A baseline early warning and attribution advice generation system and method for insurance operations data

Abstract

The present application relates to the technical field of artificial intelligence and insurance data processing, and specifically discloses a baseline early warning and attribution suggestion generation system and method for insurance operation data. The technical scheme of the present application automatically monitors the changes of insurance operation indexes and generates attribution trigger signals by presetting baseline threshold rules, standardizes the modeling of index formulas, grouping dimensions and analysis modes with the help of scene configuration registration, automatically loads and adapts configurations and calls corresponding attribution algorithms according to trigger signals or natural language requests by AI agents, locates the root cause layer by layer along the dimension node with the highest contribution degree in the recursive drilling mode, and finally inputs the attribution calculation results into a large language model to automatically generate a contribution degree matrix, a structured analysis report and targeted management suggestions, realizing the full-process automation and standardization of insurance operation data anomaly monitoring, root cause positioning, report generation and management suggestion output, and improving the objectivity, timeliness and decision support capability of operation analysis.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and insurance data processing technology, and in particular to a baseline early warning and attribution suggestion generation system and method for insurance business data. Background Technology

[0002] As the digital transformation of the insurance industry deepens, operational data analysis is shifting from simply presenting results to in-depth causal analysis. Business managers not only need to understand the outcomes of changes in metrics, but also hope to quickly identify the driving factors behind these changes and obtain targeted management recommendations. However, existing attribution analysis systems have many shortcomings in supporting this need.

[0003] The manual analysis model currently widely used in the industry relies heavily on the personal experience of business personnel. Analysts typically use spreadsheets to manually filter data and perform simple year-on-year and month-on-month calculations to determine the reasons for indicator anomalies. This approach lacks standardized algorithmic support, and the analytical conclusions vary from person to person, making it difficult to distill them into a unified methodology. Consequently, the objectivity and reproducibility of the attribution results are hard to guarantee.

[0004] Meanwhile, traditional analytics tools typically only support single-dimensional data slicing, failing to automate recursive drill-down across multiple dimensions such as product, channel, personnel, and customer. Analysts must repeatedly switch between different reports for comparison to gradually narrow down the scope, making it difficult to quickly pinpoint the root cause combination of dimensions that truly triggers metric changes, significantly limiting analytical efficiency. Furthermore, insurance operating metrics are diverse, including absolute value metrics such as premium income, ratio metrics such as cost ratio, and composite metrics such as loss ratio. Their attribution calculation logics differ, requiring different algorithms such as chain substitution, logarithmic mean Dietrich index, or weighted average. However, existing systems lack a unified method selection mechanism and cannot automatically adapt the optimal algorithm based on metric characteristics.

[0005] More significantly, while existing business intelligence systems offer data dashboards and basic alert functions, secondary analysis still requires manual intervention after an alert is triggered. There is a significant time lag between problem discovery and root cause identification, resulting in a disconnect between baseline alerts and attribution analysis. In terms of results presentation, traditional attribution outcomes are often presented in fragmented tabular form, lacking structured analytical reports and data-driven management recommendations. Analysts still need to manually compile written reports after data calculations, leading to a heavy workload and poor timeliness. Although large language modeling technology brings new possibilities to business analysis, existing solutions often use it as a simple question-and-answer tool, failing to achieve a deep integration of the reasoning and orchestration capabilities of large language models with standardized attribution algorithms. Users struggle to flexibly describe their analytical needs using natural language and automatically complete the entire closed-loop process from data querying and attribution calculation to report generation.

[0006] Therefore, there is an urgent need to provide a technical solution to address the above problems. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a baseline early warning and attribution suggestion generation system and method for insurance business data.

[0008] Firstly, the present invention provides a baseline early warning and attribution suggestion generation system for insurance operation data, the technical solution of which is as follows: The data acquisition and early warning module is used to calculate the rate of change of indicators between the base period insurance operating data and the current period insurance operating data according to the preset baseline threshold rules, and generate an attribution trigger signal when the rate of change of indicators is greater than the preset threshold. The scenario configuration registration module is used to construct factor attribution scenario configuration models and dimensional attribution scenario configuration models. The factor attribution scenario configuration model includes indicator formulas and grouping dimension parameters, and the dimensional attribution scenario configuration model includes analysis mode parameters, which include independent analysis mode and recursive drill-down mode. The factor attribution scenario configuration model and the dimensional attribution scenario configuration model are respectively associated with scenario identifiers. The AI ​​intelligent agent orchestration module is used to call the attribution analysis skill guidance file to obtain attribution analysis process guidance according to the attribution trigger signal or the natural language attribution analysis request input by the user, load the corresponding factor attribution scenario configuration model or the dimension attribution scenario configuration model according to the attribution analysis process guidance and the scenario identifier, and output tool call instructions. The attribution calculation module is used to receive the tool call instruction, call the corresponding attribution algorithm according to the factor attribution scenario configuration model or the dimension attribution scenario configuration model to perform attribution calculation on the base period insurance operating data and the current period insurance operating data, and output the attribution calculation result; when the loaded configuration model is the dimension attribution scenario configuration model and the analysis mode parameter is the recursive drill-down mode, the attribution calculation of the next level dimension is performed on the dimension node with the highest contribution in the attribution calculation result; The results output module is used to generate a contribution matrix and an attribution analysis report based on the attribution calculation results, and input the attribution calculation results into a large language model to generate management suggestions.

[0009] The beneficial effects of the baseline early warning and attribution suggestion generation system for insurance operation data of the present invention are as follows: The system of this invention automatically monitors changes in insurance operation indicators and generates attribution trigger signals by setting preset baseline threshold rules. It standardizes the modeling of indicator formulas, grouping dimensions, and analysis modes through scenario configuration registration. The AI ​​agent automatically loads the appropriate configuration and calls the corresponding attribution algorithm based on the trigger signal or natural language request. In the recursive drill-down mode, it delves deeper into the root cause layer by layer along the dimension node with the highest contribution. Finally, the attribution calculation results are input into a large language model to automatically generate a contribution matrix, a structured analysis report, and targeted management suggestions. This solves the problems of manual analysis relying on experience, difficulty in multi-dimensional root cause location, chaotic adaptation of different indicator algorithms, separation between early warning and attribution, and poor timeliness of report compilation. It realizes full-process automation and standardization of insurance operation data anomaly monitoring, root cause location, report generation, and management suggestion output, improving the objectivity, timeliness, and decision support capabilities of operation analysis.

[0010] Based on the above solution, the baseline early warning and attribution suggestion generation system for insurance operation data of the present invention can be further improved as follows.

[0011] In one optional approach, the base period insurance operating data includes at least one of the following: operating data from the same period of the previous period, operating data from the same period last year, or operating data of a preset target value; the preset baseline threshold rule includes a rate of change threshold and a change amount threshold; and the data collection and early warning module is specifically used for: The attribution trigger signal is generated when the rate of change of the indicator is greater than the rate of change threshold or the amount of change of the indicator is greater than the amount of change threshold.

[0012] The advantages of adopting the above-mentioned optional methods are: further expanding the base period data to multiple reference benchmarks such as the same period of the previous period, the same period of the previous year, and the preset target value, and simultaneously setting two types of thresholds, namely the rate of change and the amount of change, to determine anomalies, making the early warning triggering conditions more flexible and able to cover the monitoring needs under different business analysis scenarios.

[0013] In an alternative approach, the factor attribution scenario configuration model further includes a list of factor variable names and factor order configuration information. The list of factor variable names is used to identify each factor variable in the indicator formula, and the factor order configuration information is used to determine the substitution order of each factor variable when performing factor contribution calculation.

[0014] The advantages of adopting the above optional approach are: further adding a list of factor variable names and a replacement order configuration in the factor attribution scenario configuration, so that the identification and calculation order of each factor in the indicator formula can be standardized and managed, and it is convenient to flexibly adjust the attribution calculation logic for different business indicators.

[0015] In one alternative approach, the attribution calculation module is specifically used for: Extract each factor variable name from the index formula and map each factor variable name to an alias that the expression engine can recognize; Based on the factor order configuration information, the base period value or current period value of each factor variable is substituted into the indicator formula to calculate the indicator value; According to the factor order configuration information, the value of each factor variable is replaced from the base period value to the current period value in turn. The index value after each replacement is calculated. The contribution of each factor variable is determined based on the index value calculated after each replacement, and the attribution calculation result is output.

[0016] The beneficial effects of adopting the above optional method are as follows: by further mapping factor variables to aliases that can be recognized by the expression engine, and automatically completing the successive substitution from the base period value to the current period value and the calculation of the indicator value according to the configuration order, the quantification process of factor contribution is formed into a standardized automatic execution link.

[0017] In one alternative approach, the attribution calculation module is specifically used for: The contribution of the i-th factor variable is calculated using the chain substitution method, and the formula for the contribution is: in, This represents the contribution of the i-th factor variable. This indicates the index value calculated by replacing the values ​​of the first i factor variables with their current period values ​​and keeping the remaining factor variables at their base period values, and then inputting these values ​​into the index formula. This means that the index value is calculated by replacing the values ​​of the first i-1 factor variables with the values ​​of the current period, while keeping the base period values ​​of the remaining factor variables, and then inputting the index into the index formula. The contribution of each factor variable is sorted from largest to smallest according to its absolute value, and the contribution percentage of each factor variable is calculated. The contribution percentage is the ratio of the absolute value of the contribution of the factor variable to the sum of the absolute values ​​of the contributions of all factor variables.

[0018] The beneficial effects of adopting the above optional method are: further use of the chain substitution method to accurately calculate the contribution of each factor variable, and sort and analyze the proportion according to the absolute value, so that business personnel can intuitively identify the core driving factors that have the greatest impact on the fluctuation of indicators.

[0019] In an alternative embodiment, the attribution calculation module is further configured to: When the log-mean Dichotomy index method is chosen, the contribution of the i-th factor variable is calculated according to the following formula: in, This represents the contribution of the i-th factor variable calculated using the log-mean Dijkstra index method. This indicates the current period index value calculated by substituting the current period values ​​of each factor variable into the index formula. This represents the base period index value calculated by substituting the base period values ​​of each factor variable into the index formula. This represents the current period value of the i-th factor variable. This represents the base period value of the i-th factor variable; When the indicator type in the indicator formula is a ratio-type indicator, the change in the indicator between the base period insurance operating data and the current period insurance operating data is decomposed into a horizontal effect component and a structural effect component according to the weighted average method.

[0020] The beneficial effects of adopting the above optional methods are as follows: the introduction of the logarithmic mean Dijkstra's index method and the weighted average method are applicable to the attribution decomposition of multiplicative indicators and ratio indicators, respectively, so that different types of operating indicators can be matched with scientific and reasonable algorithm models.

[0021] In one alternative approach, the recursive drilling mode includes a first level of drilling and at least one next level of drilling; The attribution calculation module is specifically used to: in the recursive drill-down mode, after completing the attribution calculation of the first level dimension, obtain the contribution of each dimension node, determine the dimension node with the largest contribution value as the dimension node to be drilled down, and perform attribution calculation on the dimension node to be drilled down according to the next level dimension until the preset drill-down depth is reached or the contribution of each dimension node in the current level is less than the preset contribution threshold. The AI ​​intelligent agent orchestration module is specifically used to: identify the number of drilling levels and contribution filtering strategy specified in the natural language attribution analysis request in the attribution analysis process guidance, use the number of drilling levels as the preset drilling depth, and use the contribution threshold in the contribution filtering strategy as the preset contribution threshold.

[0022] The advantages of adopting the above optional approach are: it further supports multi-level dimensional analysis in recursive drill-down mode, and can automatically perform layer-by-layer tracing based on the drill-down depth and contribution filtering conditions specified in the natural language request, so that root cause localization can be extended from macro-level dimensions to micro-level nodes.

[0023] Secondly, this invention provides a method for generating baseline early warning and attribution suggestions for insurance operation data. The technical solution of this method is as follows: The rate of change of indicators between the base period insurance operating data and the current period insurance operating data is calculated according to the preset baseline threshold rule, and an attribution trigger signal is generated when the rate of change of indicators is greater than the preset threshold. Construct a factor attribution scenario configuration model and a dimension attribution scenario configuration model. The factor attribution scenario configuration model includes indicator formulas and grouping dimension parameters. The dimension attribution scenario configuration model includes analysis mode parameters, including independent analysis mode and recursive drill-down mode. The factor attribution scenario configuration model and the dimension attribution scenario configuration model are respectively associated with scenario identifiers. Based on the attribution trigger signal or the natural language attribution analysis request input by the user, the attribution analysis skill guidance file is invoked to obtain the attribution analysis process guidance. Based on the attribution analysis process guidance and the scenario identifier, the corresponding factor attribution scenario configuration model or the dimension attribution scenario configuration model is loaded, and the tool invocation command is output. Upon receiving the tool invocation instruction, the corresponding attribution algorithm is invoked according to the factor attribution scenario configuration model or the dimension attribution scenario configuration model to perform attribution calculations on the base period insurance operating data and the current period insurance operating data, and the attribution calculation results are output; when the loaded configuration model is the dimension attribution scenario configuration model and the analysis mode parameter is the recursive drill-down mode, the next level dimension attribution calculation is performed on the dimension node with the highest contribution in the attribution calculation results; Based on the attribution calculation results, a contribution matrix and an attribution analysis report are generated, and the attribution calculation results are input into a large language model to generate management recommendations.

[0024] The beneficial effects of the baseline early warning and attribution suggestion generation method for insurance operation data of the present invention are as follows: The method of this invention automatically monitors changes in insurance operation indicators and generates attribution trigger signals by setting preset baseline threshold rules. It standardizes the modeling of indicator formulas, grouping dimensions, and analysis modes through scenario configuration registration. The AI ​​agent automatically loads the appropriate configuration and calls the corresponding attribution algorithm based on the trigger signal or natural language request. In the recursive drill-down mode, it delves deeper into the root cause layer by layer along the dimension node with the highest contribution. Finally, the attribution calculation results are input into a large language model to automatically generate a contribution matrix, a structured analysis report, and targeted management suggestions. This method solves the problems of manual analysis relying on experience, difficulty in multi-dimensional root cause location, chaotic adaptation of different indicator algorithms, separation between early warning and attribution, and poor timeliness of report compilation. It realizes full-process automation and standardization of insurance operation data anomaly monitoring, root cause location, report generation, and management suggestion output, improving the objectivity, timeliness, and decision support capabilities of operation analysis.

[0025] Thirdly, the technical solution of an electronic device according to the present invention is as follows: The invention includes a memory, a processor, and a program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps of the baseline warning and attribution suggestion generation method for insurance business data as described in this invention.

[0026] Fourthly, the technical solution of a computer-readable storage medium provided by the present invention is as follows: The computer-readable storage medium stores instructions that, when read, cause the computer-readable storage medium to perform the steps of the baseline warning and attribution suggestion generation method for insurance business data as described in this invention.

[0027] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other objects, features and advantages of the present invention more obvious and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0028] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a schematic diagram of an embodiment of the baseline early warning and attribution suggestion generation system for insurance business data according to the present invention; Figure 2 This is a schematic diagram of the overall architecture; Figure 3 This is a flowchart illustrating an embodiment of a method for generating baseline early warning and attribution suggestions for insurance business data according to the present invention. Figure 4 This is a schematic diagram of an embodiment of an electronic device according to the present invention. Detailed Implementation

[0029] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0030] Figure 1 This diagram illustrates the structure of an embodiment of an insurance business data baseline early warning and attribution suggestion generation system 100 provided by the present invention. Figure 1 As shown, the baseline early warning and attribution suggestion generation system 100 for insurance operation data includes: The data acquisition and early warning module 101 is used to calculate the rate of change of indicators between the base period insurance operating data and the current period insurance operating data according to the preset baseline threshold rule, and generate an attribution trigger signal when the rate of change of indicators is greater than the preset threshold.

[0031] Among them, the preset baseline threshold rule refers to the pre-set judgment conditions used to determine whether insurance operating indicators have experienced abnormal fluctuations. The rule includes at least one of the rate of change threshold and the amount of change threshold, which is used to measure whether the degree of difference between the insurance operating data of the base period and the insurance operating data of the current period reaches the condition for triggering attribution analysis. For example, attribution analysis is triggered when the premium income decreases by more than 5% year-on-year.

[0032] The base period insurance operating data refers to the set of insurance operating indicator values ​​used as a benchmark. This data is compared with the actual operating data of the current period to determine the degree of change in the indicators. For example, the premium income of 10 million yuan on July 1, 2026, is used as the base period insurance operating data. The current period insurance operating data refers to the set of actual insurance operating indicator values ​​for the current analysis period. This data is compared with the base period insurance operating data to calculate the rate of change in the indicators. For example, the premium income of 9.2 million yuan on July 2, 2026, is used as the current period insurance operating data.

[0033] The indicator change rate refers to a quantitative value representing the difference between the base period insurance operating data and the current period insurance operating data. This value is calculated by subtracting the base period insurance operating data from the current period insurance operating data and then dividing by the base period insurance operating data. For example, if premium income changes from 10 million yuan on July 1, 2026 to 9.2 million yuan on July 2, 2026, the indicator change rate is -8%. The preset threshold refers to a pre-set critical value used to determine whether the indicator change rate reaches the condition for triggering attribution analysis. For example, the preset threshold is set to -5%, and attribution analysis is triggered when the premium income indicator change rate of -8% is less than the preset threshold of -5%.

[0034] The attribution trigger signal refers to a data message generated when the rate of change of an indicator exceeds a preset threshold, which is used to initiate the attribution analysis process. For example, when the rate of change of the premium income indicator is -8% and exceeds the preset threshold of -5%, the data acquisition and early warning module 101 generates an attribution trigger signal and transmits it to the AI ​​intelligent agent orchestration module 103.

[0035] Specifically, the system acquires base period insurance operating data and current period insurance operating data, subtracts the base period insurance operating data from the current period insurance operating data, and divides the result by the base period insurance operating data to calculate the indicator change rate. The indicator change rate is then compared with a preset threshold, and when the indicator change rate is greater than the preset threshold, an attribution trigger signal is generated.

[0036] The scenario configuration registration module 102 is used to construct a factor attribution scenario configuration model and a dimensional attribution scenario configuration model. The factor attribution scenario configuration model includes indicator formulas and grouping dimension parameters. The dimensional attribution scenario configuration model includes analysis mode parameters, which include independent analysis mode and recursive drill-down mode. The factor attribution scenario configuration model and the dimensional attribution scenario configuration model are respectively associated with scenario identifiers.

[0037] The factor attribution scenario configuration model refers to a predefined set of parameters for the factor attribution analysis type. This set includes indicator formulas and grouping dimension parameters, used to instruct the attribution calculation module 104 to calculate factor contribution according to the specified formulas and grouping methods. For example, for premium income change analysis, a factor attribution scenario configuration model is constructed, with the indicator formula set to "premium income = number of policies × average premium per policy," and the grouping dimension parameter set to "grouped by insurance type." The dimension attribution scenario configuration model refers to a predefined set of parameters for the dimension attribution analysis type. This set includes analysis mode parameters, used to instruct the attribution calculation module 104 to perform multi-dimensional attribution analysis according to either an independent analysis mode or a recursive drill-down mode. For example, for multi-dimensional analysis of premium decline in region A, a dimension attribution scenario configuration model is constructed, with the analysis mode parameter set to a recursive drill-down mode, where the drill-down dimension order is from the institutional dimension down to the channel dimension and then down to the personnel dimension.

[0038] The indicator formula refers to an expression describing the mathematical relationship between operating indicators and various influencing factors. This expression includes factor variable names and operators, used to calculate the base period indicator value and the current period indicator value. For example, the indicator formula is "Premium Income = Number of Policies × Average Premium per Policy," where the number of policies and average premium per policy are factor variable names. The grouping dimension parameter specifies the dimensions by which the data is grouped and aggregated in the attribution analysis. This parameter determines how the attribution calculation module 104 groups the data when calculating factor contributions. For example, if the grouping dimension parameter is set to "Group by Insurance Type," the attribution calculation module 104 performs attribution calculations separately for life insurance, health insurance, accident insurance, and other types of insurance. The analysis mode parameter specifies the execution method of the dimensional attribution analysis. This parameter includes two values: independent analysis mode and recursive drill-down mode, used to instruct the attribution calculation module 104 on which mode to perform multi-dimensional attribution analysis. For example, if the analysis mode parameter is set to recursive drill-down mode, the attribution calculation module 104 performs attribution calculations layer by layer in the order of organization → channel → personnel.

[0039] The independent analysis mode refers to an execution method of dimensional attribution analysis. In this mode, the attribution calculation module 104 independently performs attribution calculations for each dimension at the current level and outputs the contribution of each dimension node, without recursively tracing downwards. For example, when the analysis mode parameter is set to independent analysis mode, the attribution calculation module 104 calculates the premium contribution of each insurance type, channel, and personnel in region A and then stops, without drilling down further. The recursive drill-down mode refers to an execution method of dimensional attribution analysis. In this mode, after completing the attribution calculation of the first level dimension, the attribution calculation module 104 automatically locates the dimension node with the highest contribution and continues to perform attribution calculations according to the next level dimension, deepening layer by layer until the preset drill-down depth is reached or the stopping condition is met. For example, when the analysis mode parameter is set to recursive drill-down mode, the attribution calculation module 104 first calculates the contribution of each institution under the institution dimension, locates Branch A with the highest contribution, and then continues to calculate the contribution of each channel of Branch A under the channel dimension.

[0040] The scenario identifier refers to an identifier used to uniquely identify the factor attribution scenario configuration model or the dimension attribution scenario configuration model. The AI ​​agent orchestration module 103 loads the corresponding scenario configuration model from the scenario configuration registration module 102 through the identifier. For example, the scenario identifier "FACTOR_PREMIUM_001" is assigned to the premium income change factor attribution scenario configuration model, and the AI ​​agent orchestration module 103 loads the corresponding configuration model according to the scenario identifier.

[0041] Specifically, a factor attribution scenario configuration model and a dimensional attribution scenario configuration model are constructed. The factor attribution scenario configuration model includes indicator formulas and grouping dimension parameters, while the dimensional attribution scenario configuration model includes analysis mode parameters, which include independent analysis mode and recursive drill-down mode. Corresponding scenario identifiers are configured for the factor attribution scenario configuration model and the dimensional attribution scenario configuration model, respectively.

[0042] The AI ​​intelligent agent orchestration module 103 is used to call the attribution analysis skill guidance file to obtain attribution analysis process guidance according to the attribution trigger signal or the natural language attribution analysis request input by the user, load the corresponding factor attribution scenario configuration model or the dimension attribution scenario configuration model according to the attribution analysis process guidance and the scenario identifier, and output tool call instructions.

[0043] Among them, the natural language attribution analysis request refers to the attribution analysis requirement description submitted by the user to the system in natural language form. The description includes the analysis scope, indicator definition, drill-down dimensions, and filtering conditions. After semantic understanding by the AI ​​intelligent agent orchestration module 103, it is converted into a tool call instruction. For example, the user inputs "Please analyze the reasons for the decline in premiums in region A this month, and conduct drill-down analysis along the three dimensions of product, channel, and personnel" as a natural language attribution analysis request.

[0044] The attribution analysis skills guidance document refers to a pre-written, structured guidance document containing the attribution analysis operation process. This document encodes the execution steps of attribution analysis into executable process guidelines for the AI ​​agent, guiding the AI ​​agent orchestration module 103 to determine the analysis method, select parameters, and schedule tools. For example, the attribution analysis skills guidance document specifies that when a user request involves multi-dimensional analysis, the dimensional attribution analysis tool should be invoked and recursive drill-down mode should be enabled. The attribution analysis process guide refers to a sequence of specific execution steps generated by the AI ​​agent orchestration module 103 based on the guidance content of the attribution analysis skills guidance document for the current analysis task. This guide includes method selection information, parameter configuration information, and tool invocation order information. For example, the AI ​​agent orchestration module 103 generates the following process guide based on the attribution analysis skills guidance document: determining the analysis method as dimensional attribution, loading the dimensional attribution scenario configuration model, invoking the dimensional attribution analysis tool, and setting the recursive drill-down mode.

[0045] The tool invocation instruction refers to the structured command issued by the AI ​​agent orchestration module 103 to the attribution calculation module 104 for performing attribution calculation. The instruction includes the target tool identifier and input parameters. For example, the AI ​​agent orchestration module 103 issues a tool invocation instruction to the attribution calculation module 104, specifies the dimension attribution analysis tool to be invoked, passes in the base period data and the current period data, and sets the recursive drill-down mode.

[0046] Specifically, based on attribution trigger signals or in response to natural language attribution analysis requests input by users, the system invokes attribution analysis skills guidance documents and obtains attribution analysis process guidelines from these documents. Based on the attribution analysis process guidelines and scenario identifiers, it loads the factor attribution scenario configuration model or dimensional attribution scenario configuration model corresponding to the scenario identifier from the pre-built scenario configuration model. Finally, based on the attribution analysis process guidelines and the loaded scenario configuration model, it outputs tool invocation instructions.

[0047] The attribution calculation module 104 is used to receive the tool call instruction, call the corresponding attribution algorithm according to the factor attribution scenario configuration model or the dimension attribution scenario configuration model to perform attribution calculation on the base period insurance operating data and the current period insurance operating data, and output the attribution calculation result; when the loaded configuration model is the dimension attribution scenario configuration model and the analysis mode parameter is the recursive drill-down mode, the attribution calculation of the dimension node with the highest contribution in the attribution calculation result is performed on the next level dimension.

[0048] The attribution algorithm refers to a mathematical method used to calculate the contribution of each factor variable or dimension node to changes in operating indicators. Methods include the chain substitution method, the logarithmic mean Dijkstra's index method, and the weighted average method. For example, for a multiplicative indicator where premium income = number of policies × average premium per policy, the attribution calculation module 104 uses the chain substitution method to calculate the contribution of each of the number of policies and the average premium per policy. The attribution calculation result refers to the data set output by the attribution calculation module 104 after executing the attribution algorithm, containing the contribution of each factor variable or dimension node. This result serves as input data for the result output module 105 to generate the contribution matrix and attribution analysis report. For example, the attribution calculation result output by the attribution calculation module 104 includes a contribution of -600,000 yuan for the number of policies and a contribution of -200,000 yuan for the average premium per policy.

[0049] The attribution calculation of the next level dimension refers to the process in which, under the recursive drill-down mode, the attribution calculation module 104 re-executes the attribution calculation for the dimension node with the highest current contribution according to the predefined next level dimension. For example, after locating Branch A with the highest contribution under the organization dimension, the attribution calculation module 104 performs attribution calculations for Branch A's life insurance channel, bancassurance channel, and online sales channel under the channel dimension.

[0050] Specifically, based on the tool call instructions, the corresponding attribution algorithm is determined according to the loaded factor attribution scenario configuration model or dimension attribution scenario configuration model. The base period insurance operating data and the current period insurance operating data are used as inputs to perform attribution calculations, and the attribution calculation results are output. If the loaded configuration model is a dimension attribution scenario configuration model and the analysis mode parameter is a recursive drill-down mode, the dimension node with the highest contribution is extracted from the attribution calculation results, and the attribution calculation is performed again on the dimension node according to the next level dimension.

[0051] The result output module 105 is used to generate a contribution matrix and an attribution analysis report based on the attribution calculation results, and input the attribution calculation results into the large language model to generate management suggestions.

[0052] The contribution matrix refers to a tabular structure displaying the attribution calculation results. The matrix groups each factor variable or dimension node according to its positive and negative contributions, showing the contribution values ​​and percentages. For example, in the generated contribution matrix, the positive contribution list shows an average premium contribution of -200,000 yuan and a contribution percentage of 25%, while the negative contribution list shows a policy number contribution of -600,000 yuan and a contribution percentage of 75%. The attribution analysis report refers to a structured analysis document generated based on the attribution calculation results. The document includes a summary of the attribution results, the contribution matrix, and root cause analysis conclusions. For example, the generated attribution analysis report includes a summary of the reasons for the decline in premium income, a contribution matrix of the number of policies and the average premium per policy, and root cause analysis conclusions regarding the channel structure imbalance in Region A.

[0053] Among them, management recommendations refer to structured decision recommendations automatically generated by the big language model based on the attribution calculation results. These recommendations propose specific improvement directions and actionable measures for the operational problems identified in the attribution analysis. For example, based on the attribution result of a policy contribution of -600,000 yuan, the big language model generates the management recommendation "Focus on the channel expansion of life insurance products in region A, strengthen agent recruitment and training, and increase the number of policies sold."

[0054] Specifically, a contribution matrix is ​​generated based on the attribution calculation results. The contribution matrix is ​​arranged according to positive and negative contributions, with the contribution values ​​and proportions of each factor variable or dimension node. An attribution analysis report is generated based on the attribution calculation results, which includes a summary of the attribution results, the contribution matrix, and the root cause analysis conclusions. The attribution calculation results are input into a large language model, which combines the attribution calculation results to generate management recommendations. The management recommendations include at least one of the following: product structure adjustment recommendations, channel optimization recommendations, personnel management recommendations, and customer segmentation recommendations.

[0055] The technical solution of this embodiment automatically monitors changes in insurance operation indicators and generates attribution trigger signals by setting preset baseline threshold rules. It standardizes the modeling of indicator formulas, grouping dimensions, and analysis modes through scenario configuration registration. The AI ​​agent automatically loads the adaptation configuration and calls the corresponding attribution algorithm according to the trigger signal or natural language request. In the recursive drill-down mode, it delves deeper into the root cause layer by layer along the dimension node with the highest contribution. Finally, the attribution calculation results are input into the large language model to automatically generate a contribution matrix, a structured analysis report, and targeted management suggestions. This solves the problems of manual analysis relying on experience, difficulty in multi-dimensional root cause location, chaotic adaptation of different indicator algorithms, separation between early warning and attribution, and poor timeliness of report compilation. It realizes full-process automation and standardization of insurance operation data anomaly monitoring, root cause location, report generation, and management suggestion output, and improves the objectivity, timeliness, and decision support capabilities of operation analysis.

[0056] In one optional approach, the base period insurance operating data includes at least one of the following: operating data from the same period of the previous period, operating data from the same period last year, or operating data of a preset target value; the preset baseline threshold rule includes a rate of change threshold and a change amount threshold; and the data acquisition and early warning module 101 is specifically used for: The attribution trigger signal is generated when the rate of change of the indicator is greater than the rate of change threshold or the amount of change of the indicator is greater than the amount of change threshold.

[0057] The "previous period's operating data" refers to the set of insurance operating indicators from the previous operating cycle adjacent to the current period's operating data. This data serves as a source of base period insurance operating data for calculating the month-on-month change rate. For example, premium income of 10 million yuan on July 1, 2026, is used as the previous period's operating data, and compared with premium income of 9.2 million yuan on July 2, 2026, for a month-on-month comparison. The "last year's operating data" refers to the set of insurance operating indicators from the same calendar cycle as the current period's operating data but from a year earlier. This data serves as a source of base period insurance operating data for calculating the year-on-year change rate. For example, premium income of 10.5 million yuan on July 2, 2025, is used as the last year's operating data, and compared with premium income of 9.2 million yuan on July 2, 2026, for a year-on-year comparison. Preset target value operating data refers to a set of pre-set insurance operating indicators that serve as assessment benchmarks or planned targets. The data is used as a source of base period insurance operating data to calculate the degree of deviation from the target value. For example, the premium income budget target of 9.8 million yuan on July 2, 2026 is used as the preset target value operating data and compared with the actual premium income of 9.2 million yuan.

[0058] The rate of change threshold refers to the critical percentage value used in the preset baseline threshold rules to determine whether the rate of change of the indicator has reached the trigger condition. For example, the rate of change threshold is set to -5%, and attribution analysis is triggered when the rate of change of the premium income indicator is -8%, which is less than the rate of change threshold of -5%. The amount of change threshold refers to the critical value used in the preset baseline threshold rules to determine whether the absolute value of the change of the indicator has reached the trigger condition. For example, the amount of change threshold is set to -500,000 yuan, and attribution analysis is triggered when the amount of change of premium income is -800,000 yuan, which is less than the amount of change threshold of -500,000 yuan.

[0059] Specifically, the calculated rate of change of the indicator is compared with a preset rate of change threshold, and the calculated amount of change of the indicator is compared with a preset amount of change threshold. When the rate of change of the indicator is greater than the rate of change threshold, or when the amount of change of the indicator is greater than the amount of change threshold, the attribution triggering condition is determined to be met, and an attribution triggering signal is generated to start the attribution analysis process.

[0060] Among the above-mentioned optional methods, the base period data is further expanded to include multiple reference benchmarks such as the same period of the previous period, the same period of the previous year, and preset target values. At the same time, two types of thresholds, namely the rate of change and the amount of change, are set to determine anomalies, making the early warning triggering conditions more flexible and able to cover the monitoring needs under different business analysis scenarios.

[0061] In an alternative approach, the factor attribution scenario configuration model further includes a list of factor variable names and factor order configuration information. The list of factor variable names is used to identify each factor variable in the indicator formula, and the factor order configuration information is used to determine the substitution order of each factor variable when performing factor contribution calculation.

[0062] The factor variable name list refers to the collection of all factor variable names in the indicator formula. This list identifies each factor variable in the indicator formula and supports the extraction of factor variable information from the formula. For example, the factor variable name list includes two factor variable names: "number of policies" and "average premium per policy," used to identify two influencing factors in the premium income indicator formula. The factor order configuration information refers to the configuration data specifying the substitution order of each factor variable when performing the chain substitution method. This configuration information determines the order in which the attribution calculation module 104 replaces the base period values ​​with the current period values ​​for each factor variable. For example, the factor order configuration information specifies that "number of policies" is substituted first, followed by "average premium per policy," and the attribution calculation module 104 performs the substitution calculations sequentially.

[0063] Among them, factor variables refer to the components in the indicator formula that affect the value of the operating indicator. Each factor variable contains three attributes: factor variable name, base period value, and current period value. The contribution of the factor variable is calculated by the attribution algorithm. For example, the base period value of the factor variable "number of policies" is 1,000 policies and the current period value is 920 policies. The base period value of the factor variable "average premium per policy" is 10,000 yuan and the current period value is 10,000 yuan.

[0064] The substitution order refers to the order in which factor variables are replaced during the chain substitution method. The attribution calculation module 104 replaces the values ​​of factor variables from the base period value to the current period value one by one in sequence and calculates the indicator value after each replacement. For example, the substitution order is to replace "number of policies" first and then "average premium per policy". The attribution calculation module 104 first replaces the number of policies from 1,000 to 920 to calculate the intermediate indicator value, and then replaces the average premium per policy from 10,000 yuan to 10,000 yuan to calculate the final indicator value.

[0065] In the above optional methods, a list of factor variable names and a configuration for the alternative order can be added to the factor attribution scenario configuration, so that the identification and calculation order of each factor in the indicator formula can be standardized and managed, making it easier to flexibly adjust the attribution calculation logic for different business indicators.

[0066] In an alternative embodiment, the attribution calculation module 104 is specifically used for: Extract each factor variable name from the index formula and map each factor variable name to an alias that the expression engine can recognize.

[0067] Here, the factor variable name refers to the text identifier of the factor variable in the indicator formula. The identifier is used to identify and extract the factor variable from the indicator formula and supports mapping the factor variable to an alias that the expression engine can recognize. For example, the "number of policies" and "average premium per policy" in the indicator formula "premium income = number of policies × average premium per policy" are the factor variable names.

[0068] The expression engine refers to a software component used to parse and calculate mathematical expressions. The component can recognize variable names in the form of aliases and substitute the variable values ​​into the expression for evaluation. For example, the expression engine maps the factor variable name "number of policies" to the alias "a" and "average premium per policy" to the alias "b", and substitutes the values ​​corresponding to the aliases into the expression "a×b" to calculate the premium income.

[0069] Specifically, the process identifies the names of each factor variable contained in the indicator formula, maps each identified factor variable name to an alias that the expression engine can recognize, and each alias uniquely corresponds to a factor variable name in the expression engine.

[0070] Based on the factor order configuration information, the base period value or current period value of each factor variable is substituted into the indicator formula to calculate the indicator value.

[0071] The base period value refers to the specific numerical value of a factor variable in the base period insurance operating data, which serves as the starting reference value for substitution calculations in the attribution algorithm. For example, the base period value for the factor variable "number of policies" is 1000 policies as of July 1, 2026. The current period value refers to the specific numerical value of a factor variable in the current period insurance operating data, which serves as the target replacement value for substitution calculations in the attribution algorithm. For example, the current period value for the factor variable "number of policies" is 920 policies as of July 2, 2026.

[0072] The indicator value refers to the specific operating indicator value obtained by substituting the current value of each factor variable into the indicator formula. The value is used to calculate the contribution of each factor variable. For example, the indicator value obtained by substituting the number of insurance policies of 1,000 and the average premium of 10,000 yuan into the formula "premium income = number of insurance policies × average premium" is 10 million yuan.

[0073] Specifically, the factor variables to be substituted are determined sequentially according to the substitution order specified in the factor order configuration information. In each substitution process, the values ​​of the factor variables that have been determined to be replaced are replaced with the corresponding current period values, and the values ​​of the factor variables that have not been replaced are replaced with the corresponding base period values. These values ​​are then substituted into the corresponding positions in the indicator formula to calculate the indicator value after this substitution.

[0074] According to the factor order configuration information, the value of each factor variable is replaced from the base period value to the current period value in turn. The index value after each replacement is calculated. The contribution of each factor variable is determined based on the index value calculated after each replacement, and the attribution calculation result is output.

[0075] The replaced indicator value refers to the indicator value obtained by recalculating the indicator value after replacing the base period value with the current period value according to the factor order configuration information and then entering it into the indicator formula. This value is used to compare with the indicator value before replacement to determine the contribution of the replaced factor variable. For example, when the number of insurance policies is replaced from 1,000 to 920, and the average premium per policy remains at 10,000 yuan, the replaced indicator value calculated by substituting into the formula is 9.2 million yuan. The contribution refers to the value of a certain factor variable or dimension node to the total change in the operating indicator. This value is obtained by comparing the difference in indicator values ​​before and after the factor variable is replaced or by calculating the corresponding formula of other attribution algorithms. It is used to measure the degree of influence of each factor on the change of the indicator. For example, if the number of insurance policies decreases from 1,000 to 920, resulting in a decrease in premium income from 10 million yuan to 9.2 million yuan, the contribution is -800,000 yuan.

[0076] Specifically, following the substitution order specified in the factor order configuration information, the following substitution operations are performed for each factor variable in sequence: the value of the factor variable is replaced from the base period value to the current period value, while keeping the values ​​of other factor variables unchanged; the replaced factor variable values ​​are substituted into the indicator formula to calculate the replaced indicator value; after each substitution operation, the contribution of the factor variable is determined based on the change in the replaced indicator value relative to the original indicator value; after completing the substitution operations for all factor variables and determining the contribution of each factor variable, the attribution calculation results are output.

[0077] In the above optional methods, the factor variables are further mapped to aliases that the expression engine can recognize, and the successive substitution from the base period value to the current period value and the calculation of the indicator value are automatically completed in the configured order, so that the quantification process of factor contribution becomes a standardized automatic execution link.

[0078] In an alternative embodiment, the attribution calculation module 104 is specifically used for: The contribution of the i-th factor variable is calculated using the chain substitution method, and the formula for the contribution is: in, This represents the contribution of the i-th factor variable. This indicates the index value calculated by replacing the values ​​of the first i factor variables with their current period values ​​and keeping the remaining factor variables at their base period values, and then inputting these values ​​into the index formula. This indicates the index value calculated by replacing the values ​​of the first i-1 factor variables with the current period values ​​and keeping the base period values ​​of the remaining factor variables, and then inputting them into the index formula.

[0079] The contribution of each factor variable is sorted from largest to smallest according to its absolute value, and the contribution percentage of each factor variable is calculated. The contribution percentage is the ratio of the absolute value of the contribution of the factor variable to the sum of the absolute values ​​of the contributions of all factor variables.

[0080] Among the above-mentioned optional methods, the chain substitution method is further used to accurately calculate the contribution of each factor variable, and sort and analyze the proportion according to the absolute value, so that business personnel can intuitively identify the core driving factors that have the greatest impact on the fluctuation of indicators.

[0081] In an alternative embodiment, the attribution calculation module 104 is further configured to: When the log-mean Dichotomy index method is chosen, the contribution of the i-th factor variable is calculated according to the following formula: in, This represents the contribution of the i-th factor variable calculated using the log-mean Dijkstra index method. This indicates the current period index value calculated by substituting the current period values ​​of each factor variable into the index formula. This represents the base period index value calculated by substituting the base period values ​​of each factor variable into the index formula. This represents the current period value of the i-th factor variable. This represents the base period value of the i-th factor variable.

[0082] When the indicator type in the indicator formula is a ratio-type indicator, the change in the indicator between the base period insurance operating data and the current period insurance operating data is decomposed into a horizontal effect component and a structural effect component according to the weighted average method.

[0083] The horizontal effect component refers to the portion contributed by the changes in the values ​​of the internal factor variables within each dimension group when decomposing the overall change of ratio-type indicators in the weighted average method. This component reflects the degree of influence of changes in the operating efficiency or quality of each dimension on the overall indicator while keeping the weights constant. For example, the loss ratio of each branch of Company A changed from 30% in the base period to 32% in the current period, an overall increase of 2 percentage points. The horizontal effect component is +1.5 percentage points, indicating that the general increase in the loss ratio of each branch is the main internal factor for the increase in the loss ratio.

[0084] The structural effect component refers to the portion contributed by the change in the weighting of each dimension when decomposing the overall change of ratio-type indicators in the weighted average method. This component reflects the degree of influence of the change in the business structure ratio of each dimension on the overall indicator. For example, the loss ratio of each institution of Branch A changed from 30% in the base period to 32% in the current period, an overall increase of 2 percentage points. The structural effect component is +0.5 percentage points, indicating that the structural change of the increased premium scale of institutions with high loss ratios also had a certain impact on the increase in loss ratio.

[0085] Specifically, when the indicator type of the indicator formula is identified as a ratio-type indicator, the weighted average method is called to decompose the change in the indicator between the base period insurance operating data and the current period insurance operating data, and the decomposition yields the horizontal effect component and the structural effect component.

[0086] Among the above-mentioned optional methods, the logarithmic mean Dijkstra's index method and the weighted average method are further introduced, which are applicable to the attribution decomposition of multiplicative indicators and ratio indicators, respectively, so that different types of business indicators can be matched with scientific and reasonable algorithm models.

[0087] In one alternative approach, the recursive drilling mode includes a first level of drilling and at least one next level of drilling; The attribution calculation module 104 is specifically used to: in the recursive drill-down mode, after completing the attribution calculation of the first level dimension, obtain the contribution of each dimension node, determine the dimension node with the largest contribution value as the dimension node to be drilled down, and perform attribution calculation on the dimension node to be drilled down according to the next level dimension until the preset drill-down depth is reached or the contribution of each dimension node in the current level is less than the preset contribution threshold.

[0088] In this context, the first level of drill-down refers to the first attribution calculation performed from the top-level dimension in a recursive drill-down model. This level corresponds to the coarsest-grained dimension in the dimension hierarchy. For example, in the dimension hierarchy of organization → channel → personnel, the first level of drill-down is the attribution calculation under the organization dimension, calculating the contribution of each organization to the total change in premium income. The next level of drill-down refers to the attribution calculation performed on the next level sub-dimension of the current dimension node in a recursive drill-down model. This level corresponds to a more granular dimension in the dimension hierarchy than the current level. For example, after locating Branch A under the organization dimension, the next level of drill-down is to perform attribution calculations on each channel of Branch A under the channel dimension.

[0089] Among them, the dimension node to be drilled down refers to the dimension node that is determined to have the highest contribution and needs further tracking and analysis after the attribution calculation of the current level in the recursive drill-down mode. The node will be used as the input object for the next level of drill-down. For example, in the attribution calculation of the organization dimension, the contribution of Branch A is -800,000 yuan, which is the largest among all organizations. Branch A is the dimension node to be drilled down, and the system will continue to perform attribution calculation on this node under the channel dimension.

[0090] The preset drilling depth refers to the maximum number of recursive drilling levels that are pre-set. Drilling stops when the recursively executed drilling levels reach the desired depth. For example, if the preset drilling depth is set to 3, the attribution calculation module 104 will stop drilling after executing the third level in the order of organization → channel → personnel. The preset contribution threshold refers to a pre-set critical value used to determine whether the contribution of the current level's dimension nodes is sufficient to support continued drilling. Drilling stops when the absolute value of the contribution of all dimension nodes at a certain level is less than the threshold. For example, if the preset contribution threshold is set to 100,000 yuan, the attribution calculation module 104 will stop drilling when the absolute value of the contribution of all dimension nodes at a certain level is less than 100,000 yuan.

[0091] Specifically, in the recursive drill-down mode, the attribution calculation of the first-level dimension is completed first, and the contribution of each dimension node under this level is obtained. Then, the dimension node with the largest contribution value is selected from the contribution values ​​of each dimension node as the dimension node to be drilled down. Then, attribution calculation is performed on the dimension node to be drilled down according to the next level dimension. The above selection and attribution calculation operations are continuously executed until the number of drill-down levels currently executed reaches the preset drill-down depth, or the contribution of each dimension node in the current level is less than the preset contribution threshold.

[0092] The AI ​​intelligent agent orchestration module 103 is specifically used to: identify the number of drilling levels and contribution filtering strategy specified in the natural language attribution analysis request in the attribution analysis process guidance, use the number of drilling levels as the preset drilling depth, and use the contribution threshold in the contribution filtering strategy as the preset contribution threshold.

[0093] The drill-down level refers to the maximum number of recursive drill-down levels specified by the user in the natural language attribution analysis request. This value is identified by the AI ​​agent orchestration module 103 and used as the preset drill-down depth. For example, if the user specifies "drill-down analysis along the three dimensions of product, channel, and personnel" in the request, the AI ​​agent orchestration module 103 identifies the drill-down level as 3. The contribution filtering strategy refers to the strategy rules specified by the user through the natural language attribution analysis request or scenario configuration for filtering dimension nodes to be drilled down. The strategy includes the number of dimension nodes to be filtered or contribution threshold conditions. For example, if the user specifies "automatic filtering of the top 5 dimension nodes by contribution" in the request, the AI ​​agent orchestration module 103 identifies the contribution filtering strategy as filtering the top 5 dimension nodes by contribution as the objects to be drilled down.

[0094] Specifically, the attribution analysis process guide identifies natural language attribution analysis requests, extracts the number of drill-down levels specified in the natural language attribution analysis request and uses it as the preset drill-down depth, extracts the contribution filtering strategy specified in the natural language attribution analysis request and uses the contribution threshold in the contribution filtering strategy as the preset contribution threshold.

[0095] Among the above optional methods, multi-level dimensional analysis in recursive drill-down mode is further supported, and layer-by-layer tracing can be automatically performed according to the drill-down depth and contribution filtering conditions specified in the natural language request, so that root cause localization can be extended from macro-level dimensions to micro-level nodes.

[0096] like Figure 2 As shown, the overall system architecture of this embodiment includes a data source layer, a baseline early warning triggering mechanism, an attribution analysis skill guidance document, an AI intelligent agent orchestration engine, an MCP toolset, a scene configuration registration mechanism, and a result output layer.

[0097] The data source layer provides base period insurance operating data and current period insurance operating data. Base period insurance operating data includes operating data from the same period last year, operating data from the same period last year, or operating data based on preset target values. Current period insurance operating data is the actual operating data for the current period. The system periodically collects operating data, forming two datasets: base period data and current period data.

[0098] The baseline early warning trigger mechanism automatically compares the base period data of operating indicators with the current period data. When the change exceeds a preset threshold, attribution analysis is automatically triggered. The data acquisition and early warning module 101 calculates the rate of change of indicators between the base period insurance operating data and the current period insurance operating data according to the preset baseline threshold rules. When the rate of change of indicators is greater than the preset threshold, an attribution trigger signal is generated. The preset baseline threshold rules include a rate of change threshold and a change amount threshold. For example, the rate of change threshold is set to -5%, and attribution analysis is triggered when the year-on-year decrease in premium income exceeds 5%; the change amount threshold is set to -500,000 yuan, and attribution analysis is triggered when the change amount of premium income is -800,000 yuan. The system automatically calculates the rate of change of indicators between the base period and the current period according to the preset baseline threshold rules, and automatically triggers the attribution analysis process when the rate of change exceeds the early warning threshold. Users can also send attribution analysis requests to the AI ​​agent through natural language.

[0099] The scenario configuration registration mechanism is used to construct factor attribution scenario configuration models and dimensional attribution scenario configuration models, templated by the indicator formulas, grouping dimension parameters, and analysis mode parameters. The factor attribution scenario configuration model includes indicator formulas and grouping dimension parameters, while the dimensional attribution scenario configuration model includes analysis mode parameters, which include independent analysis mode and recursive drill-down mode. Each factor and dimensional attribution scenario configuration model has a corresponding scenario identifier. The system loads preset analysis scenarios with a single click using these scenario identifiers, reducing the parameter configuration cost for each analysis.

[0100] The attribution analysis skills guide is a pre-written, structured guidance document that outlines the attribution analysis workflow. It encodes best practices for attribution analysis into structured guidance steps, used to guide the AI ​​agent orchestration module 103 in determining analysis methods, selecting parameters, and scheduling tools. The attribution analysis skills guide is named attribution_guide.md.

[0101] The AI ​​agent orchestration module 103 consists of a large language model and a Skill orchestration engine, enabling natural language understanding and tool scheduling. The AI ​​agent orchestration module 103 receives attribution trigger signals or user-input natural language attribution analysis requests, calls the attribution analysis skill guidance document to obtain attribution analysis process guidance, and loads the corresponding factor attribution scenario configuration model or dimensional attribution scenario configuration model based on the attribution analysis process guidance and scenario identifier, outputting tool call instructions. The AI ​​agent performs the following operations sequentially: determining the analysis method, including chain substitution, log-mean Dichroism index method, weighted average method, or dimensional attribution analysis; if a preset scenario configuration exists, loading the indicator formula and dimensional parameters from the scenario configuration registration module 102; and calling the corresponding MCP tool to input data and perform attribution calculations. For scenarios requiring multi-dimensional drill-down analysis, the AI ​​agent calls the dimensional attribution analysis tool and enables recursive drill-down mode, setting the dimensional node number filtering strategy to filter the top 5 dimensional nodes by contribution, ensuring the analysis focuses on the most influential dimensional nodes.

[0102] The MCP toolset encapsulates attribution algorithms into standardized Model Context Protocol tools, including a general attribution analysis tool, a factor attribution analysis tool, and a dimensional attribution analysis tool. The general attribution analysis tool provides a unified entry point, automatically routing to the corresponding specific attribution algorithm based on the characteristics of the input data. The factor attribution analysis tool supports direct dataset input for factor-level attribution contribution calculation. The dimensional attribution analysis tool supports independent analysis and recursive drill-down modes for dimensional attribution contribution calculation and multi-level root cause localization. Each tool embeds a corresponding attribution algorithm, including chain substitution, log-mean Dichroism index, weighted average, and dimensional attribution drill-down algorithms. Large language models invoke attribution capabilities through the standard protocol.

[0103] The attribution calculation module 104 receives the tool call instruction, calls the corresponding attribution algorithm to perform attribution calculation on the base period insurance operating data and the current period insurance operating data according to the factor attribution scenario configuration model or the dimension attribution scenario configuration model, and outputs the attribution calculation results.

[0104] The attribution calculation module 104 uses the chain substitution method as its core algorithm, supplemented by the logarithmic mean Dichotomy index method and the weighted average method, to achieve full coverage of absolute value indicators, ratio indicators, and composite indicators. The attribution calculation module 104 supports formula parsing for Chinese variable names and uses an expression engine to dynamically evaluate formulas. Upon receiving a tool call command, the attribution calculation module 104 first identifies the indicator type of the indicator formula: when the indicator formula only contains multiplication or addition operations, the indicator type is determined to be an absolute value indicator, and the chain substitution method is selected for attribution calculation; when the indicator formula is multiplicative and the user specifies the use of the logarithmic mean Dichotomy index method, the logarithmic mean Dichotomy index method is selected for attribution calculation; when the indicator formula is in ratio form and both the numerator and denominator are operating indicators, the indicator type is determined to be a ratio indicator, and the weighted average method is selected for attribution calculation, decomposing the indicator change into level effect components and structural effect components. In the indicator formula, the components that affect the value of the operating indicator are factor variables. The factor variables include three attributes: factor variable name, base period value, and current period value.

[0105] The factor attribution scenario configuration model also includes a list of factor variable names and factor order configuration information. The list of factor variable names is used to identify each factor variable in the indicator formula, and the factor order configuration information is used to determine the substitution order of each factor variable when performing factor contribution calculation.

[0106] The attribution calculation module 104 extracts each factor variable name from the indicator formula and maps each factor variable name to an alias recognizable by the expression engine. The expression engine parses and calculates the mathematical expression, identifies the variable names in alias form, and substitutes the variable values ​​into the expression for evaluation. For example, the expression engine maps the factor variable name "number of policies" to the alias "a" and "average premium per policy" to the alias "b," and substitutes the values ​​corresponding to the aliases into the expression "a×b" to calculate premium income. The attribution calculation module 104, according to the factor order configuration information, substitutes the base period value or current period value of each factor variable into the indicator formula to calculate the indicator value; it sequentially replaces the value of each factor variable from the base period value to the current period value according to the factor order configuration information, calculates the indicator value after each replacement, determines the contribution of each factor variable based on the indicator value calculated after each replacement, and outputs the attribution calculation result.

[0107] In the core calculation process of the chain substitution method, factor variable names are automatically extracted from the user-specified indicator formula. Examples of indicator formulas include "premium income × loss ratio", "income - cost", or "(settled amount + outstanding amount) / (premium income - unearned premium reserve)". The formula extractor maps the factor variable names to aliases recognizable by the expression engine. The base period factor values ​​are then substituted into the indicator formula to calculate the base period indicator value. The information is configured according to factor order, and the value of each factor variable is replaced from the base period value to the current period value in turn. After each replacement, the indicator value is recalculated. The contribution of the factor variable is determined according to the formula. Calculation, where Indicates the first The contribution of each factor variable Indicates the previous The values ​​of one factor variable are replaced with the current period values, while the remaining factor variables retain their base period values. These are then used in the index formula to calculate the index value. Indicates the previous The values ​​of one factor variable are replaced with their current period values, while the remaining factor variables retain their base period values. These are then entered into the indicator formula to calculate the indicator value. The factor variables are arranged in descending order of their absolute contribution value, with positive and negative contributions displayed separately. Positive contributions indicate those that drive indicator growth, while negative contributions indicate those that drag down indicator growth. The contribution percentage of each factor variable is calculated using the formula... Calculate. The residual is calculated using the formula. Calculations are performed to verify the accuracy of the algorithm.

[0108] When the user specifies the use of the log-mean Dijkstra's index method, the attribution calculation module 104 uses the log-mean Dijkstra's index method for attribution decomposition. The contribution of each factor variable is determined by the formula. Calculation, where This represents the index of the log-mean Dijkstra method, calculated as the first... The contribution of each factor variable This represents the current period index value calculated by substituting the current period values ​​of each factor variable into the index formula. This represents the base period index value calculated by substituting the base period values ​​of each factor variable into the index formula. Indicates the first The current period value of each factor variable. Indicates the first The base period values ​​of each factor variable. The log-mean Dijkstra index method has the advantages of no residuals and results that do not depend on the order of factors, making it suitable for the precise decomposition of multiplicative indices.

[0109] When the indicator type in the indicator formula is a ratio-type indicator, the attribution calculation module 104 decomposes the change in the indicator between the base period insurance operating data and the current period insurance operating data into a horizontal effect component and a structural effect component using the weighted average method. The horizontal effect component reflects the contribution of each dimension's own indicator value change, while the structural effect component reflects the impact of changes in the weight ratio of each dimension, achieving a complete decomposition of the ratio indicator without residuals.

[0110] The dimensional attribution drill-down algorithm performs attribution calculations for multi-dimensional scenarios. The attribution calculation module 104 first groups and aggregates the base period insurance operating data and the current period insurance operating data according to the coarsest granularity dimension, and then performs a chain substitution method to calculate factor contribution within each group. When the loaded configuration model is the dimensional attribution scenario configuration model and the analysis mode parameter is recursive drill-down mode, the attribution calculation module 104 performs attribution calculations on the next level dimension for the dimension node with the highest contribution in the attribution calculation results. The recursive drill-down mode includes first-level drill-down and at least one next-level drill-down. After completing the attribution calculation of the first level dimension, the attribution calculation module 104 obtains the contribution of each dimension node, determines the dimension node with the largest contribution value as the dimension node to be drilled down, and performs attribution calculations on the dimension node to be drilled down according to the next level dimension. For example, the next level dimension drills down from the organization dimension to the channel dimension and then down to the personnel dimension. The attribution calculation module 104 recursively executes attribution calculations for the next level dimension until a preset drill-down depth is reached or the contribution of each dimension node in the current level is less than a preset contribution threshold. Throughout the process, the module supports filtering by the number of dimension nodes and percentage thresholds to ensure focused analysis. The AI ​​agent orchestration module 103 identifies the number of drill-down levels and the contribution filtering strategy specified in the natural language attribution analysis request within the attribution analysis process guide, using the number of drill-down levels as the preset drill-down depth and the contribution threshold in the contribution filtering strategy as the preset contribution threshold.

[0111] In scenarios where only the composition of the current overall indicator value is needed without base period comparison, the attribution calculation module 104 provides an impact calculation tool. Impact is calculated based on the formula... Calculation, where Indicates the first The weighting percentage of each dimension Indicates the first The index values ​​of each dimension, This represents the aggregated indicator value. This tool only requires current period data and is suitable for analyzing the influence of each dimension on aggregate ratio indicators. The calculation result satisfies the condition that the sum of the influence of each dimension is zero. .

[0112] The results output module 105 generates a contribution matrix and an attribution analysis report based on the attribution calculation results, and inputs the attribution calculation results into the large language model to generate management recommendations. The contribution matrix is ​​output in Markdown format, displaying the contribution values ​​and percentages of each factor variable or dimension node grouped by positive and negative contributions, including the base period value, current period value, change, contribution degree, and contribution percentage for each factor. The system automatically generates attribution analysis data files and uploads them to object storage, providing a download link for business personnel to use spreadsheet tools for downstream analysis. Management recommendations are automatically generated by the large language model based on the attribution analysis results and business knowledge. For example, if the main contributing factor to the increase in the loss ratio is the increase in the loss amount of a certain product in a certain channel, the system automatically generates management recommendations: focus on the risk management of this product in that channel, and suggest adjusting the underwriting policy or strengthening claims review. Types of management recommendations include product structure adjustment recommendations, channel optimization recommendations, personnel management recommendations, and customer segmentation recommendations. Product structure adjustment suggestions are generated based on the profitability analysis of each insurance type; channel optimization suggestions are generated based on the channel cost ratio analysis; personnel management suggestions are generated based on the personnel productivity and loss ratio analysis; and customer segmentation suggestions are generated based on the cross-analysis of customer renewal rate and loss ratio.

[0113] The core technical feature of this embodiment lies in upgrading fixed algorithmic attribution analysis to an intelligent analysis paradigm that combines LLM and MCP toolsets with Skill orchestration. Unlike traditional attribution analysis software, this embodiment does not embed algorithms in code for users to call through fixed parameters. Instead, it encodes domain knowledge and best practices of attribution analysis into LLM-understandable and executable process guidelines through Skill guidance files. Users flexibly describe their analysis needs using natural language, and the AI ​​agent autonomously selects appropriate methods, parameters, and tools to complete the analysis based on the Skill guidance. This AI-driven flexible attribution analysis model has extremely high flexibility and adaptability, and can cope with diverse needs of different business scenarios, different indicator types, and different levels of analysis depth.

[0114] Another embodiment of the baseline early warning and attribution suggestion generation system for insurance operation data of the present invention includes: The data acquisition and early warning module 101 acquires base period insurance operating data and current period insurance operating data, calculates the rate of change of indicators between base period insurance operating data and current period insurance operating data according to preset baseline threshold rules, and generates an attribution trigger signal when the rate of change of indicators is greater than the preset threshold.

[0115] The AI ​​intelligent agent orchestration module 103 responds to attribution trigger signals or natural language attribution analysis requests input by the user, calls the attribution analysis skill guidance document to obtain attribution analysis process guidance, and loads the corresponding factor attribution scenario configuration model or dimensional attribution scenario configuration model according to the attribution analysis process guidance and scenario identifier. The factor attribution scenario configuration model includes indicator formulas and grouping dimension parameters, and the dimensional attribution scenario configuration model includes analysis mode parameters, including independent analysis mode and recursive drill-down mode, and outputs tool call instructions.

[0116] The attribution calculation module 104 determines the corresponding attribution algorithm based on the tool call instruction and the factor attribution scenario configuration model or the dimension attribution scenario configuration model. It inputs the base period insurance operating data and the current period insurance operating data into the attribution algorithm to perform attribution calculation and outputs the attribution calculation results.

[0117] When the loaded configuration model is a dimension attribution scenario configuration model and the analysis mode parameter is recursive drill-down mode, the attribution calculation module 104 extracts the dimension node with the highest contribution from the attribution calculation results as the dimension node to be drilled down. Attribution calculation is performed on the dimension node to be drilled down according to the next level dimension. After each drill-down is completed, the number of attribution calculation levels that have been executed is updated. After each drill-down is completed, the root cause lock degree is calculated based on the contribution data of the current level. The calculated root cause lock degree is compared with the preset lock threshold. Based on the comparison result, it is determined whether to continue to execute the next level of drill-down or stop drill-down.

[0118] After each attribution calculation for the next level dimension is completed, the contribution data of all dimension nodes in the current level is obtained. The maximum and second largest values ​​are selected from the contribution data, and the root cause lock-in degree of the current level is calculated based on the difference between the maximum and second largest values. The root cause lock-in degree is calculated according to the following formula: in, Indicates the first Root cause lock-in at each level Indicates the first The contribution value of the dimension node with the highest contribution in each level. Indicates the first The contribution value of the second largest dimension node in each level. are positive numbers and It is less than the absolute value of the contribution of any single dimension node.

[0119] It should be noted that during the recursive drill-down process, the attribution calculation module 104 obtains the contribution set of all dimension nodes at each level after performing attribution calculations. By selecting the dimension node with the highest contribution and the second highest contribution from this set, the difference between the two is used as the core quantitative criterion for determining whether root causes are concentrated. The sum of the two, plus a very small positive number, is used as the normalization denominator, ensuring that the calculated root cause lock-in degree falls within the range of zero to one. This eliminates the impact of differences in the absolute magnitude of contribution values ​​between different levels on the consistency of the judgment criteria. The function of the above formula is that when the root cause lock-in degree reaches or exceeds a preset lock-in threshold, it indicates that the contribution distribution at that level has a significant unipolar characteristic, meaning there is a clear difference between the node with the highest contribution and the other nodes. At this point, the node with the highest contribution can be identified as the endpoint of root cause localization without further drill-down, effectively avoiding ineffective over-drilling and improving the overall execution efficiency of attribution analysis while ensuring the accuracy of root cause localization.

[0120] The attribution calculation module 104 compares the root cause lock degree with the preset lock threshold. When the root cause lock degree is greater than or equal to the preset lock threshold, the dimension node corresponding to the current level is determined as the target root cause node, and drilling continues. When the root cause lock degree is less than the preset lock threshold, the attribution calculation of the next level dimension continues until the preset drilling depth is reached or the absolute value of the contribution of all dimension nodes in the current level is less than the preset contribution threshold.

[0121] After stopping drilling, the attribution calculation module 104 outputs the attribution calculation results of all acquired levels as the final attribution calculation result.

[0122] The results output module 105 generates a contribution matrix and an attribution analysis report based on the final attribution calculation results, and inputs the final attribution calculation results into the large language model, which then generates management recommendations.

[0123] The technical solution of this embodiment obtains the contribution data of all dimension nodes at each drilling level after completion, selects the maximum and second largest contribution values ​​to calculate the root cause lock-in degree, compares the root cause lock-in degree with a preset lock-in threshold to determine the degree of unipolarity of the contribution distribution at the current level, and stops drilling when the root cause lock-in degree reaches the preset lock-in threshold. This solves the problem of excessive drilling caused by the lack of scientific stopping conditions in the recursive drilling process, reduces unnecessary data processing overhead, improves the execution efficiency of the attribution analysis process, and changes the determination of drilling stop conditions from a fixed number of layers to an adaptive determination based on data distribution characteristics, thereby improving the accuracy of root cause localization.

[0124] Figure 3This diagram illustrates a flowchart of an embodiment of a baseline early warning and attribution suggestion generation method for insurance business data provided by the present invention. This method employs the baseline early warning and attribution suggestion generation system 100 for insurance business data provided by the present invention. Figure 3 As shown, the method includes the following steps: S1. Calculate the rate of change of indicators between the base period insurance operating data and the current period insurance operating data according to the preset baseline threshold rule, and generate an attribution trigger signal when the rate of change of indicators is greater than the preset threshold. S2. Construct a factor attribution scenario configuration model and a dimension attribution scenario configuration model. The factor attribution scenario configuration model includes indicator formulas and grouping dimension parameters. The dimension attribution scenario configuration model includes analysis mode parameters. The analysis mode parameters include independent analysis mode and recursive drill-down mode. The factor attribution scenario configuration model and the dimension attribution scenario configuration model are respectively associated with scenario identifiers. S3. Based on the attribution trigger signal or the natural language attribution analysis request input by the user, call the attribution analysis skill guidance file to obtain the attribution analysis process guidance, and load the corresponding factor attribution scenario configuration model or the dimension attribution scenario configuration model according to the attribution analysis process guidance and the scenario identifier, and output the tool call command. S4. Receive the tool call instruction, call the corresponding attribution algorithm according to the factor attribution scenario configuration model or the dimension attribution scenario configuration model to perform attribution calculation on the base period insurance operating data and the current period insurance operating data, and output the attribution calculation result; when the loaded configuration model is the dimension attribution scenario configuration model and the analysis mode parameter is the recursive drill-down mode, perform the next level dimension attribution calculation on the dimension node with the highest contribution in the attribution calculation result; S5. Based on the attribution calculation results, generate a contribution matrix and an attribution analysis report, and input the attribution calculation results into the large language model to generate management suggestions.

[0125] It should be noted that the beneficial effects of the baseline early warning and attribution suggestion generation method for insurance operating data provided in the above embodiments are the same as those of the baseline early warning and attribution suggestion generation system 100 for insurance operating data, and will not be repeated here. Furthermore, the method and system embodiments provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the system embodiments, and will not be repeated here.

[0126] The baseline warning and attribution suggestion generation system 100 for insurance business data of the present invention can be a computer program (including program code) running on a computer device. For example, the baseline warning and attribution suggestion generation system 100 for insurance business data of the present invention is an application software that can be used to execute the corresponding steps in the baseline warning and attribution suggestion generation method for insurance business data of the present invention.

[0127] In some embodiments, the baseline warning and attribution suggestion generation system 100 for insurance business data of the present invention can be implemented in a combination of hardware and software. As an example, the baseline warning and attribution suggestion generation system 100 for insurance business data of the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the baseline warning and attribution suggestion generation method for insurance business data of the present invention. For example, the processor in the form of a hardware decoding processor can be one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0128] The modules described in the embodiments of this invention can be implemented in software or hardware. The names of the modules are not, in some cases, limiting the scope of the module itself.

[0129] An electronic device according to an embodiment of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the baseline warning and attribution suggestion generation method for any of the above-mentioned insurance business data. That is, an electronic device according to an embodiment of the present invention may include, but is not limited to: a processor and a memory; the memory is used to store the computer program; the processor is used to execute the baseline warning and attribution suggestion generation method for insurance business data shown in any embodiment of the present invention by calling the computer program.

[0130] In one alternative embodiment, an electronic device is provided, such as Figure 4 As shown, Figure 4The illustrated electronic device 4000 includes a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, for example, via a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 4004 is not limited to one type, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of the present invention.

[0131] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 4001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0132] Bus 4002 may include a path for transmitting information between the aforementioned components. Bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 4002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 4 The bus 4002 is represented by only one thick line, but this does not mean that there is only one bus or one type of bus.

[0133] The memory 4003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0134] The memory 4003 stores application code (computer program) for executing the present invention, and its execution is controlled by the processor 4001. The processor 4001 executes the application code stored in the memory 4003 to implement the content shown in the foregoing method embodiments.

[0135] Among them, electronic devices can also be terminal devices. A terminal device can be any terminal device that can install applications and access web pages through applications, including at least one of smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, smart TVs, and smart in-vehicle devices.

[0136] It should be noted that, Figure 4 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0137] An embodiment of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a baseline warning and attribution suggestion generation method for any of the aforementioned insurance business data.

[0138] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.

[0139] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the aforementioned baseline warning and attribution suggestion generation method for insurance business data.

[0140] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0141] It should be understood that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0142] The computer-readable storage medium provided in this invention can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0143] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the method shown in the above embodiments.

[0144] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0145] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and represent a limitation on a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0146] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this invention can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, this invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.

[0147] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A baseline early warning and attribution suggestion generation system for insurance operation data, characterized in that, include: The data acquisition and early warning module is used to calculate the rate of change of indicators between the base period insurance operating data and the current period insurance operating data according to the preset baseline threshold rules, and generate an attribution trigger signal when the rate of change of indicators is greater than the preset threshold. The scenario configuration registration module is used to construct factor attribution scenario configuration models and dimensional attribution scenario configuration models. The factor attribution scenario configuration model includes indicator formulas and grouping dimension parameters, and the dimensional attribution scenario configuration model includes analysis mode parameters, which include independent analysis mode and recursive drill-down mode. The factor attribution scenario configuration model and the dimensional attribution scenario configuration model are respectively associated with scenario identifiers. The AI ​​intelligent agent orchestration module is used to call the attribution analysis skill guidance file to obtain attribution analysis process guidance according to the attribution trigger signal or the natural language attribution analysis request input by the user, load the corresponding factor attribution scenario configuration model or the dimension attribution scenario configuration model according to the attribution analysis process guidance and the scenario identifier, and output tool call instructions. The attribution calculation module is used to receive the tool call instruction, call the corresponding attribution algorithm according to the factor attribution scenario configuration model or the dimension attribution scenario configuration model to perform attribution calculation on the base period insurance operating data and the current period insurance operating data, and output the attribution calculation result; when the loaded configuration model is the dimension attribution scenario configuration model and the analysis mode parameter is the recursive drill-down mode, the attribution calculation of the next level dimension is performed on the dimension node with the highest contribution in the attribution calculation result; The results output module is used to generate a contribution matrix and an attribution analysis report based on the attribution calculation results, and input the attribution calculation results into a large language model to generate management suggestions.

2. The baseline early warning and attribution suggestion generation system for insurance operation data according to claim 1, characterized in that, The base period insurance operating data includes at least one of the following: operating data from the same period last year, operating data from the same period last year, or operating data with a preset target value; the preset baseline threshold rule includes a rate of change threshold and a change amount threshold; the data collection and early warning module is specifically used for: The attribution trigger signal is generated when the rate of change of the indicator is greater than the rate of change threshold or the amount of change of the indicator is greater than the amount of change threshold.

3. The baseline early warning and attribution suggestion generation system for insurance operation data according to claim 1, characterized in that, The factor attribution scenario configuration model also includes a list of factor variable names and factor order configuration information. The list of factor variable names is used to identify each factor variable in the indicator formula, and the factor order configuration information is used to determine the substitution order of each factor variable when performing factor contribution calculation.

4. The baseline early warning and attribution suggestion generation system for insurance operation data according to claim 3, characterized in that, The attribution calculation module is specifically used for: Extract each factor variable name from the index formula and map each factor variable name to an alias that the expression engine can recognize; Based on the factor order configuration information, the base period value or current period value of each factor variable is substituted into the indicator formula to calculate the indicator value; According to the factor order configuration information, the value of each factor variable is replaced from the base period value to the current period value in turn. The index value after each replacement is calculated. The contribution of each factor variable is determined based on the index value calculated after each replacement, and the attribution calculation result is output.

5. The baseline early warning and attribution suggestion generation system for insurance operation data according to claim 4, characterized in that, The attribution calculation module is specifically used for: The contribution of the i-th factor variable is calculated using the chain substitution method, and the formula for the contribution is: in, This represents the contribution of the i-th factor variable. This indicates the index value calculated by replacing the values ​​of the first i factor variables with their current period values ​​and keeping the remaining factor variables at their base period values, and then inputting these values ​​into the index formula. This means that the index value is calculated by replacing the values ​​of the first i-1 factor variables with the values ​​of the current period, while keeping the base period values ​​of the remaining factor variables, and then inputting the index into the index formula. The contribution of each factor variable is sorted from largest to smallest according to its absolute value, and the contribution percentage of each factor variable is calculated. The contribution percentage is the ratio of the absolute value of the contribution of the factor variable to the sum of the absolute values ​​of the contributions of all factor variables.

6. The baseline early warning and attribution suggestion generation system for insurance operation data according to claim 4, characterized in that, The attribution calculation module is also specifically used for: When the log-mean Dichotomy index method is chosen, the contribution of the i-th factor variable is calculated according to the following formula: in, This represents the contribution of the i-th factor variable calculated using the log-mean Dijkstra index method. This indicates the current period index value calculated by substituting the current period values ​​of each factor variable into the index formula. This represents the base period index value calculated by substituting the base period values ​​of each factor variable into the index formula. This represents the current period value of the i-th factor variable. This represents the base period value of the i-th factor variable; When the indicator type in the indicator formula is a ratio-type indicator, the change in the indicator between the base period insurance operating data and the current period insurance operating data is decomposed into a horizontal effect component and a structural effect component according to the weighted average method.

7. The baseline early warning and attribution suggestion generation system for insurance operation data according to claim 1, characterized in that, The recursive drilling mode includes a first-level drilling and at least one next-level drilling; The attribution calculation module is specifically used to: in the recursive drill-down mode, after completing the attribution calculation of the first level dimension, obtain the contribution of each dimension node, determine the dimension node with the largest contribution value as the dimension node to be drilled down, and perform attribution calculation on the dimension node to be drilled down according to the next level dimension until the preset drill-down depth is reached or the contribution of each dimension node in the current level is less than the preset contribution threshold. The AI ​​intelligent agent orchestration module is specifically used to: identify the number of drilling levels and contribution filtering strategy specified in the natural language attribution analysis request in the attribution analysis process guidance, use the number of drilling levels as the preset drilling depth, and use the contribution threshold in the contribution filtering strategy as the preset contribution threshold.

8. A method for generating baseline early warning and attribution suggestions for insurance operation data, characterized in that, The method employs the baseline early warning and attribution suggestion generation system for insurance operating data as described in any one of claims 1 to 7, the method comprising: The rate of change of indicators between the base period insurance operating data and the current period insurance operating data is calculated according to the preset baseline threshold rule, and an attribution trigger signal is generated when the rate of change of indicators is greater than the preset threshold. Construct a factor attribution scenario configuration model and a dimension attribution scenario configuration model. The factor attribution scenario configuration model includes indicator formulas and grouping dimension parameters. The dimension attribution scenario configuration model includes analysis mode parameters, including independent analysis mode and recursive drill-down mode. The factor attribution scenario configuration model and the dimension attribution scenario configuration model are respectively associated with scenario identifiers. Based on the attribution trigger signal or the natural language attribution analysis request input by the user, the attribution analysis skill guidance file is invoked to obtain the attribution analysis process guidance. Based on the attribution analysis process guidance and the scenario identifier, the corresponding factor attribution scenario configuration model or the dimension attribution scenario configuration model is loaded, and the tool invocation command is output. Upon receiving the tool invocation instruction, the corresponding attribution algorithm is invoked according to the factor attribution scenario configuration model or the dimension attribution scenario configuration model to perform attribution calculations on the base period insurance operating data and the current period insurance operating data, and the attribution calculation results are output; when the loaded configuration model is the dimension attribution scenario configuration model and the analysis mode parameter is the recursive drill-down mode, the next level dimension attribution calculation is performed on the dimension node with the highest contribution in the attribution calculation results; Based on the attribution calculation results, a contribution matrix and an attribution analysis report are generated, and the attribution calculation results are input into a large language model to generate management recommendations.

9. An electronic device, characterized in that, The electronic device includes a processor coupled to a memory, the memory storing at least one computer program, which is loaded and executed by the processor to enable the electronic device to implement the baseline early warning and attribution suggestion generation method for insurance business data as described in claim 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which, when executed by a processor, implements the baseline early warning and attribution suggestion generation method for insurance business data as described in claim 8.

Images