Bairong Inc Stock

Distinguished investors, analysts and friends from the media, good day. I am Xuan Qin, Investor Relations Director of Bairong. I sincerely thank all investors joining us online for your continued attention to Bairong. The company released its 2025 annual results announcement after the market closed yesterday, March 26. In this results presentation, we'll report and share the company's operating achievements for 2025, look ahead to 2026 and answer questions of interest.

This results presentation consists of 3 parts. The first part features the CEO introducing business progress and outlook. The second part features the CSO explaining financial performance. The final part is the Q&A session.

Participants online can submit questions in the live stream question by phone. Numbers for Mainland China are (023) 627-37123; For Hong Kong, (852) 301-83602; for international callers, +86-23627-37100. Chinese conference passcode is 595868414. English conference call is 291375421. Management will answer questions after the presentation.

This presentation contains forward-looking statements reflecting the company's current beliefs and expectations about the future. These statements include words like anticipate, believe, intend, estimate, expect and words with similar meanings. All statements in this presentation, other than historical facts, are forward-looking statements. These forward-looking statements reflect only the views of the company's management as of the date of this presentation and are not guarantees of future performance. The company, any member of the group or any of their relevant affiliated parties or any of their respective directors, officers, employees, advisers or representatives assumes no obligation and expressly disclaims any obligation or commitment to disseminate any updates or revisions to any forward-looking statements.

Now I'll introduce the company's management attending this result briefing. They are Mr. Zhang Shaofeng, Founder, Chairman of the Board and Chief Executive Officer of Bairong; Mr. Kelson Chen, Executive Director and Chief Strategy Officer; Mr. Tan Ying, Co-Partner and Senior VP.

Now I invite the Founder, Chairman and CEO of the company, Mr. Zhang Shaofeng, to introduce the 2025 business progress.

Thank you, [ Sandy. ] Dear shareholders, investors, analysts and friends from the media, good morning. Welcome to Bairong's 2025 Annual Results Presentation. It is a great honor to report on Bairong's performance progress and growth and to share with you some of our thoughts on long-term development in the agentic AI era.

First, company introduction. Let me briefly introduce Bairong, the company. We're a leader in enterprise-grade AI agents. And we are the inventor of results as a service business model. We were the first to brought up this concept globally. Through the results SaaS service or RaaS business model, we deliver silicon-based employees or agents to institutional clients. Our positioning is to design, develop, deploy and train AI agents for enterprises.

Relying on the results cloud for enterprises, we integrate our self-developed BR large language model with the [ Bairong ] enterprise-grade AI agent operating system. We help clients build their own silicon-based employee systems. And we have a role-based charging models or we could charge for services. In other words, one department can be outsourced to Bairong, and we will charge by results. And therefore, we either deliver by roles or staffing or by services.

To date, we have served over 8,000 institutional clients with core customer retention rate in some key business areas reaching 98%. Representative clients include industries such as Internet, retail, telecommunications, education and health care with applications scenarios like marketing, recommendation, operations, the customer services, outbound costs, contract invoice system pricing, credit granting, anti-fraud, claim settlement and archive management, recruitment and training, among others.

Next, performance overview. In 2025, the macro environment remained challenging. However, by leveraging solid business insights and customer foundations continue to deepen its AI strategic transformation and achieved steady development.

First, regarding revenue, our MaaS business, which is our ballast, has shown steady recovery with a year-over-year increase of 9%, surpassing the 5% growth we recorded in 2024. As you know, there were policy environment pressure, including #9 document. We were able to achieve 9% year-over-year growth, really demonstrating our advantages in this era. And we were able to achieve increase in both the number of clients and average revenue per client. Furthermore, retention rate for MaaS key clients improved to 98%, a level that leads the enterprise service field. This reflects our clients' high recognition and continued reliance on our AI decision-making capabilities.

Regarding the BaaS business, or Business as a Service business, as I communicated with everyone at midyear last year, policies like document #9 and integrated filing and implementation have varying degrees of impact on the revenue of both the financial scenario and insurance scenario. However, it is noteworthy that the total premium facilitated by the company last year increased by 20% year-over-year. First year premium, a very important indicator facilitated increased by 25% year-over-year. Life insurance premium renewal rate remained stable above 90%, making us among the best in the industry. The pressure on revenue was mostly because of policy pressure from the regulatory environment. However, we're still improving in terms of our positioning in the market. And once the regulatory environment becomes more favorable and as there's less competition, we'll continue to grow.

And I know everyone is concerned about the group's overall revenue and profit performance last year, especially the significant pressure on profit. Let me try to elaborate on this. At the beginning of last year, the company signaled to the market that it would significantly enhance R&D and business investment for the full year, seizing market shares at all costs. This was noted in our annual report. If we add back investment, especially AI investment from last year, I think we can all demonstrate -- we can show that the company's mature business maintains a stable profitability. So what were these AI agent or AI-related investments went to.

First, talent reserves. Last year, we spent a significant amount of funds on recruiting AI talent to strengthen our algorithm, engineering and product teams, building AI-native organizational capabilities. If you followed our annual report, you will see that our R&D headcount increased by 383 year-over-year with total headcount -- R&D headcount now exceeding 1,100. And this accounts for a substantial 64% of the company's total workforce. The R&D team is the long-term foundation for our growth in the AI space.

The second part where we spent money was on proprietary data. We expanded the procurement and labeling of vertical data, enhanced model training corpus and consolidated our moat in data assets.

Third, computing infrastructure. This involved expanding our IDC facilities, adapting to domestic GPUs and achieving self-reliance in our training and inference platform to reduce long-term computing costs.

Fourthly, we spent money on operational support. This covers expenses related to workplace expansion and personnel growth, ensuring the organization maintains high efficiency and synergy during a period of rapid scaling. In comparison, I think Bairong's cost effectiveness in preemptive AI investment is relatively high. Last year, our gross margin remained high at above 72%, and we maintained profitability while heavily investing in the future. This is -- this puts us at an excellent level among current AI agent companies out on the market. Moreover, these investments are both necessary and urgent for a company that has the ambition to become a leading firm in AI agents.

Many people have read a report published by -- Research at the beginning of the year, about 10,000 words in this lengthy report, which projected that in 2 years' time or in 2028, global intelligence would scale to such an extent that agents will be widely adopted. The dawn of future is near and the time was no one. If we did not invest in AI, then we would have just watched, for example, the open flaw development on the sideline during the Chinese New Year this year. And I think for the new business, we have high hopes for, I completely understand that investors' eagerness to see investments translate into financial returns.

Here, though, I'd like to offer one way to look at it. In the early development stage of AI agent business, there is a significant amount of opportunities to scale up. For example, in particular, we're talking about project-based demand. In other words, we can send employees to outside -- to the clients' place and to deliver projects through customization. This is certainly something that we can do to scale up revenue. However, we know that the last cycle of software boom in China did not end very well. If we accepted all such demand, revenue would show very attractive short-term returns. However, that would not be sustainable.

We aim to build long-term sustainable RaaS cooperation with B2B clients. We will not aim sacrifice long-term sustainability for short-term profitability. If we went for projects or short-term projects, then at the beginning of the company, we wouldn't have been where we are today. There are some listed companies in China who are who are doing well with those large projects. But I think we are on a path where if we overcome a short-term pressure, long term, we can become more sustainable and scalable. And I think we look at more important leading indicators such as -- similar to what tracks in this company. And for example, we look at efficiency improvement, improved -- achieved on the product side, a number of high-quality clients acquired and continuous output of silicon-based. In other words, we will not sacrifice our long-term sustainability for short-term profitability.

So on the product side, last year, we built a -- actual silicon-based employee matrix covering customer experience, CX, and employee experience, EX. I think that this categorization is also something that is quite common in the industry. One is for improving external service quality and the other is for improving internal efficiency. So the silicon-based marketing and services , which is paying is already used in over 1,000 centers in wealth management, telecom communications and retail. [indiscernible] has achieved a 90% cost savings and reducing annualized attrition from 70% to 0%. Human employees would leave, right? And attrition is high, but silicon-based employees do not leave their jobs unless you cut the power.

Consultation conversion rate improved by 210%, and customer satisfaction increased from 16% to 55%. The silicon-based analyst called [indiscernible] by connecting multiple AI agents, health professionals quickly achieve a close look process for listening, recording, summarizing, writing. Previously, an -- report would take maybe 10 people 20 days. But now you only need a couple of days, 4 days, for example. And there's also a silicon-based legal business finance and tax effort [indiscernible], and it's for employee experience that handle standardized tests like data collection, case retrieval, document drafting and process management. So a lot of the services, for example, for overseas market expansion, location selection, among other things, that is able to help our annual client cost from CNY 3 million to CNY 5 million a year to CNY 1 million a year. The silicon-based recruitment specialist, [indiscernible], driven by AI interviews and intelligence screening, shortens the recruitment conversion cycle from 28 to 2 days, improved trial placement matching rate from 60% to 90% and increases individual HR monthly recruitment efficiency from 5 people to 20 people. Additionally, internally, the company built the silicon-based employee home based on the Bairong AI agent operating system. So this operating system manages over 200 types of internal silicon-based employees. [indiscernible] corresponds to the workload of multiple human employees for carbon-based employees and can be frequently called by a multiple users, for example, by finance, legal, HR operations among others. And we are even achieving lead to cash for client development through the system. We have over 2,500 clients that are being served by our silicon-based employees from onboarding to invoicing to renewal to, for example, purchase of more add-on modules, et cetera. And we were able to cut the staffing on the team from 50 to only [ 5 plus 18 ] silicon-based employees.

And on the client side, last year, with silicon-based employee core, we achieved commercialization breakthroughs from 0 to 1 across hundreds of industries, which validates the strong universality of AI agents, cross-industry migration capabilities. So previously, we were in, for example, banking sector for the most part, but now in the securities space, China securities, for example, are using wealth AI silicon-based employees. And we are also seeing banks that are purchasing our silicon-based AI service agents at CNY 5,000 a month, which help them cut costs.

And then we also have a trust client that's been using our customer service AI agent. And then there is also an international investment bank that uses our telephone and WeChat session management AI agents, and these agents are able to interact with their end client in a very realistic scenarios. And already, our services are paying for themselves for the end client. And also a state-owned joint stock bank became the first among the 6 major banks in China to sign an AI expert service framework agreement with Bairong.

In telecommunications, Bairong deeply participated in the digital intelligent transformation of the telecoms industry, achieving systemic verification and scaled implementation in provinces such as Guangdong. In Guangdong, we were able to facilitate over 1,000 telecom packages for a client in the industry.

In health care, we are providing mental health consulting AI agent for a renowned medical university. And that university trains doctors, but they -- the students there also need psychological consultation and our AI agents can help them in that regard. And we also have a chronic disease management agent for a medical technology company. We also are able to provide --, which offers recruitment services for blue collar recruitment processes for many large clients. And for emerging carmakers, for up and coming car-making space, we're able to provide AI agents that help us sales lead management. And we are also able to provide AI contract support for maternal and infant industry leader who is in the high-end [ post-modern ] care space. And we also have a dual carbon intelligent AI agent that provides audit for Environmental Policy Research Institute, which really verifies our ability to migrate from financial health to complex industrial scenarios. Actually 90% of nonfinancial industries are not as complex as the requirements in the financial sector. So we have strong abilities for migration. And I think for clients, their willingness to pay is high, especially versus traditional software because AI agents can pay for themselves to a large degree.

Also, I want to report on last year's R&D process. We've increased investment to attract top AI talent, enhanced data and algorithm input and solidified infrastructure foundation. We've achieved breakthroughs in technology and have built differentiated advantages. The company has obtained 573 patents and software copyrights. We are set up a joint lab together with the -- University. And we achieved in-depth integration from recruitment to engineering, R&D development among all the steps in this closed loop. And in particular, we work with University Research Institute on MOE application, especially for the automated AI agents who can cover long processes and complex processes.

At the application level, we are -- we have completed a key leap from usable to user-friendly in speech model. We upgraded from the traditional 3-stage architecture to an entry and large model. In other words, we don't go from voice to text to voice, but rather just go from voice to voice or speech to speech. This can greatly improve our service quality. And our model already supports 52 voice [indiscernible], including Cantonese, Shanghai and other dialects and as well as foreign languages. Decision latency could be compressed from ours to milliseconds, less than actually 200 milliseconds. Understanding accuracy, [ exceeding 98%]; emotion recognition exceeding 5%, interruption recognition can reach 100%, and response speed or response time now is under 40 milliseconds.

Also, in the silicon-based employee onboarding and application space, we have a life cycle management of a system of AI agents, similar to how we recruit human employees. We're able to reduce the development cycle for silicon-based employees from 2 months to 2 weeks. And particularly, we're able to allow for self-evaluation of silicon-based employees before deployment, but also allow them to self iterate after deployment. And this is very impressive.

And at the large model level, our large models underwent 6 iterations, covering scenarios like telecom package, recommendation of wealth management, product marketing, credit card installation, recruiting screening as well as other scenarios that I mentioned. And not just large language models, our vision models achieved breakthroughs in financial and medical document classification and complex structure extraction. It was a breakthrough from [ 0 to 1. ] And this opened a new path for future commercialization.

At the cost efficiency level, we understand that the cost profile is very important. And we actually addressed the industry -- where cost determines the ceiling by restructuring our self-developed training and inference platform, where latency was reduced by 1/3, resource usage saved by 50% and throughput increased by 1.5x.

We're able to use a mix of different GPUs and including domestically the policy chips and utilizing the BR Vortex inference engine integrating computing, intelligence -- and multilevel caching computing research utilization increased by 30% on -- throughput also increased throughput as threefold.

In terms of awards and honors, Bairong's AI capabilities are moving from industry recognition to standard setting. We were selected for Morgan Stanley's China top 60 AI companies and received comprehensive coverage from IDC. And from industry large models, AI agent development platforms and deep application scenarios like finance, marketing, HR, legal, we will receive coverage from IDC. We were consecutively listed on KPMG's Double 50 list for 2 years. And at the national level, Bairong platform and the BRL large model passed filing with the Cyberspace Administration of China. Also, we became a leader in the standard setting volume for intelligent AI ecosystem construction group under China [indiscernible] of information and information and commutations technology. And we also obtained DevOps Level 3 certification. At the industry level, we were named 2025 China's financial AI agent service excellent provider included in IDC's peer scale best practice cases for banking insurance AI agents. And we also received recognition as a trusted database research partner and credit risk of prevention and control industry benchmark, winning over such a best digital financial work from Xian International Bank and the annual AI interaction innovation from -- New Media.

Next, on ESG as a benchmark practitioner for new quality productive process, Bairong received A rating in wind ESG assessment. Last year, we're -- we strengthened the industry university research cooperation. We jointly developed an AI joint laboratory with people with disabilities to [indiscernible] agent operating system. We get a full process intelligent solution for [indiscernible] person's employment guidance center in its [indiscernible] platform we embedded intelligent Q&A agents based on a large language model and Bairong effectively solving problems like employment information and symmetry scattered training resources and low efficiency in -- matching for the disable population of building a smarter, more efficient and more precise employment channel for approximately 600,000 people with disabilities in Chongqing.

In addition, we also introduced AI agents into employment service support and add an employment service center in the city in West China. We established standard workflow of intelligent outreach information collection, policy interpretation and follow-up steps. The project now is fully operational. And it's a model that can be replicated elsewhere. The AI agents now can go from extensive work standardized application to more targeted application, and we're able to improve efficiency and our system processes up to -- close to 1,000 [indiscernible] data points [ steady ] with a connection rate of close to 80% far exceeding traditional outbound costs.

Information collection is also more standardized and comprehensive. And this is able to provide a solid data foundation for subsequent services. We believe that best technology is not to be admired from afar but to make people see, understand and trust to truly solve pain points in everyone's work in life.

Next, I want to provide a short outlook for Bairong's future. 2026 will be a critical year for the company's strategy to transition from the capacity-building period to value-recognition period as underlying technologies mature and industry applications accelerate will be driven by the dual engines of scale profitability in mature scenarios and forward-looking positioning for innovative scenarios. And this way, we're able to grow sustainably, and we can also grow for the future. We serve over 8,500 institutional clients. We've committed a lot of experience. We're going to take advantage of the know-how and vertical data points that we get [indiscernible] over the last decade to build a competitive moat, and we will achieve industry-leading or positioning in the era of agentic AI.

In the new year, our key focus areas will include the following: First, industry focus areas. Last year, with the issuance of -- #9, I think the potential risk in the financial and insurance sectors are now all very transparent. In the foreseeable future, the likelihood of the industry going through new policy adjustments is quite low. And we have many actually opportunities that we can explore. For example, in the AI agents that can, for example, manage mid- to low interest rate credit marketing -- wealth management as well as a nonperforming asset or loan management. This is something that is -- where there's a big demand. Additionally, and we think there are some scenarios where we think there are many opportunities.

First, AI coding or that's where I think PMS got achieved in the first instance. Yes. And the second, I think, it's a contact center. It's only after skilled AI coding. So AI agents can receive calls or conduct outbound calls, send messages on WeChat and other channels. So this is what we call CCS, contact center scenario. Contact center scenarios are usually covered by carbon-based or human agents. If you ask an LM today, you will see that there are about 30 million human agents who are working in China. So you have a few million agents who are receiving inbound calls, a few million who are for nonperforming loan management. And we also have a few million, to say the least, that are doing marketing. And the cost per seat is about CNY 100,000 a year. This is a huge market. If we're able to capture just 10% of that market, that's already CNY 1 trillion opportunity. We are eating into the wages there because we're not just software.

And then we also want to be in more specialized domains, for example, legal services or, let's say, consultation such as services provided by, for example, McKinsey [indiscernible] and these contracts can be a few million each, and not to mention financial and tax professional services. These services can be greatly empowered by AI because the output essentially is a document -- various documents. AIs systems today can develop high-quality documents. We can provide AI-native financial and tax consultation platforms for specialty Chinese companies expanding overseas. If you're a Chinese company and you engage with a multinational corporation for consultation and one of such contracts can easily set you back by a few million yuan. I think Chinese companies are going global. I believe that's going to be the main theme in the next at least 10 years. Bairong needs to tap into this trend, and we want to help Chinese companies expand overseas. Through AI agents, we hope to support them as they seek financial and tax services. And this system is called -- and -- will help, especially SMEs that are seeking to extend overseas, and we provide -- we will provide specialized financial and tax computation to these companies. And so far, we're the only one in this space, and we don't have any peers to compete with.

And also, the other thing is we will open -- make it open our core capabilities such as voice interaction and intelligent decision-making to power industries. We think the -- these capabilities will replace the traditional HMI. For example, smartphone user interface and it's all graphic. But the future human machine interface will be voice-based. And you can talk to AI directly and you don't necessarily have to type or click or tap.

We also developed with ecosystem partners such as software vendors, distributors and system integrators to make it even more powerful. We will also leverage the strength of capital. We're trying establish AI industry funds with top-tier capital providers such as [indiscernible] and [indiscernible] for co-investment. And I hope you will see our exciting new -- soon.

And we want people to know that Bairong is willing to leave profit on the table. We not only provide AI agents, we also provide capital support so that we can attract more traditional software companies to join us on the platform, and we're happy to share profit with them at 30% or 50%. Just to -- just for example, we have 2 leading AI companies in the U.S. OpenAI and Anthropic. They are also doing this dual-track approach. There's organic business expansion and also there is capital-backed expansion. And the latter scenario would make it natural for the partners to use their AI tools for AI systems. Similarly, Bairong has invested in an AI BPO company. And we are already helping them to replace human employees with silicon-based employees to help them improve efficiency. We are also helping our wealth management companies and other companies for empowerment. In the future, there's also a possibility that we might acquire such companies. And I think expansion is on the business side and that's based on organic growth and also business expansion similar to Tencent's growth story more than a decade ago. So that in the future, we can build an ecosystem that revolves around the Bairong ecosystem. This is going to, we believe, be critical for Bairong's success in the long term. The ecosystem takes time and efforts to build, but we are patient, and we're more than willing to be patient there. Similarly for NVIDIA, right, the -- ecosystem took them decades to build. But -- is a stronger competitive moat even than their other -- even other chips.

Page break

So as we all know, with start-ups in the U.S. OpenAI and Anthropic. OpenAI being the most well known and followed by Anthropic after OpenAI focuses on consumer sector, it wants to replace Google. And for Anthropic, they target enterprise sector. They want to empower companies with higher productivity, organized momentum and valuation once far surpassed Anthropic valuation and momentum. However, in the past 6 months, with Anthropic's tremendous success in the U.S. enterprise AI agent market with open cloud, it has been very successful for Anthropic, both in terms of revenue and the market valuation that has the trend of catching up with OpenAI, even surpassing OpenAI.

And let me talk here analytical institutions believe that the launch of OpenCloud marks the turning point where Anthropic surpasses OpenAI. The underlying reduction of the widespread industry view, previous generation of innovation focuses more on the consumption and distribution side. But this time, this current AI transformation brings about a greater lead in production and supply. Consequently, the industry is beginning to recognize that Anthropic growth potential exceeds OpenAI because Anthropic helps the supply side and production side. Historically, the most significant industrial revolutions, the steam engine revolution, the electric revolution and the information computer revolution were all, first and foremost, revolutions on the production and supply side, which later benefited the consumption and distribution side. This is also why the current AgI revolution is considered comparable to the first 3 revolutions. While the Internet and mobile Internet revolutions have not yet reached that scale because they were focused on the consumers and the consumption side. The big revolutions need to start with the supply/production side. Anthropic has already proven that enterprise-grade agents have surpassed the critical inflection point from concept to implementation, from tools to productive forces.

And Bairong is positioned to become the Chinese Anthropic. We have the confidence that we -- and the -- because from day 1, we have been delivering results to companies instead of just tools. For a long time, our models were not recognized as the industry benchmark or as the industry mainstream, but we have never wavered in our direction. We have been deeply involved for years and have accumulated instrumental results and experience in multiple industry scenarios. Of course, we are also clearly aware that the domestic enterprise service market has its unique rhythm. And we have been accumulating more house in that particular reason. The Chinese customers tend to have longer decision-making chains, longer validation cycles and more cautious requirement for value delivery. The willingness to pay or the habit of paying is not as strong as that in the U.S., but we have already validated our -- proven that companies are looking to pay. It may take us longer to gain verification in terms of financial returns, but this does not change our long-term optimistic view of this direction. Bairong has never had a classic software-type business since its inception. We know that companies in China are not willing to pay for the tools, they only pay for the results. So all our revenue is priced based on results. This is baked in our genes. We want to deliver results, not just tools.

Leveraging this inherent DNA of delivering results rather than tools, we officially and comprehensively launched [indiscernible] to the market in 2025. This is another year of our -- business. And we started, I believe, a big campaign of reform in the industry around the world. Our AI strategy conference received 400,000 people's attention, and we post which the conference on December 18 last year was the major event. It might not be delivering a big difference in the short run, however, in just 3 to 5 years' time, it will create an insurmountable competitive gap. And previously, the ecosystem of tech companies in China and the U.S. differed greatly. China focused on consumer sectors, while the U.S. emphasized by the enterprise sector. After Facebook, there was no major B2C companies in the U.S., the strength slicing B2B enterprises. But in China, the B2B companies were struggling. The revenue models of Chinese and U.S. enterprise software companies vary significantly. And so the Chinese B2B players in software statutories were loss-making significantly.

For this time around was our silicon-based employees and agentic AI employees. If you look at the service model and products and revenue models, I believe there's going to be a historic convergence. This presents a tremendous historic opportunity for Chinese enterprise -- for Chinese tech companies to overcome past challenges. Although the scale may not be the largest, but we will look beyond the service industries we serve such as Internet e-commerce and communication and finance. And if we just look beyond those industries and carefully analyze our [indiscernible], which is end-to-end result delivery and our [indiscernible] model charging based on low compensation or transaction sharing. You understand that the business we conduct is essentially a pure blooded agentic economy, a concept that is considered advanced even in the United States.

From day 1 of the company's founding to present, our revenue from traditional tool-based software models has remained virtually 0. What we'll provide is, in essence, the new quality labor force for the entire human labor market as -- Chairman of Alibaba recently pointed out at Siemens RXD conference, the global white collar labor market is worth nearly $50 trillion, and this value is expected to be restructured or enhanced by intelligent agents. And just 1 or 2 weeks ago, Alibaba launched their new model --, which focuses silicon-based employees. This concept was proposed by Bairong -- was first proposed by Bairong the world. So basically, we perceive well -- we strongly validates that bb strategic path where we receive is no longer tool software fees but labor compensation. So fundamentally, we're the same as Anthropic. We specialize in priming and dispatching silicon-based employees, we're an AI-related company, and we're going to be part and parcel of this major historic shift will contribute a role model and benchmark for Chinese tech companies.

What type of silicon-based employee is the most promising? Programmers, contact centers, CRM systems, customer agents and also our structural for the agents that handle unstructured data and consulting writing reports; and third would be -- complex processes and even dynamically generating workflows. OpenCloud has being such a big hit because they're able to handle complex processes without any human interference or human intervention. These are the most promising rural growth cases, but personal OpenCloud cannot be used for corporate purposes to consider a lot of authorization and security issues. And for Bairong, we are the first one in China to launch a large-scale, high-capacity validated AI agent for enterprise purposes.

According to my research report, the AI agent market size is expected to expand continuously projected to grow from RMB 57.4 billion in 2023 to RMB 3.3 billion -- RMB 3.3 trillion in 2028. This fully demonstrates the immense market potential for enterprise-grade AI agents as their genuine tools for unprecedented the advanced market opportunity. This window is swift. And within 2 years, it's going to -- it will become clear who possesses changing qualities to a strongly enabled Just as -- enable -- seize opportunity in personalized information distribution.

During the mobile Internet era, ultimately establishing it as dominant force in revenue among Chinese tech companies, Bairong will go all out to see this critical transition period. We aim to establish a leadership position in the silicon-based economy wave and create long-term sustainable value returns for our shareholders. The core management team and I have unwavering confidence in this since the company's listing. I, as the founder, have not sold a single share, which fully demonstrates our steadfast conviction in the company's business model, technological path and future prospects. The above is my brief summary of Bairong's 2025 annual performance and outlook for the future. Thank you for taking time to attend our annual results exchange. Thank you.

Thank you, Chairman Zhang, for your presentation. Now I'm going to give the floor to Mr. Zheng, Executive Director and CFO, to introduce 2025 annual performance and financial highlights.

Thank you, [indiscernible], dear analysts, shareholders and investors and friends from media. I will now introduce Bairong's 2025 annual performance.

In 2025, the company proactively positioned itself ahead of the site opportunity of the AI while increasing strategic investment achieved steady operational performance and financial results. Full year revenue reached CNY 2.92 billion, basically flat compared to the same period last year. Gross profit reached CNY 2.102 billion with gross margin and remaining stable to 2%, mainly benefiting from the scalable [ ROS ] business, which is further realizing economic economies of scale, operating profit was CNY [ 16.38 ] million, net profit CNY 73.87 million with operating margin and net margin [ 3% and 3%, ] respectively. Under new IFRS, net profit reached [ 79 ] million with net margin of 6%. EBITDA reached CNY 2.4 million with an EBITDA margin of 8%. In a year full of challenges, Bairong successfully stabilized the revenue and maintained continued profitability, demonstrating the resilience of this business model and its market competitiveness.

Improved [indiscernible] business revenue declined by 5% year-on-year to CNY 1.901 billion, accounting for 65% of total revenue. Within advanced business, financial Industry Cloud revenue decreased slightly by 3% to RMB 1.371 billion, accounting for 72% of [indiscernible] and 47% of the total revenue. [indiscernible] Industry Cloud continue to respond to regulatory changes with revenue decreasing by 10% to RMB 530 million, accounting for 28% of [indiscernible] and 18% of total revenue. Our cloud business, MaaS business achieved revenue growth of 9% year-on-year to RMB 1.019 billion, accounting for 35% of total revenue. The revenue structure shows the pattern of BaaS as the mainstay and MaaS as the stabilizer. Our MaaS business institutions, you're making -- decisions through various silicon-based employees across different roles.

In 2025, within the RMB 1.019 billion MaaS revenue, core customers, those contributing annual revenue above 300,000 accounted for 78% of revenue. The vendor of core customers increased by 12 year-on-year to 223 customers. The average revenue contribution per core customer increased by 6% to CNY 3.59 million. The MaaS business maintained steady growth and continued to provide sustained cash flow to the company.

Our BaaS business is based on generative AI technology and leverages self-developed silicones to marketing service specialists, institutions, intelligent marketing and intelligent operations, significantly improving asset operational efficiency in wealth management, insurance and internet technology industries. BaaS managed to Industry Cloud, mainly about a business outsourcing AI BPO cooperation model charging based on the business results or scale achieved and does not charge any fees before assisting institutional clients in generating any revenue.

In 2025, as a leading player in the implementation of AITC acquisition, our BaaS financial industry cloud revenue decreased slightly by 3% year-on-year to RMB 1.371 billion mainly due to the impact of the policy environment of which roles Industrial Cloud provides comprehensive customer insights and accurately -- insurance products through decision-based AI and conduct high-value policyholder operation through a silicon carbon integration approach. In 2025, revenue from our BaaS insurance industry cloud business was CNY 530 million. From the revenue structure perspective, new policy contribution revenue was CNY 450 million, a decrease of 8% compared to CNY 487 million in the previous year. The life insurers premium persistency rate exceeded 90%, ranking at a forefront of the industry. Renewal contribution was CNY 80 million, a decrease of 19% compared to CNY 99 million in the same period last year. Total premium reached CNY [ 16.52 ] billion, up 20% compared to CNY [ 5.442 ] billion last year. These reflect our reseller the momentum of our business is stabilizing and rebounding. In 2025, our R&D expenses reached RMB 637 million, an increase of 25% compared to RMB 509 million in the last year, accounting for 22% of total revenue and up 5 percentage points year-on-year.

The incremental investment mainly focused on construction of AI-native capabilities including the expansion of talent, procurement of data assets and depend in R&D of infrastructure and operational expenditure. Sales and marketing expenses reached RMB 1.142 billion, accounting for [ 39% ] of total revenue, flat from last year. General and administrative expenses was CNY 328 million, accounting for [ 7% ] of total revenue, flat from last year. [indiscernible] will continue to maintain investment in core technologies and computing power to ensure long-term competitiveness while focusing on improving marketing efficiency and controlling management expenses and while ensuring growth, further enhancing profitability and return on capital.

As for cash and assets, in 2025, our cash and cash equivalents and similar financial assets amounted to RMB 3.377 billion, down by 8% compared to RMB 3.657 billion at the end of last year, mainly due to Bairong improving capital utilization efficiency. RMB 3.377 billion includes RMB 728 26 million in cash and cash equivalents, CNY 1.7 billion in large denomination certificates to bond deposits as well as current financial assets measured at fair value to profit and loss. In addition, the company is confident in its long-term future and financial importance to shareholder returns.

In 2025, the company repurchased a total of 9.6 million Class B shares from the open market, totaling HKD 90 million as well as the few AI companies in China that have identified a real deployment scenario for agenetic AI and achieved profitability. In 2025, the company maintained a stable revenue scale and sustained profitability in the future Bairong opportunity and continue to empower thousands of industries and business with AI. Thank you.

Thank you, [indiscernible], for your wonderful remarks. We'll now move on to Q&A. [Operator Instructions] I will read out questions later. [Operator Instructions] We'll begin the Q&A session now.

[indiscernible] you have the floor.

This is Jon Dann from Securities. We see that Bairong's AI agent platform is stabling very well, which reminds me asking you currently popular enterprise-level AI agent giant as rocket and the open-source autonomous AI agent platform, OpenCloud. They are very similar. Not only capable of answering questions, but also completing tasks, making decisions and taking action with minimal user intervention. I believe that in the future, the technical barriers to building AI agents will become increasingly lower and the deployment will become more convenient. And as this backdrop can be on mission of achieving scale commercialization through AI agents be realized as expected in which affects does the company's differentiated competitive advantage line?

Thank you, from Zhongtai Securities. I'll give the floor to Mr. Zhang Shaofeng, CEO and Chairman of the company.

Thank you for this great question. This is definitely pertinent to the core of our business. As you said, the OpenCloud source models that are becoming more and more agentic AI capabilities and the company's understanding and recognition of agentic AI has been improving OpenCloud from the Spring Festival to [indiscernible] last year. It was very inspiring and exciting. And I think I received a lot of questions from many companies, friends, and they say, why don't I just download my own OpenCloud and use it directly. Well, if you remember, you went on seeking out OpenCloud, we had just followed and for ourselves. That's what you thought. However, if you look at what happened last year, how many companies truly -- by themselves by just downloading it, what we need is the agentic AI at a corporate level. Corporate agentic AI is much more demanding than the individual consumer-grade AI. They have to be very responsive, at the same time, cost-effective OpenCloud is fundamentally an open source agenetic AI framework.

Last week, there was a problem. Meta, internally, because they were too aggressive, they used to be open class framework and there is a lot of security and could result in breaches of information and the leakage of the confidential information. So it's very hard to just download it and use it directly as a company because there are many things to take into consideration.

How well do you understand your own security environment? How can you manage the tools? At Bairong, we had a home for silicon-based employees. That's a platform for managing simplistic AI employees. We started 2020, early 2020, and then there was no OpenCloud back then. I think OpenCloud is kind of accelerating people's understanding of the industry. It has not fundamentally changed our strength. It has not compromised our own capabilities. We have our specialized in corporate level agentic AI. We have done a lot of different OpenCloud kind of applications that we have in promoting OpenCloud type of business to the whole industries. So I think OpenCloud and the presence and popularity of it is a good sign. And the plans we have been talking to our customers will be proven right as now they have the contact staff OpenCloud because our ideas are now being kind of validated in the mainstream channels and also a lot of the big companies may provide a generic general for those AI agents, AI models that can do a lot of things, including writing poetry, et cetera, like we have launched -- Alibaba and talk they have their own like reversion of OpenSound and do it to Tencent or in power cohort, Microsoft may have targeted B2B sector with the general purpose AI. It's all work-related scenarios and those work-related scenarios, be it when it comes to medical communication or insurance, it does not matter as general purpose. It could be B2B or B2C, it is a specific use cases, but for Bairong, we have the industry expertise and know-how, we integrate our tools into the process of our company's operation, which our agents are giving a lot of authorization. So I think it's a very good sign because it will enable our silicon-based employees to more easily to navigate, it's more of a collaborative symbiotic relationship. So our legal agent called -- no organically integrated with into and the fast back -- so it's going to be embedded in the workspace agentic AI or AI agents, but the type of agents with -- not the same as the AI agents.

And another way, another important factor about access information, what type of employees have access to information. So for a large tech companies, if something can be addressed by comments works software, then that's basically an upgrade of Microsoft Office. So for Bairong, be it other source or close source or be it AI agents, it's not a directly competitive relationship. We have never thought about making those into their markets. Will they try to take our market? Well, I have 2 thoughts. As -- received an interview where he said, as large as the potential license application. And when you've asked why you don't want to specialize in this industry sub applications -- said, we don't have enough resources. It's not possible for us to have enough high-quality private domain data. So that's why we only want to focus on the model level. So -- not going to fundamentally exchange that judgment, which we have believed in over the last 2 years. So Bairong, for general purpose AI agents and also corporate AI agents, we need to build back on, which is our -- we have looked further in-house our -- is kind of a builder of corporate-level OpenCloud, and we will continue to use -- we are going to open up areas where we do have full expertise of -- we are going to do in -- agent developer. So basically, they will come in and develop their own specialized agents, and then we're going to do revenue share.

And just like Tencent. Tencent is traffic plus the resource allocation, we will do corporate infrastructure empowerment as technological empowerment. And in recent years, we're seeing more traditional RSV using Bairong to create their own agents. So that's my quick response.

Thank you very much. We'll give [indiscernible] from Gotham Securities.

This is [indiscernible] from Securities. I have a couple of questions. And just now we're very much agreeing on the churn investment of the industry. And my question has to do with our silicone-based employee strategy like what's our revenue expectation for silicon-based employee. And also you mentioned that -- different AI agents, what's the strength of our silicon-based employees? And as more companies provide AI agents, how will that affect our business going forward? And thirdly, AI cloud price increase, it has a direct impact on our cost. Will that affect the profit margin of our silicon-based employees.

Thank you, Mr. Lei. Let me just quickly repeat your question. First, you want to understand the progress of our silicon-based employees? And I'll to give the floor to Mr. Zhang Shaofeng.

Mr. Yang, we talked about this overview, I said it is kind of unique. We do not propose silicon-based employee today, let's say, slowing. We have been changing since day 1. We have always been delivering end-to-end solutions stacked as there was no silicon-based employees concept. But once you break down idea like our MaaS or BaaS, H2 business for us. So basically, H1, our MaaS is growing business, a mature business, and BaaS is more growing business, and H3 is like the incubation business growing fast. None of these businesses are tools per se. We have always been delivering results or we have been delivering silicon-based employees from day 1 to risk control and then approval and review. We help institutions use our tools review to review applications from financial customers. We can handle tens of thousands of applications. In the past, it was people handling just paper with the 10 and Bairong is the first one who for the industry. We have -- that if you converted the cost to -- the human employees, they can handle probably dozens of cases per day, and we have tens of thousands of tons of thousands of like this control agentic AI agents. So certainly the cost significantly. And also for our BaaS business, if you compare with traditional banks want to issue credit cards, they have to outsource business to on-the-ground companies basically hand out pamphlets RMB 200 to RMB 300 as commission [indiscernible] of how the division of labor is organized.

If we look on the guidance, peak performance -- at this peak, we had 200,000 silicon-based employees using AI cold calls and WeChat to the financial case -- and there's another third scenario which is for telecom companies, AI is helping carriers and telecom players to renew services for customers selling packages to customers. For example, if you are going to other province, you need to turn on roaming services. This kind of service has expanded quickly in scale to hundreds of thousands. We're seeing the hope at the end of the tunnel. Some areas going to slow, some are growing really fast at dozens of -- per year. And also for the -- like post the other tier centers. There has also been a lot of use cases. And now the question about AI agents. What's the competitive edge [indiscernible]?

Yes, that's my question. You are correct. So competitive edge of our own AI agents.

Well, the large tech companies, they don't want to develop too many corporate level applications, I think you're well aware, the product they launched today are kind of like upgraded version of [indiscernible], upgraded version of corporate WeChat or -- office scenarios. We don't want to compete with those kind of apps -- provide a corporate service or corporate agents. We are not really seeing any major corporate customers. But using roughly 200,000 silicon-based employees [indiscernible] if you look at companies where the revenue in that scale is more traditional software companies and the products they deliver our tools, they charge a project-based fee and then sell a license. Like our model, we only charge based on the workload completed and the number of positions filled by silicon-based employee. So it's definitely not comparable, and we are the largest in this sector. And we have always been dedicated to building silicon-based employees. So we have deep insights and know-how on there's a lot of pitfalls.

As [indiscernible] that traditional software companies, you want to branch out into this business, you have to walk through all the pitfalls that experienced we already. For example, if you use on for a dialogue interaction and then use that directly, but the result is not going to be as good as our performance because for Bairong, in addition to AI models, they map infrastructure, for example, soft exchange of communications.We noticed that when we do the voice agent, the simultaneous tendering capacity is not enough. With another generic -- communications soft exchange -- typically do not have it. We don't know what's lacking and they want to sell it or offer it to you. Even the ByteDance have it.

So the large tech companies, they don't know what are looking for the customers and for small and medium-sized companies, and they are not AI native. The first generation was decision-making and second generation generated AI previously also insourced and internalized, but now we're able to output. Many companies don't even realize that as well as topline as to your second question.

As for your third question, price increase or rather the cloud cost increase, how would that affect our margin for silicon-based employees? So definitely, that will lead to some price or cost increase. But for the cloud platforms to raise price, it could be a double-edged sword because many of our silicon-based employees underlying models are developed by ourselves. We are not using any third-party models. So compared to other suppliers, we are at an advantage because they rely more on the generic cloud providers. And also, we sometimes seek general purpose AI because users tend to ask a few very strange questions, which is considered a corner case that's where we allocate the assignment to general models. But overall, it's a good sign because we have -- capabilities for R&D. People are bullish on -- to [ Google ] because they have their own [indiscernible] is not affected by NVIDIA. I hope that answers your question.

That's all the questions I had.

Thank you so much, Mr. Yang from International. Next question? Next, we're going to give the floor to [indiscernible] from International.

I have a question about performance. We achieved double-digit growth in the first half of last year, and in the second half, there was the decline. [indiscernible] decline?

Second question. We mentioned AI investment and [indiscernible]. Could you please provide a breakdown of this investment? Have you done any tests -- stress tests? And will -- why that affected our gross margin so significantly? How will [indiscernible] affect our business going forward? And also revenue and profitability guidance for 2026?

Thank you, [indiscernible]. I'll give the floor to Mr. [indiscernible].

Thank you for the question. Documents officially took effect in October 2025. And as we all know, there were some policy shifts last November, which was now as document #9, which shocked caused by the policy in the financial industry is mainly because some of our clients took some products offline. And it takes time for their healthier, better, new products to be launched. But it takes time for them to get ready. And this time frame in the second half of 2025, it was a transitional period. That's why it affected our silicon-based employees marketing scale, and that's why it hits our revenue especially in the second half of the year. At the same time, the company's intensity of preemptive AI investment did not weaken compared to the first half. We aim to better seize the golden window in the next 2 years to rapidly accumulate first-mover advantages and expand the client scenarios as CEO and CFO mentioned, if we add back these preemptive AI investments, which shows that the company's mature business like MaaS and BaaS still maintain stable profitability. And the investors who have followed Bairong for a while will note that in 2020, affected by policy uncertainty, our revenue declined year-on-year for the first time by 10%. And in 2020, our revenue declined by 10%. Back then BaaS -- free to decline. But in 2021, the revenue rebounded strongly with a year-on-year growth rate as high as 23%. This demonstrates, on the one hand, the abundant cash flow in our industry and on the other, Bairong's high organizational resilience, capable of enduring cycles and supporting long-term business development.

In 2026, high-risk business will be largely cleared and we will experience a brief period of adjustments. However, if you focus on various low-risk businesses I mentioned earlier, such as AI agents for low interest rate credit marketing, wealth management and higher-performing asset disposal in countercyclical growth -- and is still possible to achieve considerable rapid growth. We are standing at a brand-new starting point.

Here's another way to look at it. Imagine us as the emerging AI agent structure but backed by over RMB 2 billion in cash, that combination of innovation and financial strength creates substantial long way for growth. That's my quick response to your questions. Thank you.

Thank you, Mr. Glen, and thank you to analysts from [indiscernible] International sector. Next question, please. From [indiscernible] , please

Am I audible?

Yes, you're audible.

Sounds great. So dear management, which is -- from SDIC International. My question is about our MaaS business that has some barriers in terms of technology and other barriers that want to look new business, [indiscernible] corporate data to treat the model against that backlog. We might be slightly different from general purpose AI or we're starting at the same starting line. But for the big tech companies, do you have any advantages in terms of cost and the time frame. There's a lot of small companies, these general purpose launch models will that create competition for us? Will that put us at a disparage?

Thank you, Jan. I'll give the floor to Mr. Shaofeng, Zhang Shaofeng.

As I touched upon something relevant to this question. I think it's a key question at -- through that it might cause problems. Even for a company like Anthropic, there [indiscernible] the application is the biggest usage use case is an application lab in platform. So thank you for the question for me to elaborate on this issue.

General purpose AI model are more about AGI or [indiscernible] need to know everything even like in poetry and answering questions. But if we think about it, how many customers are truly aiming for that kind of unique or only -- AI agents. We don't want to corporate AI to answer how to book a restaurant or give you a guide towards the destination. So it's still restricted to a relatively small scope of work. So if you use general purpose AI, if you look at the number of parameters, it's typically -- in size in terms of number of parameters. Even though there is the structure, it's not going to be as lean and efficient as our dedicated industry-specific AI.

So for our vertical industry AI, I want to limit to just 10 billion parameters. Our smallest model is only 1 billion in parameter. The advantage of that is it's much cheaper. Somebody asked a question about cost because if you don't upgrade the chips, the cost is going to be correlated with the square total offshore parameters and for DeepSeek -- DeepSeek did very well that there was a lot of hallucination, however the tolerance for hallucination in the corporate use case is much smaller compared to B2C products. So from a general pattern perspective, that's the first initial response I want to make. We don't want to use pure general purpose AI, but if there is a good specific vertical model that will be much more efficient.

And second, a lot of the knowledge is not available on the public website. It's credit to my knowledge. Let me give you a quite example. If you use DeepSeek and OpenAI and if you ask for payment collection, that's really hard to achieve that. It's almost impossible. The models silicon-based employees doing to pay and then the AI model would just say, okay, you take the rest, you find a job. They're not able to collect the payment for those who defaulted because the different data points to learn from this. Taking payment collection is just a simple example. Do you have data? It cannot even buy data because it has to do with user privacy. So it's called private domain data, dedicated specialized data. And without that use case, you're not able to train the model, but Bairong advantages that we have specialized model that is smaller in size, faster in response, lower in cost with lower in hallucination. At the same time, we have accumulated a lot of different use cases with ever growing pool of -- from the domain data. And even if we have the same number of parameters, we're going to deliver better effects. These small companies, can they do it themselves? Well, DeepSeek was launched last year, but how many companies have managed to build their own agents based on DeepSeek? It's where the requirements are higher is becoming more demanding, the overall level of requirement has been elevated. Now the technology might have improved this year, our competitor is upholding a higher bar, so their deliverable will be much more demand compared to last year. But if you are able to deliver what Deepseek give it to you. It does not matter because the competitors will be using something far better [indiscernible]. So it forces you to use more advanced technologies because of competition you face. Just now I said about AI agents in the cloud. Can you just develop a corporate cloud yourself? It's very hard even Meta has problems.

Thanks, Mr. [indiscernible] I will pass to the next question.

This is Robin from Bloomberg. You mentioned that this year, there will be a lot of touched on the investment in AI agents. Could you help provide a detail of this investment, including the subsidiary competition and the proportion of various expenditures such as R&D expenses -- procurement and talent agreement? What is the -- between expensing and capitalizing of AI-related investments? AI-related deliver cost in 2025 or '26 onetime investments are part of that continuous plan with the labor cost budget and high explanation plan with 2026?

Thank you, robin, from Bloomberg Intelligence. Mr. Calcin, I will answer the question. Last year, we added a lot of upfront AI expenditure. The total R&D expenses is CNY 637 million, again, an increase of 25%. The first half expense were CNY 302 million, increase of 33%; second year of the year, CNY 335 million, up 18%. As [indiscernible] other AI-related cash expenditure, you will see that it continues to grow over the year, throughout the year. Our [indiscernible] but if you take into consideration other AI-related cash expenditures such as servers and [indiscernible] investment in ecosystem development and integration than the overall increase investment for new business is even larger. Our largest R&D investment is in talent. In 2025, investment in AI R&D talent alone increased by more than RMB 100 million, up 24%. However, our return on talent investment is very high. As the Chairman mentioned earlier, 2026 will be a key year for the company's AI strategy to move from the capability building base to the value realization phase. Therefore, we will continue to emphasize optimization and upgrading of AI talent -- speaking, our investment in AI talent is highly worthwhile.

Everyone can calculate Bairong's productivity. Our revenue curve employee was already close to RMB 2 million in 2023, exceeded RMB 2 million in 2024, while on the average salary over the past few years has been announced at RMB 500,000 to RMB 600,000. The return on talent investment is as high as 3x. The return on R&D investment is also very high. Currently, most mainstream AI companies have not yet reached the stage of focusing on profitability. Bairong is one of the few companies that is truly doing AI and has real profit and sustained real profits. Since we have decided to do AI and aim to become the Google or agentic AI, we must have the determination of the investment.

Some companies in Hong Kong has suffered a loss of more than CNY 6.2 billion over the past 3 years at some AI companies reported losses of more than CNY 2.6 billion in the first 3 quarters of last year. Anthropic requires USD 50 billion if we single data entry and some rest [ 2.3 billion ] chip AI and improving systems over the next 5 years. In comparison, Bairong investment is highly cost effective because we are more practical and more -- and more results oriented, which enables us to deliver ahead of others. Thank you for your question.

Thank you. Thank you, Robin. In the interest of time, this concludes our Q&A section. Thank you all for your participation, and thank you management for responses, and thank you all for participating in our results conference for Bairong's 2025 annual performance. If you have any questions, please visit the company's IR website at ir.brgroup.com or email [email protected]. Also, you are welcome to scan the QR code to follow our digital IR assistant Bairong. This IR assistant Bairong can interact with you directly and regularly release the company's business news and updates. We look forward to meeting you and conversing with you again at our next meeting. Thank you.

[Statements in English on this transcript were spoken by an interpreter present on the live call.]