Earnings Call: H1 2025

Aug 29, 2025

Summary

H1 revenue rose 22% to CNY 1.612bn, with 73% gross margin and CNY 254m adjusted net profit. BaaS cloud and MaaS grew, while insurance regulation and telecom restrictions pose H2 headwinds; guidance remains double-digit growth.

Dear investors, analysts, and media friends, good morning. I am Sammy Ching, Executive Director and Director of IR at Bairong. I would like to thank all the investors online for your attention to Bairong. The company released its 2025 interim results yesterday, August 28, and will report and share the company's 2025 interim operating results, as well as outlook for the second half of 2025, and answer your questions. The presentation will be divided into three parts. Our CEO will introduce the business progress and outlook, followed by our CFO, who will interpret the financial performance, and finally, the Q&A session. Online participants can enter their questions into the question bar of the live broadcast interface. We will call the following phone numbers to answer questions. 023-6362-333 or 4008-063-263 in mainland or in Hong Kong, 3018-3602 or 961505, and internationally, 8323-6273-7100. The password for the Chinese line is 320556788, and the English line is 649518155. Management will be happy to answer your questions. These results announcement contain forward-looking statements that reflect the company's current beliefs and expectations regarding the future, and as their words suggest, anticipate, believe, intend, estimate, expect, and words of similar meaning. All statements other than statements of historical fact contained in this meeting are forward-looking statements. These forward-looking statements reflect the views of the company, management as of the days hereof and are not guarantees of future performance. Neither the company, any member of the group, nor any of its relevant affiliates or any of their respective directors, officers, employees, consultants, or representatives undertakes any obligation and expressly disclaims any obligation or undertaking to disseminate any forward-looking statements that are updated or revised. Next, I would like to introduce the company's management team at the announcement meeting today. First, Mr. Zhang Shaofeng, Founder, Chairman of the Board, and CEO of Bairong. Mr. Zheng Wei, Executive Director, Senior VP, Chief Financial Officer, and Head of BaaS Insurance Scenario, Bairong. Mr. Duan Ying, Senior VP of Bairong and Head of MaaS and BaaS Financial Scenario. Now, I would like to invite Mr. Zhang Shaofeng, the founder of the company, to report on the business progress of Bairong in the first half of 2025. The floor is yours. Distinguished shareholders, investors, analysts, and friends from the media, good morning. Welcome to the 2025 interim performance announcement of Bairong. It is a great honor for me to report to you on Bairong's performance progress and to share with you our views on long-term development. Let me briefly introduce Bairong first. Our business primarily utilizes AI technology to assist thousands of business and institutions in two main focus areas. First, we help them manage risks, whether it be individual or business customers in terms of their risk profile and risk preference. So this is a part of the KYC business that we offer so that our users can better understand what their customers are like and what they are doing. Secondly, we provide generative AI services to help the user better market and engage with customers, which will increase customer satisfaction. So by leveraging AI modeling to aid intelligent decision-making, known as a Model as a Service or MaaS, we are currently able to handle a daily average of over 300 million calls. Additionally, through generative AI and VoiceGPT to support intelligent marketing and operations known as Business as a Service for BaaS. We have also made significant progress in both technology and application. We built a first-class platform of infrastructure. To date, we have accumulated over 8,000 institutional clients. We have developed AI products such as digital human all-in-one machine called AvatarGPT and the CybotStar AI agent platform that can help institutions solve practical business problems. Our institutional clients include household names such as Alibaba, Baidu, ByteDance, Meituan, Xiaomi, JD, as well as financial institutions such as ICBC, Bank of Communications, Ping An, Taikang, among others. Some of them are major clients that contribute tens of millions of CNY revenue to the company. Next, I'd like to go over our business performance. The overall increase of household loans remained around the same level as last year, and we were able to keep NPL ratio at a stable level. We navigated smoothly through last year's minor downturn, and we've achieved robust growth in the first half of 2025. MaaS revenue grew 19% year-over-year, accelerating by 8% compared to the second half of last year, with the core customer retention rate further increasing to 98%. Our BaaS revenue showed impressive growth, thanks to the progress in generative AI technology, and the revenue of BaaS surged by 45% year-over-year. Gross margin has consistently stayed at a high level of 73%, demonstrating that our investment in AI has yielded advantages and scale effects. Net profit for first half 2025 exceeded 200 million CNY, with both net profit margin and adjusted net profit margin returning to double digit levels at 12% and 16%. We could not have done it without the enduring support, trust, and partnership of our shareholders, investors, and analysts. On behalf of Bairong, I extend my heartfelt gratitude to you all. Bairong is at a historic opportunity as artificial intelligence unlocks immense business value across a variety of industries. We believe all the industries are going to be transformed by AI, technology such as LLM and generative AI. We're in the middle of that transformation, and we fully embraced AI agents since last year. We've been an AI native company, but last year we firmly positioned ourselves in generative AI and LLM. We also wanted to explore not only the financial industry, but also the broader industries out there. We have also evolved our business model from traditional software sales to digital employee employment. In other words, our digital employees are able to help you write documents, help you do research, and essentially, our clients can pay for those digital employees, be it in the form of wages or in other payment terms. This allows our clients to move from CapEx to OpEx, and Chinese clients usually prefer OpEx, which offers greater flexibility. Our product strategy prioritizes plug and play. In other words, our clients can use our products right out of the box with a minimal amount of tooling and interface tweaks. We also allow our clients to train their own AI agents. We allow our users to create synergy between their human employees and their AI agents. So we're lending our underlying capabilities to the client, not just the agents themselves. We launched our own proprietary system with our knowhow and our API interface that's able to help our clients continue to involve their use cases with the AI agents and AI capabilities. We have made great progress there. As highlighted in JP Morgan's first half 2025 report, China AI, the sleeping giant awakes. Among China's top 60 AI companies, Bairong is the only company that was mentioned in the financial vertical. This year, we will continue to fully embrace AI. That's our top priority. Our resources continue to be concentrated on generative AI and decision-making AI. Our clients from big techs to startups will all benefit from our faster iteration process from the lab to scaled commercialization. We've established an AI bootcamp with weapons and ammunitions. The weapon in this case will be the infrastructure, and the ammunition here will be the tech stack that we offer. We've also developed algorithms with our own team to help accelerate commercialization from the lab. We continue to improve AI native adoption at the client end, and ultimately we want our clients all become AI native institutions. We expect to have a kilo-card where over 1,000 GPU clusters will continue to improve our computing capabilities. The ammunition, which is our tech stack, includes NLP, LLM, privacy computing, and AI agents. We focus on the orchestration of AI agents to complete real life complex tasks. We are firmly established in AI agent application in the verticals that we operate. We will pursue both depth and breadth. We will double down on finance, including banking and insurance, to utilize AI to enhance operational granularity and risk control precision. In terms of breadth, we will expand into non-financial sectors, and we hope our AI infrastructure, AI capabilities, can empower other industries as well to help them improve their productivity. We will also lend our capabilities in terms of privacy computing, and generative AI capabilities to our non-financial clients to maximize value for them. Overall, in the first half of 2025, we saw an acceleration along our strategic path, a more robust tech foundation, higher talent density, and broader scenario replication. AI has emerged as the core engine powering Bairong's growth. I also want to update you on the development of our enterprise-grade agent platform, CybotStar. Its mission is unequivocal: to empower enterprises across diverse industries to effectively utilize the benefit from sophisticated large model technologies, solving real-world business challenges. CybotStar was thoughtfully designed for two primary scenarios. It's based on the bottom underlying layer to provide top-layer, workflow-oriented instructions. The design is very considerate because it takes into account two primary business scenarios. Internally, it is referred to as the EX, employee experience. EX serves as a powerful tool for employees, enabling intelligent training for customer service, effectively conducting customer background checks, ensuring legal and regulatory compliance, protecting consumer rights, optimizing HR workflows. We want silicon-based assistance for all employees at our company. Externally, it acts as a catalyst for superior customer experience. We refer to as CX, customer experience. Delivering automated customer service responses, intelligent marketing recommendations, timely repayment reminders, proactive customer follow-ups, personalized care service, and convenient self-service options, and reminders of birthday congratulations, all enhancing customer engagement and satisfaction. Results are the ultimate validation. I'm thrilled to announce that in the first half of 2025, CybotStar accelerated significantly on its commercialization journey. We successfully signed formal cooperation agreements with several large. We have signed agreements with several large and medium-sized institutional clients. A particularly noteworthy achievement is our breakthrough in combining proactive large model capable of independent thinking, decision-making, and action with emotional, expressive text-to-speech technology deployed at scale. This intelligent solution has demonstrably outperformed traditional manual benchmarks in traded scenarios, showcasing AI's immense potential in handling complex interpersonal interactions. Generative AI in combination with decision-making AI is very competitive, and continuous technological innovation drives our progress. To make the platform's brain learn faster and smarter, we introduced the powerful GRPO reinforcement learning framework. This year, you might have noticed that DeepSeek adopts this as one of the underlying technologies. This acts as a super coach. GRPO now trains our proactive large language models for more efficient experiential learning and self-optimization. It is also applied to automatically optimize the effectiveness of RAG, Retrieval Augmented Generation, dramatically accelerating model iteration and enabling us to respond faster to new clients' demand and enhance existing applications. At the same time, we recognize that powerful computing infrastructure is fundamental. Consequently, the first phase of the company's purpose-built high-performance data center is now fully operational, planned with up to 1,500 high-end GPU cards. It is a very smart system of infrastructure. It can smartly coordinate amongst all the GPU cards. To maximize computational resource flexibility and efficiency, we launched a fully self-developed heterogeneous computing cluster. The system features an intelligent orchestrator capable of seamless integration and virtualization management for diverse GPUs, including both domestic brands and NVIDIA products, regardless of size. Crucially, the cluster is deeply optimized for our platform's core model technologies, including large language models, TTS models, and multimodal models. This significant infrastructure investment lays an exceptional solid computational foundation for the large-scale deployment of intelligent applications across business scenarios. In the first half of 2025, Bairong's AI capabilities penetrated even deeper, leveraging our intelligent voice product and robust CybotStar enterprise agent platform. We embedded AI capabilities like integrated circuits into the core business processes of key high-potential sectors. We provide benchmark customers in different sectors with innovative experiences and applications, such as enabling customer self-service and precise recommendation via AI assistance, delivering highly natural AI voice interactions, implementing personalized AI-driven marketing outreach, and our customers will not be able to tell whether it is a real human or a robot interacting with them. It can be used for various scenarios, whether it is marketing, consulting, or just a regular interaction with customers. We are integrating AI with customers' core business logic to create transformative services. All of this provides customers with comprehensive cutting-edge AI experience. By persistently cultivating these high-value scenarios, we are further cementing our leadership position in the commercialization of AI across diverse industries. The fifth part I want to present to you are the awards we have received for our generative AI and large model. We have widespread recognition in the first half of 2025. In the first half, based on Bairong's large language model and enterprise-level agentic model, we have already filed successfully with the Cyberspace Administration of China, CAC, earning authoritative validation for its precision. At the same time, following that, Bairong self-developed large model, BR-LLM, also completed its official registration. The dual registration demonstrates that Bairong has fully established end-to-end AI capabilities covering large model training, intelligent agent development, and enterprise scenario deployment, meeting stringent standards for customer-facing service in content generation control and compliance management. As of now, Bairong has formed a comprehensive AI product matrix centered on the base model BR-LLM, intelligent voice interaction models, and code model BRCoder. BR-LLM and BRCoder and Digital Human Avatar, demonstrating formidable capabilities in multi-turn dialog, content creation, automated code generation, and tool invocation. We have demonstrated significant capabilities. Recognized for having the largest and deepest large language application practices in the financial sector, Bairong was named by KPMG as one of the winners on the Dual 50 list for the second consecutive year. Furthermore, at the 2025 Innovation and Development Forum co-hosted by Boston Consulting, the Chinese Academy of Sciences, Internet Weekly, and eNet Research Institute on this specific forum, Bairong was honored with the 2025 AI Large Model Innovative Application Award in the artificial intelligence field for its emotionally interactive robot. This prestigious award recognizes Bairong's outstanding achievements in AI and its significant contribution to advancing the industry and leading technological innovation. This marks the second consecutive year Bairong has received this award. As of May 2025, the company holds 461 patents and software copyrights covering artificial intelligence, machine learning, privacy computing, human computing collaboration, and multimodal technologies. Sixth, as a leading data-driven enterprise, we are committed to our social responsibilities. Bairong continues to actively and consistently fulfill its corporate social responsibilities. We previously introduced our intelligent voice interaction all-in-one device, Da Sheng, designed for adolescent emotional companionship. This is used to help them foster emotional and mental robustness and to prevent tragedies from happening. We brought the mythical Monkey King into reality as a companion who listens attentively and offers comforting understanding. Powered by Bairong's self-developed adolescent emotional companionship large model, Da Sheng can answer imaginative questions like "How big is the universe?" While also detecting subtle cues of sadness or frustration in a child's voice within 0.3 seconds, responding with genuine warmth instead of the coldness of a textbook. Da Sheng is now currently deployed in schools in Hebei and Shenzhen, and before the end of this September, more Da Sheng units will be deployed across many schools in China. We are also showcasing this product, the Bairong AI Digital Human Avatar, at the China Welfare Institute Children's Palace, serving as an escort and safeguard for children's mental health and development. Utilizing generative AI and affective computing technologies, this Digital Human Avatar senses emotions in real time, adjusts its conversational strategy accordingly, and provides professional emotional wellness services, safeguarding adolescents' mental health. Beyond companionship, we strive for early detection. We hope that we'll be able to look out for the children and help them grow in a healthy and happy manner. Seventh, I want to share with you that in the longer term, we are very positive on Bairong's prospects, especially with regard to the application of agentic AI. But in the short term, there are some headwinds we expect to persist. Some people have asked us, for example, in May, Circular No. 9 published by the NFRA, how has that affected us? I want to talk to you directly about the NFRA article. As we all know, as you can see on the financial statement, insurance industry has come under pressure as a result of regulatory changes. A lot of the older traditional products are now removed from the shelf. At the same time, regulators are further limiting the commission rate of insurance sales. Therefore, our BaaS Insurance service has come under continuous pressure, and we expect the pressure to continue in the next two years. That's the first point. Second, many people have also asked us about the Circular No. 9 published by the National Financial Regulatory Administration, which might limit the sales of credit products by financial institutions. In the second half of this year, we expect more financial institutions specializing in credit to be subject to that change. Their business will shrink as a result of the new regulatory policies. I would say it is a big hit to the relevant companies. Our MaaS and BaaS products are both affected by this new regulatory shifts. Thirdly, VoiceGPT. It is based on our AI algorithm, but at the same time, it is also based on telecom providers' phone connections. There are two events that affected the supply of the connection. One is the parade for the victory of the Chinese people's war of resistance against Japanese aggression. Basically, in order to reduce complaints by residents, the telecom players have decided to limit phone connections. Secondly, due to the telecom scams happening in Cambodia, they are stepping up the campaign against such fraudulent behaviors, further limiting the phone connection we have access to. Altogether, these three events, insurance product, credit product limitation, and telecom connection weakness. These three things combined are going to pose some challenges to our performance in the second half. We think there is some impact on our business, and we also expect our insurance to be under pressure in the next few years. As regulators continue to monitor the growth of the economy, the strength of the economy, there might be regulatory changes, but we have a wait-and-see approach when it comes to those changes. In the mid to long term, we will continue to explore non-financial related AI agent business. In the second half of the year, I think there is indeed some pressure on our short-term performance, but we believe as AI adoption continues to increase across industries, we are still optimistic for the long term. Bairong is a long-termist, and short-term investment is for long-term returns. We have strong belief in AI and its potential to service all the businesses and individuals. We will continue to develop AI agents in terms of AI platforms as well as AI applications. We believe that AI is able to help improve productivity and reduce costs. Our decision-making AI and generative AI models will be able to help us support millions of businesses as they build their own AI platforms, AI agents, and their adoption. We will continue to make progress in our journey to bring intelligence to individuals and businesses, and as a leader in the age of AI. We hope to be on this journey with investors who share the same long-term vision. Thank you so much. That is all from my presentation. Great. Thank you so much, Mr. Zhang, for your presentation. I will now hand it over to Mr. Zheng Wei, our CFO and BaaS scenario VP, Mr. Zheng Wei, for financial highlights. Thank you. Dear investors, shareholders, analysts, and friends from the media, I will now go over Bairong's financial highlights. In the first half of 2025, the company achieved stable business performance financial results. Revenue in the first half of the year reached CNY 1.612 billion, an increase of 22% compared with CNY 1.321 billion in the same period last year. In a challenging year, Bairong successfully achieved stable revenue growth and maintained continuous profitability, reflecting the resilience and market competitiveness of the business model. The gross profit of the company reached CNY 1.182 billion, an increase of 22% compared with CNY 967 million in the same period last year. The gross profit margin remained stable at 73%, mainly benefiting from the scalable AI cloud business model. The operating profit and net profit both reached CNY 201 million, and the margins are at 12%. Net profit of non-IFRS reached CNY 254 million, and the net profit margin of non-IFRS reached 16%. Non-IFRS EBITDA reached CNY 283 million and EBITDA profit rate 18%. We continue to optimize costs and operating efficiency. In the first half of the year, we saw 23% in our BaaS business driven by generative AI to CNY 1.11 billion accounting for 69% of the total revenue. In the BaaS business, cloud revenue increased by 45% year-over-year to CNY 857 million, accounting for 77% of BaaS and 53% of total revenue. The insurance industry responded to regulatory changes, and revenue declined by 19% year-over-year to CNY 253 million, accounting for 23% of BaaS and 16% of total revenue. Our BaaS business MaaS revenue increased by 19% year-over-year to CNY 502 million, accounting for 31% of total revenue. The revenue structure shows the pattern of AI at BaaS at the core and stable MaaS expectations. As you know, our MaaS business aims to assist institutional customers in making decisions and diversifying application scenarios by models and result evaluation. In the first half of 2025, our MaaS business revenue reached CNY 502 million, with an annual contribution of our core customers, each about CNY 300,000 and more. We continue to support the productivity of verticals. We are also able to achieve repeat application and scalability in insurance, banking, wealth management, as well as other financial sectors. We are able to leverage our BaaS services to support retail credit, small and medium-sized credit, as well as e-commerce, among other scenarios. In the first half of 2025, our BaaS financial cloud business increased by 45% to CNY 857 million, showing our product market fit and customer penetration continued to increase. Through Liming Insurance Agency, we are able to operate across China. Our decision-making AI offers customer insights as well as a precision recommendation of insurance products. We charge a commission based on a premium received. In the first half of 2025, our BaaS insurance cloud income was CNY 253 million. Specifically, new policy revenue contribution decreased about 20% to CNY 204 million. Life insurance premium renewal continued at 90%, and renewal contribution was CNY 49 million, down 12%. Total premium written was CNY 3.1 billion, up 9%, which shows our resilience and recovery in this segment. In the first half of 2025, our GP margin stayed stable at 73%, benefiting from scalable AI cloud model as well as scalability potential. Our R&D expenses reached CNY 302 million, up 33% compared with CNY 226 million last year, or 19% of revenue. We continued to invest in AIGC and LLMs to support our business development. During the reporting period, we put into scalable application LLM and multi-motion voice synthesis models, which has created a business value. We have also improved iteration speed and stability with our reinforced learning framework as well as active model training and RAG. We have also brought online a variety of computing power initiatives, including the first phase of our high-performance computing center. We have laid a solid foundation for greater scale multi-scenario deployment. Our selling and marketing expenses reached CNY 606 million, flat over last year in terms of percentage of income at 38%. SG&A CNY 140 million, down at about 9% of revenue. Net profit CNY 201 million, up 41%, and net profit margin 12% on an IFRS. Net profit margin CNY 254 million at 16% margin. We will continue to keep our long-term competitiveness with investment and core technology progress, with focus on marketing expense control as well as marketing efficiency improvement for greater return. In the first half of 2025, our cash and cash equivalent was CNY 3.7 billion, up 2%, thanks to greater efficiency of use of capital, which includes cash and equivalent, our deposit as well as our current financial assets with fair value recognized in P&L. We are committed to long-term vision and shareholder return. We bought back 3.27 million of shares or HKD 26 million in the first half of 2025. We are one of the few companies that has achieved profitability with generative AI, and we will continue to serve our customers with AI empowerment. Thank you so much. This is great. Thank you so much for your presentation, Mr. Zheng Wei. We now open the line for questions. For those of you who are joining from phone, please press star one to get in the queue and then press one to unmute. For those of you who are joining online, please use the chat box to input your questions. Please begin your questions by stating your name and your affiliation. Thank you. Great. Our first question is from Yang Chu from Morgan Stanley. Thank you so much, management, for giving me the opportunity. My name is Yang Chu from Morgan Stanley. I cover the financial sector. My question is, in digital human and agent space, there is a lot of competition. Your new products, AvatarGPT and CybotStar have done well, but what about the number of customers and orders in hand? Also, just wondering, what is the charging model for CybotStar? Is it by tasks or by revenue split? As Shaofeng mentioned, about 25%-30% of codes are already generated by AI internally at Bairong. Could you please elaborate on that and how much cost has been saved? Also, if we can get some color on new product customer expansion as well as scenarios that you think will have great potential. Yeah. Thank you for your question. CybotStar or Baigong, as it is known in Chinese, is our agent platform. It's actually a reference to the CybotStar. In other words, CybotStar is able to generate bots or smart agents. In other words, it's a factory or a platform that is able to create agents for users. CybotStar has not only helped Bairong internally, but we've also been helping our clients with ESG application, with financial leasing, as well as other use cases. We've been in contact with close to 100 clients, and we've already signed deals with a number of them. Revenue contribution so far is limited because there is, I think, a process that we need to complete because, again, we are empowering our users rather than just selling software. That's where we innovate. But obviously, you don't need to pay salary to those digital employees. But I think there's a combination or a variety of pay models that we can explore, including salary plus bonus or just bonus. We're now exploring the market, and we're educating our clients, and we're helping them convert from just purchasing software to paying for a job done by agents. We're not giving the top priority to revenue today. It's about market share. I want more companies to start using AI agents as workers, as employees. I think this kind of precursor to future growth trajectory is actually more important than revenue per se in this day and age. Our agent builder is based on a large language model, which is comprised of audio, video, and texts. So it's called base models. It's not just a limited speech text model. It has memory capabilities, execution, planning capabilities. All these dimensions combined form the agentic AI, and it has the base models down below. On top of that, it has the ability to evolve and improve itself constantly. It's just like a human. It's a digital employee in a bona fide sense. So digital employee and so-called silicon-based and carbon-based employees as a comparison. Those AI agents, they are interacting with each other more smoothly so that they are able to coordinate with each other to handle more complicated, sophisticated issues that used to be the limited preview of human employees. To give you an example, to support a company in terms of loan issuance, you have to judge whether this company is going to use the proceeds of the loan to a green cause. Then in the documents they submit, there might be a lot of unstructured data, like images. It's highly unstructured. In the past, we relied on human employees to write reports to analyze the data and documents, and then giving them a score in terms of how green their investment is, their demand for loan is. But today, the agentic AI is able to handle all sorts of unstructured data in a way that humans do, and then they will be able to come up with an evaluation report and give you a final conclusion about how green the project is, whether it's, say, X out of five or how many scores, how many points, 100 grade system or points grading system. That kind of agentic AI is now supporting Huaxia Bank across 1,100 branches of the bank. It has boosted the management and operational efficiency by 33%. Another area is wealth management. It's kind of in a high-end area where you have to talk to the customers and better understand their needs to see what kind of financial assets can match their demands. Right now, we're seeing better matching through AI agents, better matching between customer needs and financial wealth management products. This area upgraded from human employee only to an area where there is AI empowerment. More and more use cases will be unlocked. As to whether there is going to be significant short-term profit, I don't think any quick profits will come that easily because it takes time. It's still a work in progress. Companies' willingness to pay needs to be educated. As we all know, in China, software market is not that successful, or it's a big failure in a sense. It can be said that China does not have the kind of software industry as traditionally defined in the U.S. There is only software human resource outsourcing industry. In our understanding, we want to dispatch AI agentic agents. It's like a fixed income plus bonus, or if it's just a bonus, basically, we need to figure out a way for companies to pay for AI employees. It requires the holistic shift of mindset by companies. They need to allocate budget from the CapEx basket to the OpEx basket. That is why this year we are not setting any specific revenue targets. We're just hoping more and more people will start adopting Bairong's agentic AI to address their work scenarios, to handle their workload. That's my quick response to your question. Thank you, Shaofeng. Next question. Next, we're going to give the floor to Wuma Henxu from CITIC. Thank you, Mr. Zhang. My question is that the company's cost and fee expenses. At the beginning of the year, we noticed that the semi-annual revenue, we exceeded the target for semi-annual revenue target. At the beginning of the year, you also said at an earnings call that the company is going to invest heavily into new AI business. By rough estimation, total investment in 2025 for new AI business is going to exceed, or it's going to be roughly approximately CNY 300 million. In looking at the interim report, R&D increased by CNY 70 million, sales expenses increased by CNY 100 million. These two together, it's about CNY 150 million of new investment, which is in line with the CNY 300 million number proposed at the beginning of the year. Regarding the investment rhythm in the second half, do you think you are going to step up or do you think it's going to accelerate or decelerate? What are the directions of cost control and fee control? Also, if you look at different business lines, could you give us an annual revenue growth target trajectory by business lines? Thank you so much, Mr. Wuma. I'm going to give the floor to the CFO. Thank you, Mr. Wu. As you can see in the first half, our performance. In the first half of 2025, Bairong continued to counter external uncertainties with its own certainty, steadfastly executing its AI strategy to drive balanced growth in both quality and speed. By continuously strengthening capabilities in AI, R&D, application, and commercialization, the company has achieved rapid revenue and profit growth, demonstrating robust resilience and sustainable growth momentum. We will pay attention to all these areas, revenue as well as profit, in order to deliver more value to shareholders. With regard to the specificity of your questions, in the first half, the growth is driven by the commercialization of AI products. We are building standardized AI capabilities configured to specific use case of scenarios to turn it into scalable products, shortening the cycle from POC to scaling. At the same time, due to the engineering systemic model, we have strengthened the whole chain from data to model training to deployment, which has helped to accelerate the iteration as well as boosting stability. This has significantly cut our delivery and operational and maintenance cost. The unit economics has more room for improvement as a result. With regard to financial and generic customers' use cases, we are continuously building the knowledge pool of our industry space, and we are also improving the interpretability and controllability of our models so as to strengthen accuracy and consistency, so as to establish better business outcomes and better reputation among customers, so as to drive cross-selling and repeat purchase. This will in turn drive the positive growth for both revenue and target. In the second half of the year, I believe we will be able to continue to maintain that target for double-digit growth for revenue and profit. At the same time, we are going to invest more in R&D to respond to the pressure in the market. Externally, we face certain macroeconomic stress, and we have to strengthen our endogenous capabilities of quality growth so as to be more anti-cyclical. What we want to emphasize is that the investment into training has to be more focused on our strategic focus and dynamically adjusted. Our resources will be prioritized to match AI infrastructure construction, and we are going to do more to invest into high potential business areas. Based on market feedback, we are going to adjust our approach in light of market trends and cycles. At the same time, based on the threshold we have designed, we will ensure our cash flow security and safety. We are not going to sacrifice long-term strategy for short-term fluctuations. Overall, we want to leverage intelligence to drive efficiency instead of simply cutting costs. Just like our Chairman has said, we are going to increase the adoption of so-called silicon-based employees to improve our internal operational efficiency. For example, by applying agentic AI to all internal processes, we will be able to increase our management and operational efficiency so as to accelerate the implementation of our products, which may help to further scale our revenue and growth. Based on the investment we made in the first half, second half number might be higher. However, for the specific number, we will dynamically adjust in light of market realities. Overall, we will continue to take AI as the key engine for growth, and we will continue to pay attention to AI as our deepest moat. Long-termism as the guidance and engineering-oriented implementation and our deep insights into industry, all these things combined will help us deliver high-quality, sustainable growth. Then we will be able to deliver better revenue and profit numbers to shareholders. Thank you so much. Please read out the next question. Meeting assistant, please go ahead to the next question. Next, we have Jin Yujie from CICC. Thank you, management, for the opportunity to raise this question. This is Jin Yujie from CICC. I have a question about competitive advantages. I want to ask the management, as major tech companies increasingly leverage general-purpose large models to penetrate vertical sectors, how can financial institutions maintain their model advantage in core fields like finance and insurance? What specific quantitative metrics demonstrate the competitive edge of proprietary models over solutions refined from general models, particularly regarding iteration speed, training cost, and performance metrics? Thank you. Thank you, Mr. Jin from CICC. For this question, I will give the floor to Mr. Zhang Shaofeng, the CEO of the company. All right. When models first came out, everyone was paying attention to generic models, general-purpose models. They have some pre-training, they have some knowledge base. It is very generic, however. Whether there is truly an omnipresent, omnipotent general-purpose model, just like the presence of God, I think after two years of trials and tribulation, at least here in China or maybe in U.S. or other countries, there is no company who is able to claim that there is an omnipotent general-purpose AI model that excel in all areas. If you look at OpenAI, they are pretty strong in terms of reputation, but their coding capability is weaker than Claude from Anthropic. For DeepSeek, there is the hallucination problem, and the speed is rather low. Every general-purpose model has its own fair share of weaknesses. For us, we are developing our own model. At the same time, we are doing weekly horizontal comparison with other models regularly. So far, we have not noticed any large model who is able to excel on all fronts. Even for companies that specialize in general-purpose model, they know their models are specialized in one way or another. Whether it is for a specific vertical or segment, what kind of advantages do they have? Industry-specific models, what is their advantage? Once you train this model specific to the industry, it can reduce hallucination because it has been strengthened and reinforced to that specific industry, so it can correct errors. Second, the scale of parameters will be smaller and more manageable. One advantage of that is going to be faster in response. Second, the operational cost will be much lower in terms of the operational speed, operational cost. It is all correlated to the square, the total of the parameter scale. In other words, the model doesn't need to know everything. Because, if you really want to make an LLM that is omniscient, then I think it would be very challenging for them to have the right amount of depth for any verticals. The other thing is, the LLMs are being used inside institutions. I think institutional clients usually are reluctant to lend their knowledge to LLM for training or vertical specific in-depth training. That, I believe, is also making it difficult for LLMs to work well. Therefore, we believe we are going to have more proprietary or vertical LLMs. There are some corner cases where the vertical LLMs probably were not trained on, were trained for. Those corner cases can be perhaps managed by generic LLMs, because generic LLMs can have more versatile answers, even though the quality of those answers might not be the best. I think perhaps you are going to see a combo of these things. We are going to continue to invest in both paths, including proprietary or vertical LLMs, as well as generic LLMs. In particular, our CybotStar is more of a vertical LLM that has deep knowledge of the internal workings of institutional clients, less hallucination, and making sure that the answers are ethical and compliant. In terms of large scale LLM or AGI, that is not the focus of our business. That is my short answer to your question. Thank you. That is great. Thank you. Next question, please. Sure. Next question is from Minsheng Securities, Zhang Haifeng, please. The floor is yours. Thank you, management, for the opportunity. My name is Zhang Haifeng from Minsheng Securities. My question is, again, regarding the Circular No. 9 policy, or Circular No. 9 document. Just wondering what the impact is going to be long-term from that piece of document. Do you expect the document to be executed or implemented to the letter, or do you think there is some flexibility? What is the company's plans in response to this particular policy? As mentioned by management, regulators are very much keen on cracking down on telephone frauds. Just wondering, what is going to be the implications, if you can elaborate on the phone line access that you have. Sure. The Circular No. 9 document is essentially about regulating commercial banks' internet loan facilitation business with a focus on risk control and a focus on transparency of fees, as well as arbitrage of different interest rates. The overall financing cost needs to consider also platform service fees as well as credit enhancement service fees, among others. The loan total cost needs to be controlled within 24%, which is legally protected by the court. We believe this is going to be a new chapter, which will make the loan facilitation business more compliant and to develop new profitability scenarios. These policies are focusing on institutional clients that Bairong serves, including commercial banks, internet fintech companies, and consumer finance companies. Bairong is an independent technical service provider. It puts compliance first, and it is already adjusted its compliance and business mix in response to the document. We are working closely with clients in this transition period. We are tracking very closely the progress that our clients have been making, and we are working closely with clients to continue to grow their business within the regulatory framework. Moving forward, we will continue to grow with clients under the new framework, and we think that is going to be a place where we can continue to offer value to the table. In terms of the crackdown on phone scams, I think the challenge for us is mostly on the higher cost of telephone lines, which could limit high frequency call-outs and put pressure on scalability. For the company, we will optimize our phone lines and to access the phone book that is approved or number segments that are approved by the telecom operators, and we will also monitor the health of those phone lines with also whitelist system. We will continue to upgrade our technologies with AI optimized scripts and also privacy and data security measures to upgrade our compliance capabilities. We will grow together with our clients in light of the new changes in the regulatory environment. Our next question is from Zhongtai Securities. Thank you, management. My name is John Dunn. I am from Zhongtai Securities. Congratulations on the solid performance in the first half of the year. My question is mostly on BaaS Insurance. I think in the first half of the year, your numbers were under pressure because of the requirement to align sales practices with regulatory filings. I am wondering if commission has stabilized or is there still some downside? Would commission impact proactiveness of agents? How do we work closer with agents to not only scale up premium, but also use AI capabilities to improve insurer stickiness? Thank you for your question. I wanted to emphasize that the impact is not only from the particular policy you mentioned, which requires alignment of sales practices with regulatory filings. There is impact from that policy, but I think more importantly, it is about the economic transition that China is in that is putting pressure on growth. As you know, the insurance industry relies on loading plus and mortality surplus among other surpluses. I think the focus of this round of regulatory impact has mostly been on the fee surplus and the interest surplus. The idea is to guide the industry for high-quality growth moving forward. I think, again, it is not just aligning practices with regulatory filings. It is also about reducing interests to reduce the pressure of the liability side of insurers. So that insurers can minimize their risk in an environment where interest rates are decreasing, so that they do not get exposed to interest loss. I think if you look at overseas experience, including in Japan, I think the fact that insurers have inelastic payment where liability, some of them were exposed to interest loss risks in an environment where the economy was slowing. So the regulator is moving ahead of time to protect insurers from such future losses or risks. That is the first thing. Secondly, because the assumed interest rates continue to get reduced, I think the regulator is also guiding the insurance sector towards a future that is more focused on protection rather than return on investment. Going back to your question regarding the alignment, I think that policy has been in focus for some time now. But in April this year, the regulator issued a new notice called Documents number 13 that focuses on the marketing system for life insurance industry. The implications of that particular piece of policy is still getting played out. These adjustments reflect the regulator's understanding of an assessment of current financial environment and risk factors that exist in the insurance sector. I think these adjustments are moving the industry from a high-fee, high-risk model to a more sustainable, more prudent model. Going back to commissions, I think we are in a transition period where commissions are going to drop before they stabilize. Short-term commission is expected to have some downside still, but deceleration has already taken place. If you look at the first half of 2025, The speed of deceleration has already narrowed. I think all in all, we believe the insurance industry is facing short-term challenges, but there is still long-term optimism. Demographic shifts are creating inelastic demand for insurance. An aging population will require more life insurance and pension support, and there is also greater awareness of the protection that insurance can offer, and insurance will be an irreplaceable part of risk management. In the first half of 2025, our BaaS new premium and renewal premium continued to see growth, with renewal rate over 90%, which means we have exhibited resilience and certainty in our value proposition despite industry weakness. Our AI technologies have empowered us to better support our customers, and we are able to have a more favorable cost structure to respond to a current challenging environment. We have adopted a flatter management style and a commission-based mechanism. This can help the BaaS industry, insurance industry cloud, to better transform with greater resilience. Compared to our peers, we are able to stay more flexible in response to different circumstances. We are in a better position to capture customer value and manage them in a precise manner. Thank you so much. In the interest of time, we will take some questions from online, and I am going to read out the questions on behalf of the online participants. The first from Guotai Junan, Mr. Sun Kun. The company's MaaS business is regaining momentum and restoring double-digit growth. Is it because the macro environment is getting better, customers' ability to pay is getting stronger, or is it because the company's product market competitiveness is further strengthened? What is the market share of MaaS business today? For this question, I will give the floor to Mr. Duan Ying. Driven by national policy support and recovering market demand, the company's MaaS business achieved double-digit growth in the first half of the year. Specifically, our core customer contributed 16% more revenue than the same period last year, with their number increasing by 1%. The core customer count increased by 1%. Average revenue per user, ARPU, rising by 14%. This indicates the ARPU growth has become a bigger contributor to core customer revenue, strengthening the company's customer loyalty. This is a sign of very good customer stickiness. Meanwhile, core customers' share of MaaS revenue decreased from 78% last year to 76% in the first half of this year, which represents a 2% point drop. Conversely, non-core customers' revenue grew 32% year-on-year, outpacing the overall MaaS business growth. This demonstrates increased client payment and higher payment amounts, along with expanding market share. This is very good thing because I believe we have evidence to prove that our MaaS market share is on the rise. The company's land first, expand later business model, combined with strong product competitiveness, has gained market validation. Specifically, financial sectors such as banking, insurance, consumer finance, fintech, and auto finance remain key drivers of MaaS revenue. Thank you so much. Thank you, Mr. Duan. For our one last question, I am going to give the floor to Mr. Chen Yuxuan from Huaxin Securities. The question is: The company has established a stable profitability with ample cash reserves, yet its share buybacks in the first half of the year totaled only HKD 25 million, which represents a significant slowdown compared to the over HKD 200 million spent on share buybacks during the previous two years. Will the company implement more capital market revitalization strategies in the second half, such as increase the share buyback or dividend plans? Earlier this year, management emphasized capturing market share at all costs. How will the company balance short-term expansion with shareholder returns? I'll give the floor to Mr. Zhang Shaofeng. At beginning of the year, we mentioned that we need to spend money more on establishing and enhancing our AI capabilities, AI infrastructure, including buying computing power, acquiring more talent, and also data needed for training large model. Previously we mentioned, just now some analysts asked about we're not spending as much on data as we expected. That's because we have quality requirement for data. It's not like we're just going to taking any data. We need quality data for training purposes. We'll spend the money in a way that benefits the mid to long-term interest of the company. We're not short-term oriented. In the first half of the year, the change in the buyback scale is a reflection of the rhythm and the balance between mid to long-term development and short-term growth. The company has always given priority to shareholder returns. In the first half of the year, the buybacks were conducted with the prerequisite of ensuring adequate and sufficient operational cash flow. We have to make different considerations as we approach buybacks. In the second half, based on the changes in the capital market and also the need for business development and R&D, and combined with our own financial compositions, we're going to reevaluate our positions on dividend payout and buybacks. Last year, annually, we authorized maximum buyback at 10% of total equities. I believe there is enough support for actual buyback execution. If we design any specific buyback plans, we're going to publish and announce them accordingly in time so as to make sure that we keep shareholders in the loop. Regarding short-term expansion and shareholder balance, that's more of a long-term perspective. At the beginning of the year, I mentioned that 2025 is the year zero of agentic AI, which is what I call silicon-based employees. In the market, a lot of our fellow peers or analysts have very vague understanding of this concept. What does it mean by year zero? What does it mean by agentic AI? I think today people have a better understanding of what it is. AI service is still in a rapid growing phase, and they're gaining enterprise recognition. I think the top target for Bairong is to increase market share and capturing more customer mind share. We're not in a stage of apparent cash returns yet. The company believes in short-term investment, be it for market expansion, computing power, or R&D investment. All those short investments are always in line with long-term shareholder value. We're not going to do anything simply to boost the financial performance in the short run, but at the expense of mid to long-term profitability. We do everything in light of what they could do for the company's financial performance in the next two to three years instead of the financial returns in the next month or in the second half of the year. That's the consistent position of the company when it comes to financial investment. So that is why we have disclosed to you that there might be some business pressure in the second half of the year. But I hope that you will be able to understand that we are operating within a reasonable range, so we have reasonable expectations to the second half performance. All actions we take are for the benefit of long-term company value and interests. If there is any further actions, we are going to disclose them accordingly, and we welcome you to pay close attention to our announcements. Thank you so much, Mr. Zhang Shaofeng. In the interest of time, we have concluded our Q&A session. Thank you all for your questions, and thank you, the management, for your responses. I would like to once again thank everyone for participating in Bairong 2025 interim results announcement meeting. If you have more questions, please visit the company's investor relations website at ir.brgroup.com or send an email to ir@brgroup.com for inquiries. Additionally, you are welcome to scan the QR code to follow our digital IR assistant, Bairong Rong, which is the QR code on the screen. This AI assistant, Bairong Rong, 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 earnings call. Thank you. Thank you.