so they can focus on higher-value final interviews and salary discussions. At the same time, the multi-interviewer system is expanding from a single AI interview into end-to-end talent assessment services such as AI exams and AI tests. In the future, it is expected to form a more complete closed loop in enterprise talent selection and development.

On the C side, the AI job-search partner “Dora” was fully launched in the fourth quarter of 2025. It can communicate deeply with users from a seasoned headhunter perspective, accurately distill core competitiveness, identify personal strengths, deeply optimize user resume quality, and support one-click export. Based on its deep understanding of resumes and job openings, Dora can efficiently screen and recommend highly matched positions for users. It also includes features such as full-simulation “press-the-questions” practice and immersive mock interview functions, helping job seekers improve on-the-spot performance and overall job-search competitiveness.

On the H side, “XiaoYi” upgraded to version 2.0—from keyword matching to semantic understanding—becoming an amplifier of headhunter professional capabilities.

By this point, it is clear that Liepin has built an end-to-end AI system covering the BHC three sides: the enterprise-side AI Agent addresses talent screening and delivery efficiency; the headhunter side’s XiaoYi improves productivity for operations; the interview side’s multi-interviewer platform enables intelligent assessments; and the job-search side’s Dora empowers the user experience. With four stages connected, it forms a complete chain from job posting to interview assessment.

02

AI reshapes hiring efficiency; the moat in the mid-to-high-end market is getting thicker

In the recruitment business, the most core cost is not the storage cost of the resume database, nor the server computing power cost, but the cost of mismatching. If you hire the wrong person, you waste a few months of salary in the light case, and in the worse case you delay an entire business line. Therefore, the competition among recruitment platforms is essentially about lowering mismatch rates. In the past it relied on headhunter experience; now it relies on data and algorithms.

The special nature of mid-to-high-end hiring is also a high degree of non-standardization. Many requirements can be understood but not explicitly stated, which puts higher demands on matching algorithms. This also explains why its online process moves slower than general recruitment; but once a moat forms, it becomes difficult to overturn.

Liepin’s vertical large model, “Tongdao Huicai,” iterates precisely along this direction.

In 2025, through training reinforced with data in the platform domain, the company not only reduced operating costs but also continuously improved human-job matching accuracy. For employers, the large model deeply parses requirements through semantic understanding; for job seekers, it comprehensively understands, infers, and predicts demands based on domain and behavioral data. During the year, Liepin’s large model intelligent agent application technology also received national patent authorization.

The value of technology ultimately has to land in operating results, and this technology foundation is being transformed into a solid commercial moat.

In the second half of 2025, by optimizing organization and smart management of opportunities, the company’s sales efficiency stabilized and increased. The operating leverage effect gradually began to show. The operating middle platform uses AI technology to empower the sales side. Relying on the large model’s analysis capabilities, it digs deep into high-potential users, promotes intelligent management of business opportunities, and drives a year-end improvement in customer coverage and cash conversion rates by 286% compared with the first half.

This means AI is not only improving matching efficiency, but also changing the way work is done on the sales side—moving from mass-recruiting tactics to precise outreach driven by data.

Other businesses also grow steadily, but the logic and core focus remain consistent with the main business.

As for Wenjuanxing, it successfully transitioned to a SaaS model, cumulatively publishing 350 million questionnaires and collecting 24.6 billion completed responses, and active users in the fourth quarter rose year over year. Behind its growth are features such as AI voice answering and AI questionnaire poster promotion, which horizontally replicate Liepin’s AI capabilities into more scenarios.

The flexible staffing business scale has grown steadily as well. Through refined operations, it reduces costs and increases efficiency, similarly reflecting an efficiency-driven pattern.

In the talent development business, as of the end of 2025, the company registered 116 million individual users, up 10% year over year. Active users increased significantly, with the average monthly active users in 2025 up 15.3% year over year.

The key to user operations is that as B-side and C-side AI products fully launch, the platform’s appeal and user engagement are effectively boosted. At the same time, by continuously optimizing its resume grading system and focusing on the structure of core users, the company makes targeted efforts across user acquisition, retention, and conversion—further strengthening its ecosystem advantage.

Overall, Liepin demonstrates an operational shift from traffic-driven to efficiency-driven. The full deployment of the AI product matrix not only brings direct revenue contributions, but more importantly, it reconstructs the cost structure across all stages of recruitment.

When the cost of mismatches is systematically compressed, competition among platforms is no longer about who has a larger resume database, but about who can understand more precisely the hidden matching logic between people and job openings. The accumulation of this capability is making Liepin’s moat in the mid-to-high-end market increasingly thick.

03

Conclusion

Beyond business changes, Liepin’s shareholder returns are also worth paying attention to. In 2025, the company announced a special dividend plan for the first time, paying a special dividend of 42 HK cents per share. Meanwhile, at year-end, the company actively carried out share repurchases in the Hong Kong stock market, sending positive signals to the market.

This year, the company will continue to push for year-end dividend payments. The board of directors recommends a final dividend of 20 HK cents per share to the company’s shareholders, totaling HK$101 million. For a company in an AI-driven transformation period, continued and stable dividends and buybacks not only reflect management’s confidence in cash flow, but also show its capability to balance business investment with shareholder returns.

Looking back at Liepin’s annual report, several keywords are worth noting: collections turning positive in the second half of the year, revenue contribution from AI products, BHC ecosystem continuing to expand, and positive returns to shareholders. After years of focusing on the mid-to-high-end market, Liepin has built a complete three-sided ecosystem, and its AI product matrix has also run a commercialized path. This combination has scarcity in the current Hong Kong stock market.

Of course, the macro recovery timeline remains unclear, and industry competition continues. But a stock that combines a margin of safety (strong cash position), a growth rationale (AI-driven), and valuation advantages (historically low levels) is still worth tracking over the long term.