Guide:
Didi’s unique travel scene big data has a very broad application prospect in the financial field. In the future, it can cooperate with banks, insurance, payment and wealth management institutions to help traditional financial institutions improve resource allocation efficiency and reduce customer acquisition and risk management costs. . Big data in travel scenarios has important commercial value in transaction fraud identification, risk pricing, precision marketing, full life cycle risk management, and growth operations. The ability to apply and analyze big data is becoming a core competitive element in the future development of financial institutions. This article starts from the perspective of auto finance and car loan products, combines scene data with traditional credit risk control concepts, and accurately identifies credit risk changes in the process of business development, which has played a positive role in improving business models and reshaping user value.
0. directory
- What is Auto Finance?
- What is Didi Auto Finance doing?
- Application of Didi Big Data in Risk Control of Auto Finance
- Existing problems and solutions from the perspective of assets
- Existing problems and solutions from the perspective of whole-process risk management
- Three optimization points in data application
- The application prospect of Didi big data in the auto finance risk control scenario
- Enterprise credit intelligent risk control
- Intelligent risk control of retail credit
1. What is auto finance?
Auto finance mainly refers to financial services related to the auto industry, and is a financing method involved in various links such as auto R&D, design, production, circulation, and consumption. It mainly includes fund raising, credit installment, mortgage discount, financial leasing, and related insurance and investment activities.
▍Business model
In the retail business, commercial banks and financial leasing companies are the funders, dealers/4S stores/leasing companies are the sales channels, and the automobile e-commerce platform plays a diversion role, jointly providing financial products for individual consumers who need to buy cars by installments and service.
From the perspective of the competitive landscape, banks and manufacturer finance are the main players in the retail market, and they have absolute advantages in terms of capital costs and channel acquisition. In addition, the auto e-commerce platform, as an online diversion service provider, improves the efficiency of customer acquisition for traditional financial institutions, and has also been active in the auto finance market in recent years. From the perspective of product types, after-sale leaseback is the mainstream in the market, while direct leasing needs to develop rapidly.
2. What is Didi Auto Finance doing?
1) At this stage, Didi’s auto finance business is positioned to serve the travel ecology. Everything starts from user value and provides low-cost car purchase financial solutions for drivers who need to buy cars.
2) Internally build an auto financial risk control system, through the accumulation and application of online car-hailing scene data, continuously improve the comprehensive risk management capabilities, generate high-quality online car-hailing financial assets, and gradually form risk pricing capabilities.
3) Provide high-quality financial assets and systematic risk control capability output to traditional financial institutions, realize efficient matching of funds and assets, and accumulate financial asset management capabilities. At the same time, as a bilateral platform connecting funds and assets, it has established long-term partnerships with mainstream financial institutions and continued to provide financial support for the online car-hailing system.
In the future, the business scope of Didi Auto Finance will continue to enrich with the development of the travel industry ecology, extending to the entire travel industry chain, providing financial services for car dealers, 4S stores, agents and other car sellers to purchase cars and operating equipment, To meet the financial needs of each link in the upstream and downstream of the industrial chain, and gradually form a new financial format for the automotive industry that integrates information flow, capital flow, and logistics.
3. Application of Didi big data in risk control of auto finance
Under the traditional credit framework, the risk control model that judges the repayment ability based on the lender's central bank credit investigation no longer meets the risk management needs of online car-hailing finance. In the online car-hailing scenario, the risk control of auto finance puts forward higher requirements on the authenticity and stability of loan assets, and the timeliness of risk warning. It is particularly important to establish an intelligent marketing and intelligent risk control decision-making system based on big data.
▍From the perspective of assets:
Problems at the C-end of car loans: Before the loan, the data in the unused scene is used as a supplement for personal credit investigation, the data in the loan is missing, there is no matching risk warning plan, and the collection efficiency after the loan is low. It is necessary to form a dynamic credit score for the online car-hailing lender .
Solution: use Didi big data to supplement the traditional retail scorecard model, apply the data that can reflect the characteristics of personal credit risk in the scene to the field of auto finance, and formulate risk control policies and access standards. At the same time, establish a PD (probability of default) scoring model for car owners in the system, pay attention to significant changes in PD parameters, and provide risk warning solutions under big data. Gradually build a comprehensive risk management system under the online car-hailing scenario, and improve the risk management capability of the whole process.
Car loan B-end problem: Traditional financial institutions lack of CP (Car partners) credit data, which makes them unable to effectively identify channel risks, especially for small and medium-sized CPs, it is difficult to obtain credit from traditional financial institutions.
Solution: With the help of the big data of the Didi platform, support the employer's credit approval for CP. Specifically, the basic information of the channel and the data dimensions that can reflect its asset scale, asset utilization efficiency, and driver management ability are systematically sorted out to form input variables, while continuously accumulating bad samples in the system to establish a CP semi-supervised model . The output of the model is the comprehensive score of CP credit rating, which intuitively reflects the risk level of CP. At present, the CP rating of auto finance is a monthly output, which can dynamically reflect changes in the CP risk level.
▍From the perspective of whole-process risk management:
In the actual operation process, we found the following problems in the three stages of retail car installment loans before, during and after the loan.
Pre-loan access risk: The loan applicant is not the driver who actually operates the vehicle after the loan is granted, that is to say, A borrows and B repays. This kind of problem usually occurs in the channel incoming link. There are certain operational risks in the sales process of auto finance products. In order to increase the order completion rate, offline channel salespersons find people with good credit qualifications who are more likely to pass the pre-loan review to apply for loans instead of drivers. However, drivers who actually run Didi have credit The assets are poor, the repayment ability is not enough to support the monthly payment, and the probability of PD default is high. Then the credit risk of this car installment loan will be gradually released during the asset performance period after the loan.
▍In the first order, the lender and driver information did not match:
Operational risks during the loan: The lender returns the car within the duration, and the vehicle will be compensated by the leasing company. After the leasing company finds a new driver, the new driver will operate and continue to repay the loan. In this case, the judgment of traditional risk control on the initial lender before the loan, and the GPS positioning of the vehicle can no longer effectively reflect the risk changes of the operating vehicle after the loan. When the loaned vehicle is matched with multiple Didi drivers successively during the duration, the leasing company faces great challenges in vehicle operation management, cash flow management and driver management. Sometimes multiple drivers collectively return the car, which will cause channel concentration risk.
▍A car is matched with multiple drivers at different points in time during operation:
Post-loan overdue collection: Traditional credit risk control lacks post-loan data for online car-hailing cars. Without access to lender income and operating behavior data, it is impossible to determine the lender’s repayment ability and repayment willingness behind each overdue debt , so it is impossible to give priority to debt collection for lenders with high income-repayment ratio and repayment ability. In this case, it is necessary to formulate a collection scorecard based on the order data of the lender platform and the operation data of the loaned vehicle, and carry out classified management of collection.
▍DiDi big data can solve:
The establishment of a comprehensive risk management system for online car-hailing finance.
During the preparation of retail data and the development of model variables, a long list of models is formed from the basic dimension of lender credit to the four major risk factors of city, channel and vehicle to realize dynamic monitoring covering the entire life cycle of loan assets. At the same time, through the continuous accumulation of model dependent variables (bad samples) through the asset performance of the invested companies, we can effectively grasp the changes in risk levels, establish an early warning and response mechanism, and reduce the loss rate.
Each risk factor drills down to form multiple risk indicators, which are combined to form a risk control strategy. Through the comprehensive application of single strategy and multiple strategies, early warning and timely prevention of risks in loans are realized.
Specifically, the optimization direction has the following points:
Optimization point 1: From the traditional lender risk assessment at the time of lending, it is optimized to dynamic multi-dimensional risk monitoring throughout the process.
Traditional credit risk control only focuses on the single-dimensional credit risk measurement of the lender, but in the online car-hailing scenario, city policy compliance, vehicle operation status, and channel management capabilities will all play a decisive role in the change of credit risk throughout the credit process . In this regard, we use the continuous accumulation of Didi car-hailing scene data and bad samples to supplement the traditional credit data dimension and optimize A-card and B-card.
Early warning needs analysis:
Timing of disbursement:
Anti-fraud information verification, data dimensions include but are not limited to verification of drivers, vehicles, passenger-vehicle matching, and channel basic information on the platform side, and at the same time check channel incoming risk.
After disbursement, real-time changes in the credit risk of the lender are reflected through monitoring during the loan, and a big data risk early warning system is established.
Establish a big data internal evaluation verification governance structure, internal evaluation verification process methods, and provide different levels of optimization strategies and real-time processes. In the early-warning model, the typical mid-loan early-warning strategy is as follows:
Driver dimension strategy: flow stability, earning capacity, whether a witness has been issued, etc.
Vehicle dimension strategy: the operation status of the vehicle on the platform, the matching status of the vehicle and the driver, the mileage of the vehicle, whether the vehicle license has been obtained, etc.
CP channel strategy: channel negative information scanning, channel concentration risk events, compliance ratio, channel concentration overdue, etc.
City compliance strategy: whether the online car-hailing platform certificate has been obtained, the progress of the city compliance certificate, whether to classify management, etc. .
With the continuous enrichment of data dimensions, the drill-down dimensions of the four major risk factors will gradually increase. At the same time, we are also verifying one by one in the actual business, and carry out strategy iteration through the results of the driver A card B card model.
Post-loan collection:
optimize the collection scoring model. Real-time analysis and monitoring of the overdue days, billing behavior, and average monthly income of overdue drivers, and a comprehensive score list of repayment ability and repayment willingness corresponding to each overdue debt, helping to improve the efficiency of post-loan collection.
Optimization point 2: Increase the time width and time-point observation depth of data observation, and introduce forward-looking on this basis.
Through long-term observation of data, iteration of a single risk strategy, and continuous verification of multi-strategy applications, we will obtain the historical average level and law of driver credit risk changes, and make forward-looking adjustments based on the current stage of the business and future development trends. After the PD (probability of default), the significant changes in credit risk are quantitatively and qualitatively assessed.
Optimization point 3: Relying on big data analysis capabilities, form a comprehensive judgment on the overall risk and return changes of the business.
Through the whole-process risk management of financial leasing vehicles at the C-end, the driver credit portrait and CP channel portrait in the form of financial leasing products are gradually outlined, and the operational risks of auto finance in business models and products can be quickly identified, such as financial leasing packages, economic leasing , CP compensation, concentrated default risk, etc. Furthermore, there is a clear and accurate measurement of the quality of auto finance assets, so as to achieve a balance of risk and return on the asset side and the capital side.
4. The prospect of wide application of Didi big data in auto finance scenarios
▍Enterprise Credit Intelligent Risk Control
Direction: In the entire travel industry ecology, there are a large number of scattered small and medium-sized service providers/channel providers. The daily operating data of these small and medium-sized enterprises on the Didi platform reflects their operating capabilities, capital liquidity management, and driver management capabilities. Multi-dimensional business data can fully support data risk control to obtain funds, and provide decision-making innovation solutions for business, including identification of abnormal customer behavior, differentiated credit approval, whole-process risk control and early warning, limit setting, etc.
Progress: At present, some licensed auto finance institutions that have business relations with Didi platform partners have conducted in-depth discussions with us on the credit granting method of data risk control. Multi-dimensional data establishes a risk control model to provide financial support for high-quality car rental companies to grant credit to the public.
▍Retail Credit Intelligent Risk Control
The Didi platform has obvious bilateral effects, that is, both the supply side and the demand side complete transactions through the platform, so a large amount of transaction and operation data will be deposited on the platform. When the target of auto finance services is the people who own cars in the system, Didi big data can be used to supplement the deficiency of traditional retail scorecards, and the non-credit data in the system can be applied to auto finance business scenarios, for example, to formulate product-level risk control Policies and access standards, output automated credit scoring, anti-fraud, risk exposure management, risk pricing, etc.
Gradually establish a risk management system in the online car-hailing scenario, and realize the innovation of the internal evaluation model at the data, decision-making, and algorithm levels.
Including: pre-screening customer groups, feature model establishment and training, anti-fraud rule design, online strategy verification, joint modeling with partners, online post-loan overdue management, etc.
With the accumulation of big data risk control capabilities, no matter whether the product form is new car financing lease or vehicle mortgage loan, an intelligent risk control system can be established for different business types. On this basis, the dynamic monitoring of platform data can help screen personal credit users with good asset performance, form a white list, automate loan approval, and improve asset matching efficiency.
Author of this article:
Tang Pei
Didi | Auto Finance Business Analyst
An engineering student with a management consulting background in the financial industry believes that the meaning of life is closely related to valuable work. He has been looking for smart, witty, in-depth thinking habits, highly sensitive to business, and a broad-minded partner to join the team.
At the same time, you are also welcome to pay attention to Didi Technology Official Account, we will bring you the latest open source information and technical dry goods!