Bank financial resource cloud customer potential tapping method based on machine learning
By constructing a customer response probability prediction model using machine learning methods, the shortcomings of traditional bank treasury cloud customer screening methods are addressed, enabling precise marketing and resource optimization, and improving customer screening efficiency and marketing effectiveness.
Patent Information
- Application Number
- CN202511564355.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-30
- Publication Date
- 2026-02-06
AI Technical Summary
Traditional bank treasury cloud customer screening methods are logically simple, fail to reflect the true needs of customers, rely on human experience leading to low efficiency, lack machine learning algorithm support, and cannot achieve refined analysis and quantitative evaluation, resulting in low marketing conversion rates and high costs.
By employing machine learning methods, through defining sample data, customer profiles, data processing, model building and validation, application and evaluation, the LightGBM ensemble learning algorithm is used to construct a customer response probability prediction model, outputting a customer value score to achieve precision marketing.
It improved the accuracy and efficiency of customer screening, reduced labor costs, enabled the quantitative assessment of customer value and the optimization of marketing strategies, and improved marketing conversion rates and resource allocation efficiency.
Smart Images
Figure CN121481691A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of financial cloud business, and in particular to a bank financial cloud customer potential tapping method based on machine learning. BACKGROUND
[0002] Financial cloud business is a common business in the banking industry, that is, a multi-channel, multi-subject and multi-bank account financial comprehensive management service provided to enterprise customers. In the marketing of traditional bank enterprise customers, how to select suitable financial cloud customers has long relied on the experience of bank staff to judge, and can only rely on simple data analysis for analysis, such as screening according to simple indicators such as customer registered capital, asset size, and deposit. According to this traditional financial cloud customer screening method, there are many drawbacks in the actual target customer list development and use process: first, the screening logic is too simple and cannot reflect the real financial cloud demand of enterprise customers, such as whether there is a group fund management demand, a multi-member fund management demand, etc., resulting in low conversion rate of the list screened out; second, manual acquisition of various data, large amount of manual work, low efficiency; and group relationship, transaction characteristics, etc. cannot be analyzed through simple and intuitive data; third, the bank internal data has certain limitations and cannot reflect the real customer business situation and potential; and the customer's in-house and out-of-house situation changes all the time, and only analyzing static data cannot reflect the customer business trend and business potential cannot be quantified.
[0003] The existing financial cloud customer potential analysis, the business department can use limited technical means, mainly based on the extraction and manual processing and sorting of simple customer registration information, deposit information and account settlement information, lacking refined analysis tools.
[0004] And now big data technology and machine learning integrated algorithms have been widely used in the financial field, such as risk management and fraud monitoring, loan loss early warning, potential customer mining, customer value stratification, etc. However, in the potential customer mining scene of Huaxia financial cloud product, it still relies on manual experience to mine potential customers, without the application of machine learning algorithms.
[0005] The existing financial cloud potential customer mining mode relies heavily on the subjective judgment of business personnel, lacks objective data support and quantitative standards. At the same time, personal experience is difficult to standardize and scale, resulting in unstable business effect, and is easily affected by cognitive bias, which cannot meet the needs of large-scale customer mining. In addition, manual mining needs to invest a lot of human resources to evaluate customer value, and the marketing cost is high, which will seriously restrict the growth efficiency of the business.
[0006] The financial cloud potential customer mining method lacks the application of machine learning technology, does not construct a customer response probability prediction model and a customer value scoring system based on multidimensional data, and thus cannot realize the continuous optimization iteration based on the feedback mechanism. In the absence of the support of the machine learning algorithm, the system cannot automatically discover the behavior patterns and demand characteristics of potential customers, and finally limits the improvement space of customer mining effect and the business innovation potential. SUMMARY
[0007] In view of the above problems, the present application is proposed in order to provide a bank financial cloud customer mining method based on machine learning to overcome the above problems or at least partially solve the above problems.
[0008] According to one aspect of the present application, a bank financial cloud customer mining method based on machine learning is provided, and the mining method comprises:
[0009] Step S1: defining sample data;
[0010] Step S2: customer profiling according to the sample data;
[0011] Step S3: data processing of the customer profiling;
[0012] Step S4: model construction and verification;
[0013] Step S5: application of the model;
[0014] Step S6: evaluation of the model.
[0015] Optionally, the step S1: defining sample data specifically comprises:
[0016] Based on the product cooperation breadth, depth, fund adequacy, transaction activity, and group relationship indicators of the public customer, the target customer group of the financial cloud is screened.
[0017] Optionally, the step S2: customer profiling according to the sample data specifically comprises:
[0018] Collecting product cooperation, transaction conditions, asset conditions, basic information, and group relationship data of the financial cloud customers;
[0019] Customer profiling is performed on the customers.
[0020] Optionally, the step S3: data processing of the customer profiling specifically comprises:
[0021] An index system is constructed from the dimensions of basic information, association conditions, asset conditions, transaction conditions, held product conditions, and signing conditions;
[0022] After the data processing is completed, the data is cleaned.
[0023] Optionally, the basic information includes: the branch to which the company belongs, the nature of the company, the size of the enterprise, the registered capital, and the paid-in capital.
[0024] The associated information includes: the number of group companies, the number of associated companies, and the number of companies at the same address.
[0025] The asset information includes: total assets, N-month average monthly deposit, deposit balance, and change trend.
[0026] The transaction information includes: the number of transaction channels, transaction amount, number of transactions, number of counterparties, and transfer-in and transfer-out amount ratio.
[0027] The holding product information includes: the total number of holding products and the number of accounts of each sub-product.
[0028] The signing information includes: the signing status of financial services and the signing duration.
[0029] Optionally, the step S4: model construction and verification specifically includes:
[0030] Divide the data into training set and test set, and use machine learning algorithm to train the model;
[0031] The model training process adopts K-fold cross-validation strategy to evaluate its generalization ability, and the key parameters of the model are automatically optimized through hyperparameter optimization algorithm;
[0032] The performance of the model is comprehensively evaluated by multiple indicators.
[0033] Optionally, the multiple indicators include: AUC, KS, and recall rate.
[0034] Optionally, the step S5: applying the model specifically includes:
[0035] Target customer selection: apply the trained optimal model to the candidate customer group at the latest time point, output the prediction probability score of each customer, and select the top-ranked customers for marketing;
[0036] Customer value assessment: comprehensively consider the purchase probability, fund situation, and transaction size of the customer, and score the customer by weighted comprehensive scoring method; and sort the customers based on the comprehensive score to classify the customers;
[0037] Model explainability analysis.
[0038] Optionally, the model explainability analysis specifically includes:
[0039] Global explainability: calculate the average absolute SHAP value of each feature to obtain the global feature importance ranking, which is used to guide business insight;
[0040] Local explainability: generate the SHAP value of each feature according to the prediction result of each customer.
[0041] The application provides a bank financial cloud customer tapping method based on machine learning, the tapping method comprises the following steps: step S1, defining sample data; step S2, performing customer portrait according to the sample data; step S3, performing data processing on the customer portrait; step S4, performing model construction and verification; step S5, applying the model; and step S6, evaluating the model. The LightGBM integrated learning algorithm can automatically learn the complex relationship between customer characteristics and purchase tendency, output the response probability of the customer, and realize accurate marketing of the customer. The customer omission caused by manual experience in tapping potential customers is avoided.
[0042] The above description is only a summary of the technical scheme of the application, in order to more clearly understand the technical means of the application, the content of the specification can be implemented, and in order to make the above and other purposes, characteristics and advantages of the application more obvious and easy to understand, the following specific embodiments of the application are described. BRIEF DESCRIPTION OF DRAWINGS
[0043] In order to more clearly illustrate the technical scheme of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the application, and those skilled in the art can obtain other drawings according to these drawings without creating any creative labor.
[0044] Figure 1 The flow chart of the bank financial cloud customer tapping method based on machine learning provided by the embodiments of the application. DETAILED DESCRIPTION
[0045] The exemplary embodiments of the present disclosure will be described below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.
[0046] The terms "include" and "have" and any variations thereof in the specification embodiments of the application and claims and drawings are intended to cover non-exclusive inclusion, for example, including a series of steps or units.
[0047] The technical solutions of the present application will be described in further detail below in combination with the drawings and embodiments.
[0048] As shown in the drawings, Figure 1 A bank financial cloud customer tapping method based on machine learning, the tapping method comprising: step S1: defining sample data; step S2: customer profiling according to the sample data; step S3: data processing on the customer profile; step S4: model building and verification; step S5: application of the model; step S6: evaluation of the model.
[0049] The definition of the sample includes:
[0050] Customer segmentation: based on the indicators of public customer product cooperation breadth, depth, fund adequacy, transaction activity, group relationship, etc., the target customer group of the financial cloud is screened.
[0051] Customer profiling includes:
[0052] Collecting product cooperation, transaction, asset, basic information, group relationship, etc. of the financial cloud customers, customer profiling is conducted to facilitate business understanding of various behavioral characteristics of customers, and to provide data support for precision marketing, personalized recommendation, etc.
[0053] Data processing includes:
[0054] An index system is constructed from the dimensions of basic information, association, asset, transaction, product holding, and contract.
[0055] Basic information: branch, company nature, enterprise size, registered capital, paid-in capital, etc.
[0056] Association information: number of group companies, number of co-associated companies, number of companies at the same address, etc.
[0057] Asset information: total assets, N-month average daily balance, balance and trend, etc.
[0058] Transaction information: number of transaction channels, transaction amount, number of transactions, number of counterparties, transfer-in and transfer-out amount ratio, etc.
[0059] Holding product information: total number of holding products and number of accounts of each sub-product;
[0060] Contract information: contract status and contract duration of some financial services;
[0061] After data processing is completed, the data is cleaned.
[0062] Model building and verification includes:
[0063] The data is divided into training set and test set, and machine learning algorithm is used for model training, such as LightGBM classification model. The model training process adopts K-fold cross-validation strategy to evaluate its generalization ability, and the key parameters of the model are automatically optimized by hyperparameter optimization algorithm (such as grid search). Finally, the performance of the model is comprehensively evaluated by AUC, KS, recall rate and other indicators.
[0064] Model application, including:
[0065] Target customer selection: Apply the trained optimal model to the latest time point of candidate customers, output the prediction probability score of each customer, and select the top-ranked customers for marketing.
[0066] Customer value assessment: Considering the purchase probability, financial situation and transaction size of the customer, the customer is scored by weighted comprehensive scoring method. And based on the comprehensive score, the customers are sorted and graded. For example, divided into high, medium and low three grades, to facilitate the branch to carry out marketing work in order of priority.
[0067] Model explainability analysis:
[0068] Global explainability: Calculate the average absolute SHAP value of each feature to obtain the global feature importance ranking, which can be used to guide business insight.
[0069] Local explainability: For each customer's prediction result, generate SHAP value for each feature, which objectively shows how each feature value of the customer pushes the model's prediction output from the base value to the final prediction value, providing clear and transparent individual decision basis for business personnel and meeting model compliance audit requirements.
[0070] Model evaluation, including:
[0071] After the list is issued, the effect of the list is evaluated, such as comparing the hit number, recall rate and hit rate of target customers of different priority levels, and comparing the hit situation of different priority levels with the hit situation of the industry control group, or comparing the transaction situation and deposit situation of the hit customers in the list before and after signing the financial cloud, to judge whether the new customers are the target customers of the financial cloud product.
[0072] Multi-dimensional customer data feature engineering
[0073] Based on customer basic information, assets, transaction flow, holding products and other internal data and external data such as industry and commerce, a business quantitative index system is constructed to provide data support for financial cloud potential customer mining and customer value assessment.
[0074] Machine learning algorithm driven potential customer mining model
[0075] Adopting LightGBM ensemble learning algorithm, a customer response probability prediction model is built to output quantitative value scores, accurately identify high-potential customers, and overcome the subjectivity of artificial experience and the limitations of rule-based models.
[0076] Precision marketing and resource optimization based on value scores
[0077] According to customer value scores, differential marketing strategies are formulated and marketing resources are accurately allocated, thereby improving marketing efficiency and reducing artificial marketing costs.
[0078] Data-driven business insights
[0079] Based on the feature importance output by the machine learning model, the business is empowered in reverse, helping the business understand the decision-making motivation of customers, thereby guiding service optimization and strategic adjustment.
[0080] Model effectiveness evaluation
[0081] Statistical comparison of hit numbers, recall rates, and hit rates of target customer groups of different priorities, and comparison of hit situations of different priorities with those of the industry control group.
[0082] Beneficial effects:
[0083] Higher prediction accuracy: Based on multi-dimensional customer behavior data, the LightGBM ensemble learning algorithm can automatically learn the complex relationship between customer features and purchase propensity, output customer response probability, and achieve accurate marketing of customers. At the same time, it can avoid customer omission caused by artificial experience in mining potential customers.
[0084] Optimal resource allocation: By building a customer value evaluation system, customers are prioritized to achieve optimal resource allocation, facilitate business to carry out marketing work in order, and provide customers with value measurement indicators for business and data reference for marketing work.
[0085] Lower marketing cost: Through machine learning algorithm, a large amount of data can be processed in batches, and a marketing customer list can be directly output, greatly reducing the time cost of mining potential customers through artificial experience.
[0086] Higher interpretability: Through SHAP value, the model is made interpretable to help the business discover the key factors driving the prediction results and guide business personnel to make strategic decisions.
[0087] The above detailed description of the specific implementation is further detailed for the purpose, technical solution and beneficial effect of the present application, and it should be understood that the above is only the specific implementation of the present application and is not used to limit the protection scope of the present application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.
Claims
1. A machine learning-based method for tapping into potential customers in a bank's treasury cloud, characterized in that, The mining method includes: Step S1: Define sample data; Step S2: Create a customer profile based on the sample data; Step S3: Process the customer profile data; Step S4: Model building and validation; Step S5: Apply the model; Step S6: Evaluate the model.
2. The method for tapping potential customers in a bank treasury cloud based on machine learning according to claim 1, characterized in that, Step S1: Defining sample data specifically includes: Based on indicators such as the breadth and depth of cooperation with corporate clients, their financial adequacy, transaction activity, and group relationships, we screen the target customer group for our financial cloud platform.
3. The method for tapping potential customers in a bank treasury cloud based on machine learning according to claim 1, characterized in that, Step S2: Creating a customer profile based on the sample data specifically includes: Collect data on the product cooperation, transaction status, asset status, basic information, and group relationships of the treasury cloud customers; Create customer profiles.
4. The method for tapping potential customers in a bank treasury cloud based on machine learning according to claim 1, characterized in that, Step S3: Data processing of the customer profile specifically includes: An indicator system is constructed from the dimensions of basic information, related information, asset information, transaction information, product holdings, and contract signing. After the data processing is completed, the data is cleaned.
5. A method for tapping potential customers in a bank treasury cloud based on machine learning as described in claim 4, characterized in that, The basic information includes: branch, company type, company size, registered capital, and paid-in capital; The associated information includes: the number of group companies, the number of companies with common related parties, and the number of companies at the same address; The asset information includes: total assets, average daily deposits over the past N months, deposit balance and its trend; The transaction information includes: number of transaction channels, transaction amount, number of transactions, number of counterparties, and ratio of inflow to outflow amount; The information on the products held includes: the total number of products held and the number of accounts for each sub-product; The contract information includes: the contract status and duration of the financial service contract.
6. The method for tapping potential customers in a bank treasury cloud based on machine learning according to claim 1, characterized in that, Step S4: Model building and validation specifically includes: The data is divided into training and testing sets, and machine learning algorithms are used to train the model. The model training process employs a K-fold cross-validation strategy to evaluate its generalization ability, and uses a hyperparameter optimization algorithm to automatically tune the key parameters of the model. The model performance is comprehensively evaluated using multiple indicators.
7. A method for tapping potential customers in a bank treasury cloud based on machine learning as described in claim 6, characterized in that, The metrics include: AUC, KS, and recall rate.
8. A method for tapping potential customers in a bank treasury cloud based on machine learning as described in claim 1, characterized in that, Step S5: Applying the model specifically includes: Target customer selection: Apply the trained optimal model to the candidate customer group at the latest time point, output the predicted probability score of each customer, and select the top-ranked customers for marketing. Customer value assessment: Taking into account the customer's purchase probability, financial situation, and transaction size, a weighted comprehensive scoring method is used to score customers; and customers are ranked and categorized based on the comprehensive score. Model interpretability analysis.
9. A method for tapping potential customers in a bank treasury cloud based on machine learning, as described in claim 8, is characterized in that... The model interpretability analysis specifically includes: Global interpretability: Calculate the average absolute SHAP value of each feature to obtain the global feature importance ranking, which can be used to guide business insights; Local interpretability: Generate the SHAP value for each feature based on the prediction results for each customer.