Thoughts of our next generation fintech. With the turnaround of the fundamentals and changing, then the change of our revenue mix, we think the company has bottom, at least from the fundamental point of view. Let's dive into our business. For our credit solution business, our flagship digital lending platform, [Non-English content] , incorporates 20 years of lending experience, servicing over a 100 millions of registered users. In 2025, we have approximately. 77% of the loan was issued to repeat borrowers. This is a record high that shows our brand loyalty and the ability to manage risks with existing borrower. With the use of AI, we are able to customize the marketing contents to individual customers. The response time for our AI agent to generate personalized new contents is about half a second.
With the help of AI, we analyze and approve millions of loans automatically every year. With the use of AI and improvement in credit markets, our First Payment Default was down from recent peak of 2.1% to near 1.5%, about the same level as the May 2025. We have done a few measures in the past quarter to mitigate risk. We increase and maintain the repeat borrowing amount to 77% of total loans facilitated. The risk from repeat borrower are more predictable, therefore, that enable us to manage our risk better. Here are some of the metrics improvement from our AI. Customers on average stay longer with the AI agent outbound call compared to the traditional IVR sales pitch, contributing to a higher sales conversion ratio.
Not only the AI agent outbound call is more effective than the traditional IVR, it is also 84% cheaper. Our generative AI model is 50% faster in generating marketing messages to clients, and the conversation is more relevant and personalized. We use AI agent extensively in our debt collection. In 2024, about 45% of a first notice collection were handled by human. Now the number is down to less than a quarter. Our recovery staff on average handles 525 cases per quarter, and they are 47% more productive. It is not just the volumes. The AI is also handling more complicated cases, as you see on the middle bottom. In 2024, our AI agent only handles the 1st-day delinquent notice.
In 2025, it handles 14%-20% of a longer delinquent cases that require more personalized interaction with the borrower. The AI can, on the spot, build a repayment proposal or litigation documents within 5 minutes while chatting with the borrower. This shows the level of sophistication and efficiencies of our AI agent brings to the operation. This is one of the most powerful AI agent we developed last year, the EQ and the GQ agent. They help us to identify idle capital across our lending entities and predict the capital flow for the next month. It then automatically reallocate capital from the region platform that has a capital surplus to the one that needs more capital. EQ agent monitors transactions every month, identify the trends that is changed during the period, and recommend a capital allocation plan.
GQ agent does the execution, which facilitates the process, including going through the approval process, trigger the fund transfer, and so on. We were running this since mid-2025. So far, the capital allocation accuracy has improved by 10% to 66%, resulting in lower capital cost by 24 basis point. The processing speed also increases. Before AI, we used 2 full-time staff total of 100 person hour to make the capital plan. With the AI, it only takes 20 minutes. It is more efficient, and it allows us to do reallocation on a weekly basis instead of monthly. Our agent is able to detect when restricted cash becomes unrestricted cash for redeployment. This makes our capital deployments much more efficient, and we are looking to deploy this technology to some of our institutional clients. This is our AI-based risk management process.
It starts with a pre-approval, including the KYC. We understand our borrowers' profiles through our channels and our record about the borrower's credit history. Our model consists of over 200 risk model with the data from over 2,000 sources. The model classify them into one of the 10 risk classes. Based on the risk classes, the system will make credit decisions and pricing automatically. In a post-drawdown stage, the AI continues to monitor the borrower's behavior, predicts potential delinquency, and propose solution. On the recovery side, our smart reminder systems can predict the best time to reach borrowers and improve overall recovery efficiency. With the help of AI and improvement in credit cycle, our 30 days First Payment Default was down to 1.5% from last year's peak of 2.1%. Let's move on to our second business segment.
Our second business is insurance brokerage. We are one of the very few platforms in China that has both offline and online distribution channels that operates nationwide. We recognize customer journey has changed in the past five years. They get a lot more information from social media and other digital channels. Many even prefer the digital interaction than the face-to-face with the life brokers. As the first projects of our next generation fintech, we leverage our existing channels and funnel down customer demand and identify products that fits the customers and the carriers. That fits the customer and the carrier have difficulty reaching those customer segment. Our first product is healthcare insurance products. Our existing customer base or traffics are young professionals. This customer segment has a unique lifestyle. For example, they travel a lot for work or for leisure.
Carrier like these customers because they are young and healthy, but hard to reach. We determine a few use cases that target these customers and design our own products underwritten by our carrier partners. Use our existing digital channel to interact with customer. The result has been very good. Here are the financial results of our internet insurance business. On a quarterly basis, our Q4 2025 gross premium was CNY 50 million, compared to just CNY 4 million in the Q1 of 2025. That representing the compound quarterly growth rate of 87%. The gross premium contribution has also been increasing.
In Q4 of 2025, it was 5.8%. We expect the revenue contribution from the online channel will reach 50% of overall insurance revenue by 2027. Another interesting point is the online channel has helped reviving our offline channel as we use the online channel as a low-cost acquisition channel. We cross-sell to our customer to offline channel for high margin life or P&C products. Here are the strategic priority for this year. We will continue to drive AI innovation and operational efficiency. We will leverage our data and AI tools to support new product development as we begin to sell our AI capability as a service to external customers. We are transforming from an AI adaptive to AI native.
Here is the framework of the business model and the service offering. Starting from the bottom, we have the technology layer with the [Non-English content], our proprietary LLM, and Magic Cube, our multi-agent platform. The six AI agents above that use these technologies and other external technologies. These agents can learn, perform, and can be sold as separate agents. For the service layer, we develop functional services such as customer acquisition, credit review, and risk management targeting to corporate and consumer segments. Finally, on the business layer on the top, we develop a business model out of these services. Currently, we are turning our credit solution and insurance business to agentic business. Our growth driver for this year are recovery of our core business as the credit quality improves substantially. Expanding product scope with AI agents.
Continue to grow our online to offline insurance model as it is pivotal to our next generation fintech strategies. Finally, exploit the next generation fintech opportunities. Turning to financial now, our 2025 revenue was CNY 5.7 billion, it dipped by about 1.5%. We intentionally slowed down our loan facilitation in the fourth quarter 2025 to weather the credit down cycle. As the technology related revenue increased in 2025, the revenue contribution from lending reduced, and we expect this trend to continue in 2026. Sorry. Here's our operating metrics. Here's our operating metrics of our lending business. Starting from the left, as you see our First Payment Default, 30 days in January 26 is already down to May 2025 level when the risk began to deteriorate.
Our February and March numbers show the number is near our long-term average, as we have increased the loan facilitation effort as the result. In the middle, the cost of capital decreased by 110 basis points since October 2025, when the new lending regulation was in force. Customer acquisition costs also decreased by 80 basis point year-over-year. These two metrics show after the new regulation in October, many small players have left the market, resulting in less competitions, and also our AI has helped driving a higher sales conversion and more efficient capital allocation. This is the breakdown of our insurance portfolios. We have a solid renew book and a new P&C book. The internet channel has been driving the growth in the past three quarters.
The growth from the internet insurance also results in more spillover to the traditional line as the cross-selling is taking effect. On the cost structure, AI has substantial benefits in lowering our cost. As the sales and marketing cost as a percentage of revenue decreased from 32% in 2024 to 21% in 2025. Origination and services as a percentage of revenue decreased for both credit solutions and insurance business. We are seeing AI has a major effect on our cost structure as the internet insurance channel has virtually zero commission to the live agents. Heading to our last slide. Here's our financial highlights. Our 2025 revenue dipped by 1.5% as we decided to slow our pace of loan facilitation in the fourth quarter when the risk was high.
On the net income side, we changed our lending model to take on a higher proportion of the book under the risk-taking model. We took that approach as the funding pool was stretched during the period of regulation change in the second half of 2025. Many funding partners and guarantee companies did not want to take the credit risk at that time. We have many high quality repeat borrowers, so we took a calculated risk to guarantee those loans so that the capital can flow through. Unfortunately, our accounting standard has a revenue and provision mismatch for this risk-taking model. We need to take a provision upfront for this self-guarantee model regardless of the risk. Our revenue is amortized over the loan period.
As the risk-taking model loan book grows, we took a bigger hit upfront, it results more predictable revenue stream in the future. We will continue to monitor the market conditions and may adjust the loan mix between the risk-taking and the old model. This is the end of our formal presentation. You may follow us on LinkedIn and X. Now let's move to the Q&A from the audience. Okay, let me go through that. The first question is, given the growing interest in AI agent across industry, would you consider partnership or white label deals where Magic Cube become the engine behind other platform workflow? Yes, we are open to different partnership.
What we wanna do is we are not just exporting the technology to our clients. We will also work very closely with our clients in transforming their business using the AI agent. Meaning, we will also help them how to use the AI agent to integrate into their workflow, their existing workflow and their business model. Okay, the second question is: Yiren Digital has shown that AI can cut costs and boost efficiency. Do you still see a easy win from AI that haven't been fully implemented in the business? That's a very good question.
We think the internet insurance will still have a very good growth potential because currently we only deploy our AI agent to sell the what we call the standard product that's sold online, mostly the one-year health products. We think as the technology becomes more advanced, we can start rolling out our agent to sell more long-term product, like the life or the multi-year property and casualty products, so which has a much higher margin. Okay. The next question is. Several valuation model shows a very large upside versus the current share price. In your view, what are the one or two concrete milestone that could trigger a rerating from here? That's a very good question.
I think there are two factor that will trigger the rerating. First of all, our existing business. It went through a regulatory uncertainty last year and also together with the down cycle of the down credit cycle. This year, we have seen so far since January, the risk matching has improved much better than expected. From our existing business, that's definitely a turnaround. For our AI business, as we start generating more technology revenue from these AI products, we expect.
We are looking to sign more long-term contract with our clients. That will enable us to build a more stable revenue stream to get and a more predictable cash flow to support our ongoing R&D. I think the AI part will help us to not just help our existing fintech business, but we are also developing a new use cases in other industry. Which we hope we will make some announcement later this year to go through that. Okay. There are few audience asking about our international strategies outside of China.
For those who follow us, we started our Philippine business in 2023, and our Indonesian business just started in September 2025. In the next 12-18 months, our Indonesian business will start to scale up because we spent the first 6 months understanding the market and collecting more data. Now we, and this year, we are in the positions to scale up our loan facilitation scale and to generate higher revenue contributions. Next question. Your presentation mentioned fraud detection blocking large number of high-risk customer every day. How does this AI edge translate to a better terms or larger quota from our fund partner? That's a very good question.
Our fraud detection program actually reads about tens of thousands of documents every day. Last year alone, it saved us about RMB 180 million of potential fraud loss. I think given those potentials, the funding partner will see our, definitely our loan loss has improved a lot. Together with the macro, the industry condition with the less competition, they are more willing to partner with us at a lower cost of capital. This help us to improve our margin. Okay. One last question. We're running out of time. The AI leader globally still trade at a premium valuation.
What do you think is missing in the current perception of Yiren as an AI-powered platform rather than just a lender? I think, well, we have been building the, we have been doing our AI R&D since 2021. I think this, 2025 was the year we can we have launched the products. 2026 and 2027 is the one that we can start monetizing those technologies. Yeah. Okay. I think we are running out of time. Thank you for attending the roadshow. For those who leave your contact address, we will follow up with you of, with the answers with your questions. Meanwhile, please also follow us, you know, on our X and LinkedIn account.
With that, thanks very much.