A Customer Resource Data Analysis Method and System in the Insurance Industry

Abstract

This invention relates to the field of insurance data analysis technology, and discloses a method and system for analyzing customer resource data in the insurance industry. The method includes: acquiring channel touchpoint sequence data and competitor reach signal data, calculating the competitor pressure value within a given interval; calculating the cumulative competitor reach pressure value at each stage of the conversion path; constructing a counterfactual touchpoint sequence with missing touchpoints; inferring the probability change of downstream touchpoints based on a channel causal relationship graph, and predicting the counterfactual conversion probability; calculating the pure causal incremental contribution value of each touchpoint; and aggregating the pure causal incremental contribution value in layers according to channel identifier and competitor pressure level to generate channel value analysis results. This invention can separate the true incremental contribution of channels from the passive absorption effect, while controlling the independent influence of competitor pressure factors, thereby improving the accuracy of customer resource acquisition and deployment decisions.

Description

Technical Field

[0001] This invention relates to the field of insurance data analysis technology, and more specifically, to a method and system for analyzing customer resource data in the insurance industry. Background Technology

[0002] In customer acquisition scenarios within the insurance industry, insurance data platforms need to evaluate the contribution value of each acquisition channel to customer conversion in order to guide channel deployment decisions for customer resource acquisition. Existing technologies analyze the sequence of channel touchpoints a customer encounters before making a purchase, calculating the contribution weight of each channel based on characteristics such as the touchpoint's location in the path, interaction depth, and temporal proximity. The contribution values ​​are then aggregated by channel to form the basis for decision-making.

[0003] However, existing channel contribution assessment methods have the following technical problems: First, they treat all touchpoints on the conversion path as independent contributors, failing to consider the causal transmission relationship between channels. For example, downstream touchpoints generated by customers actively searching after brand advertising exposure cannot distinguish between the channel's real incremental effect and passive reception effect. Second, in the competitive environment of the insurance market, customers are simultaneously reached by multiple insurance companies. Existing methods do not consider the independent impact of competitor outreach on conversion probability, leading to an overestimation of the company's channel contribution under high-pressure competitor conditions and an underestimation of the company's channel purity under low-pressure competitor conditions, causing customer resource acquisition and placement decisions to deviate from actual results. Summary of the Invention

[0004] This invention provides a customer resource data analysis method and system for the insurance industry, solving the technical problems in related technologies such as the inability to accurately quantify the true contribution value of channels in a competitive environment, the difficulty in identifying the impact of causal transmission relationships between channels on conversion effects, and the lack of assessment of the impact of competitive pressure on channel effectiveness.

[0005] This invention provides a method for analyzing customer resource data in the insurance industry, comprising the following steps: Acquire channel touchpoint sequence data of customers who have made a purchase and competitor reach signal data during the same period. Align the competitor reach signals to each time interval of the channel touchpoint sequence according to time and calculate the interval competitor pressure value at the time of occurrence of each channel touchpoint. Based on the interval competitor pressure value sequence, the cumulative competitor reach pressure value of the customer at each stage of the conversion path is calculated by weighted summation according to the proximity of the reach time. For each customer's channel touchpoint sequence, iterate through each channel touchpoint in the sequence and remove the target channel touchpoint from the original sequence to form a counterfactual touchpoint sequence. Based on the channel causal relationship map, the probability change of the existence of downstream channel touchpoints in the counterfactual touchpoint sequence is inferred. The counterfactual touchpoint sequence and the cumulative competitor reach pressure value of the corresponding time period are jointly input into the conversion probability prediction model with competitor covariates, and the counterfactual conversion probability estimate is output. The counterfactual conversion probability estimate is compared with the actual conversion result, and the conversion probability difference is calculated as the pure causal incremental contribution value of the channel touchpoint. The pure causal incremental contribution values ​​are aggregated in layers according to channel identifier and competitor pressure level, generating channel value analysis results that include an incremental contribution distribution matrix and a ranking list of channel competitor sensitivity.

[0006] Furthermore, the competitor outreach signal data includes third-party advertising exposure records, competitor insurance app activity records, and insurance comparison platform access records. The interval competitor pressure value is calculated as follows: all competitor outreach signals within the time window of the channel touchpoint occurrence time are statistically analyzed, different intensity weights are assigned to different types of competitor outreach signals, and the intensity weights of each competitor outreach signal are summed to obtain the interval competitor pressure value. Among them, the intensity weight of third-party advertising exposure records is lower than the intensity weight of competitor insurance app activity records, and the intensity weight of competitor insurance app activity records is lower than the intensity weight of insurance comparison platform access records.

[0007] Furthermore, the cumulative competitor reach pressure value is obtained by exponentially decaying weighted summation of the competitor pressure value sequence in the interval. The decay weight is a negative exponential power function of the natural constant, and the exponent term of the negative exponential power function is the negative value of the product of the time decay coefficient and the time difference between the current channel touch point occurrence time and the competitor reach signal occurrence time.

[0008] Furthermore, the time decay coefficient is set differently according to the type of insurance product. A high time decay coefficient is set for short-term insurance products with short decision-making cycles, and a low time decay coefficient is set for long-term life insurance products with long decision-making cycles.

[0009] Furthermore, when constructing the counterfactual contact sequence, the time interval between adjacent channel contacts before and after the removed channel contact is adjusted, and the subsequent channel contacts of the removed channel contact are shifted forward on the time axis by the shift amount being the time interval between the removed channel contact and its preceding channel contact.

[0010] Furthermore, the channel causal relationship graph is a directed graph structure describing the causal transmission relationship between different channel touchpoints. The nodes of the channel causal relationship graph represent channel types, the directed edges of the channel causal relationship graph represent the causal influence relationship between channels, and the weight of the directed edges of the channel causal relationship graph represents the conditional probability of the upstream channel triggering the occurrence of the downstream channel. The conditional probability is obtained by statistically analyzing the ratio of the number of customers whose downstream channel touchpoints appear in sequence after the occurrence of the upstream channel touchpoint to the total number of customers including the upstream channel touchpoint. Causal edges are established to connect channel pairs whose conditional probabilities exceed a preset threshold.

[0011] Furthermore, the change in the probability of existence of downstream channel touchpoints in the counterfactual touchpoint sequence includes: after the target channel touchpoint is removed, searching for all outgoing edges with the channel corresponding to the target channel touchpoint as the source node in the channel causal relationship graph, obtaining the downstream channels pointed to by each outgoing edge and the corresponding conditional probability weights, marking the channel touchpoints belonging to downstream channels in the counterfactual touchpoint sequence as probabilistic existence states, the existence probability of the channel touchpoints in probabilistic existence states is equal to the original existence probability multiplied by a decay factor, the decay factor being 1 minus the conditional probability weight of the corresponding causal edge; in the subsequent conversion probability prediction, the channel touchpoints in probabilistic existence states participate in feature calculation by weighting according to their existence probability.

[0012] Furthermore, the conversion probability prediction model with competitor covariates takes the touchpoint sequence feature vector and the cumulative competitor reach pressure value as input, and the conversion probability as output; the touchpoint sequence feature vector includes the number of channel touchpoints, the distribution of channel touchpoint types, the statistics of channel touchpoint time intervals, and the channel touchpoint interaction depth index; before inputting into the conversion probability prediction model with competitor covariates, the numerical features are standardized, the categorical features are converted using one-hot encoding, and the cumulative competitor reach pressure value is normalized.

[0013] Furthermore, the pure causal incremental contribution value is equal to the actual conversion identifier value minus the counterfactual conversion probability estimate; the pure causal incremental contribution value of each channel touchpoint is normalized, and the normalized pure causal incremental contribution value is equal to the pure causal incremental contribution value of that channel touchpoint divided by the sum of the pure causal incremental contribution values ​​of all channel touchpoints in the customer channel touchpoint sequence; the channel competitor sensitivity is obtained by calculating the coefficient of variation of the mean pure causal incremental contribution value of the same channel under different competitor pressure levels, and the coefficient of variation is equal to the standard deviation divided by the arithmetic mean.

[0014] This invention also proposes a customer resource data analysis system for the insurance industry, comprising: The data acquisition and pressure value calculation module is used to acquire channel touchpoint sequence data of customers who have made transactions and competitor touchpoint signal data during the same period. It aligns the competitor touchpoint signals to each time interval of the channel touchpoint sequence according to time and calculates the interval competitor pressure value at the time of each channel touchpoint. The cumulative pressure value calculation module is used to calculate the cumulative competitive product reach pressure value of a customer at each stage of the conversion path by weighted summation based on the proximity of the reach time, according to the interval competitive product pressure value sequence. The counterfact sequence construction module is used to iterate through each channel touchpoint in the channel touchpoint sequence for each customer who has made a purchase, remove the target channel touchpoint from the original channel touchpoint sequence, and form a counterfact sequence. The counterfact conversion probability prediction module is used to infer the change in the probability of the existence of downstream channel touchpoints in the counterfact touchpoint sequence based on the channel causal relationship map. It inputs the counterfact touchpoint sequence and the cumulative competitor reach pressure value of the corresponding time period into the conversion probability prediction model with competitor covariates, and outputs the counterfact conversion probability estimate. The contribution value calculation module is used to compare the counterfactual conversion probability estimate with the actual conversion result and calculate the conversion probability difference as the pure causal incremental contribution value of the channel touchpoint. The analysis results generation module is used to aggregate pure causal incremental contribution values ​​in layers according to channel identifier and competitor pressure level, and generate channel value analysis results including an incremental contribution distribution matrix and a channel competitor sensitivity ranking list.

[0015] The beneficial effects of this invention are as follows: This invention constructs a counterfactual touchpoint sequence with missing touchpoints and infers the probability changes of downstream touchpoints based on a channel causal relationship graph. This allows for the consideration of the cascading effects of a particular channel touchpoint on downstream channel touchpoints when evaluating its contribution, thus separating the channel's true incremental contribution from the passive reception effect. This solves the technical problem of existing methods that treat channel touchpoints as independent contributors while ignoring the causal transmission relationship between channels. Simultaneously, by calculating the cumulative competitor reach pressure value and incorporating it as a covariate into the conversion probability prediction model, the invention allows for control over competitor pressure levels during the counterfactual conversion probability estimation process. This ensures that the conversion probability difference only reflects the pure incremental effect of the company's own channel touchpoints, solving the technical problem of channel contribution evaluation bias caused by the failure to consider competitor factors in existing methods. This achieves the technical effect of improving the consistency between customer resource acquisition and deployment decisions and actual results. Attached Figure Description

[0016] Figure 1 This is a flowchart of a customer resource data analysis method in the insurance industry proposed in this invention; Figure 2 This is a distribution diagram of the competitive pressure values ​​of each channel contact point in the example proposed in this invention. Figure 3 This is a graph showing the trend of cumulative competitor reach pressure value as a function of contact sequence in the example proposed in this invention; Figure 4 This is a comparison chart of the normalized contribution values ​​of each channel touchpoint in the example proposed in this invention; Figure 5 This is a schematic diagram illustrating the average contribution value of the channel under different levels of competitive pressure, as presented in this invention. Figure 6 This is a channel competitor sensitivity ranking chart of an example proposed in this invention; Figure 7 This is a scatter plot showing the relationship between the average channel contribution value and competitor sensitivity in the examples proposed in this invention. Figure 8 This is an intensity weight distribution diagram of the competitor reach signal type in the example proposed in this invention; Figure 9 This is a feature comparison diagram of the original sequence and the counterfactual sequence of the example proposed in this invention. Detailed Implementation

[0017] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0018] At least one embodiment of the present invention discloses a method for analyzing customer resource data in the insurance industry, such as... Figure 1 As shown, it includes the following steps: Step 1: Obtain channel touchpoint sequence data and competitor reach signal data to calculate the competitive pressure value for the interval: Obtain complete channel touchpoint sequence data for customers who have already made a purchase, as well as competitor reach signal data for the same period. Channel touchpoint sequence data includes each channel touchpoint the customer interacted with sequentially before making a purchase, along with their timestamps. Competitor reach signal data includes third-party advertising exposure records, competitor insurance app activity records, and insurance comparison platform access records.

[0019] Furthermore, third-party ad exposure records are obtained through data interfaces with third-party ad monitoring platforms. These interfaces return the timestamps and ad identifiers of competitor ads that customers encountered within a specific time period. Competitor insurance app activity records are obtained through data interfaces with mobile device behavior data service providers. These interfaces return the launch time, dwell time, and interaction records of customers within competitor insurance apps. Insurance comparison platform access records are obtained through data cooperation channels with insurance comparison platforms. These interfaces return the timestamps of customer visits to insurance comparison platforms, the types of insurance products viewed, and the dwell time.

[0020] Competitor outreach signals are aligned chronologically to the time intervals of the channel touchpoint sequence, and the interval competitor pressure value is calculated for each channel touchpoint occurrence time. The interval competitor pressure value is obtained by weighting the number and intensity of competitor outreach signals within the time window of the channel touchpoint occurrence time, and is used to quantify the degree of competitor interference experienced by the customer when a specific channel touchpoint occurs. The time window is set as a fixed duration range before and after the channel touchpoint occurrence time, for example, 24 hours before the channel touchpoint occurrence time to the moment of channel touchpoint occurrence.

[0021] Specifically, the competitive pressure value for a given time window is calculated as follows: all competitor outreach signals are statistically analyzed within the time window, and different intensity weights are assigned to different types of competitor outreach signals. The intensity weight for third-party advertising exposure records is set to 0.3, the intensity weight for competitor insurance APP activity records is set to 0.5, and the intensity weight for insurance comparison platform access records is set to 0.7. The intensity weights of each competitor outreach signal are summed to obtain the competitive pressure value for that time window.

[0022] Furthermore, the intensity weights are determined based on the degree of influence of different competitor outreach signals on customer decisions. Third-party ad exposure records represent passive information reception with low customer engagement, thus receiving a low intensity weight of 0.3; competitor insurance app activity records indicate that customers are actively seeking information about competitor products, with moderate customer engagement, thus receiving a moderate intensity weight of 0.5; insurance comparison platform access records indicate that customers are making multi-faceted comparisons, with the highest customer engagement, thus receiving a high intensity weight of 0.7.

[0023] Step 2, calculate the cumulative competitor reach pressure value at each stage of the conversion path: Based on the interval competitor pressure value sequence, the cumulative competitor reach pressure value of the customer at each stage of the conversion path is calculated by weighted summation according to the proximity of the reach time.

[0024] The cumulative competitor reach pressure value is calculated using the following formula: ; in, Indicates the first The cumulative competitor reach pressure value at each channel touchpoint. This represents the upper limit of the summation. Indicates the summation index. Indicates the first Competitive pressure values ​​for each time interval Indicates the moment when the current channel touchpoint occurs. Indicates the first The moment when a competitor sends a signal. Indicates the time decay coefficient. This represents the natural constant. This attenuation weighting method assigns lower weights to competitor outreach signals from more distant times and higher weights to competitor outreach signals from more recent times, in order to reflect the timeliness impact of competitor outreach.

[0025] Furthermore, the time variable in the formula and Use a uniform time unit for calculations, such as hours, to ensure accurate time difference calculations. Dimensionality consistency. Exponential function. As a dimensionless attenuation weighting coefficient, and related to the dimensionless interval competitive pressure value. Multiplying these values ​​yields the cumulative competitor reach pressure value. It is a dimensionless numerical value.

[0026] Furthermore, to accommodate the differences in decision-making cycles among different types of insurance products, a time decay coefficient is used. Differentiated settings are applied based on the type of insurance product. For short-term insurance products with shorter decision-making cycles, customers primarily consider recent information when making decisions, and the influence of past competitor outreach decays rapidly. Therefore, a larger time decay coefficient is set, for example... For long-term life insurance products with long decision-making cycles, the customer decision-making process is lengthy, and the impact of historical competitor outreach lasts longer. Therefore, a smaller time decay coefficient is set, for example... .

[0027] Step 3, constructing a counterfactual touchpoint sequence with missing touchpoints: For each customer's channel touchpoint sequence, iterate through each channel touchpoint in the channel touchpoint sequence, remove the target channel touchpoint from the original channel touchpoint sequence, and retain the temporal relationship and attribute information of other channel touchpoints to form a counterfactual touchpoint sequence with one less channel touchpoint, while retaining the original competitor's reach timeline data.

[0028] Furthermore, to more accurately simulate the scenario of missing channel touchpoints, the time interval between adjacent channel touchpoints before and after the removed channel touchpoint is adjusted when constructing the counterfactual touchpoint sequence. Specifically, the subsequent channel touchpoints of the removed channel touchpoint are shifted forward on the time axis by the shift amount equal to the time interval between the removed channel touchpoint and its preceding channel touchpoint, making the time distribution of channel touchpoints in the counterfactual touchpoint sequence more consistent with the hypothetical scenario where the channel touchpoint did not occur.

[0029] Step 4: Inferring changes in the probability of downstream touchpoints based on the channel causal relationship graph, and predicting the counterfactual conversion probability: Inferring changes in the probability of downstream channel touchpoints in the counterfactual touchpoint sequence based on the channel causal relationship graph. The channel causal relationship graph is a directed graph structure describing the causal transmission relationship between different channel touchpoints. Nodes in the channel causal relationship graph represent channel types, directed edges represent the causal influence relationship between channels, and the weight of the directed edges in the channel causal relationship graph represents the conditional probability of the upstream channel triggering the occurrence of the downstream channel.

[0030] Furthermore, the channel causal relationship map is obtained through statistical analysis of the co-occurrence frequency and sequential relationship of channel touchpoints in historical data. Specifically, the frequency of each channel pair appearing sequentially in the customer channel touchpoint sequence is statistically analyzed, and the conditional probability of the downstream channel appearing after the upstream channel appears is calculated. The conditional probability of the downstream channel appearing after the upstream channel appears is calculated using the following formula: ; in, Indicates channel After the touchpoint appears, the channel The conditional probability of a contact point occurring. Indicates upstream channels, Indicates downstream channels, Indicates channel After the touchpoint appears, the channel The number of customers whose touchpoints appear in sequence. Indicates that channels are included. Total number of customers at each touchpoint. Establish causal edge connections between channel pairs whose conditional probabilities exceed a preset threshold, forming a channel causal relationship graph.

[0031] Furthermore, the preset threshold is set to 0.3 based on business experience. This preset threshold is used to filter channel pairs with significant causal transmission relationships. When the conditional probability is below 0.3, it indicates that the upstream channel has a weak triggering effect on the downstream channel, and no causal edge connection is established. When the conditional probability reaches or exceeds 0.3, it indicates that the upstream channel has a significant triggering effect on the downstream channel, and a causal edge connection is established with the conditional probability as the weight of the causal edge.

[0032] The process of inferring the probability change of downstream channel touchpoints includes: when the target channel touchpoint is removed, searching for all outgoing edges with the channel corresponding to the target channel touchpoint as the source node in the channel causal relationship graph, obtaining the downstream channels pointed to by each outgoing edge and the corresponding conditional probability weights, marking the channel touchpoints belonging to the downstream channels in the counterfactual touchpoint sequence as probabilistic existence states, the existence probability of the channel touchpoints in the probabilistic existence state is equal to the original existence probability multiplied by the decay factor, and the decay factor is 1 minus the conditional probability weight of the corresponding causal edge.

[0033] Furthermore, channel touchpoints with probabilistic existence are included in feature calculations through a weighted approach in subsequent conversion probability prediction. Specifically, when counting the number of channel touchpoints in the counterfactual touchpoint sequence, channel touchpoints with probabilistic existence are counted by weight according to their existence probability; when calculating the distribution characteristics of channel touchpoint types, channel touchpoints with probabilistic existence are counted by weight according to their existence probability; when calculating features such as channel touchpoint time intervals and interaction depth, the feature values ​​of channel touchpoints with probabilistic existence are accumulated by weight according to their existence probability, so that the feature vector of the counterfactual touchpoint sequence can reflect the change in the existence probability of downstream channel touchpoints due to the absence of upstream channel touchpoints.

[0034] The counterfactual touchpoint sequence and the cumulative competitor reach pressure value for the corresponding time period are jointly input into the conversion probability prediction model with competitor covariates, and the output is an estimated counterfactual conversion probability value controlled by competitor covariates. The conversion probability prediction model with competitor covariates takes the touchpoint sequence feature vector and the cumulative competitor reach pressure value as input, and outputs the conversion probability. The touchpoint sequence feature vector includes the number of channel touchpoints, the distribution of channel touchpoint types, the statistics of channel touchpoint time intervals, and the channel touchpoint interaction depth index.

[0035] Furthermore, before inputting the touchpoint sequence feature vector and cumulative competitor reach pressure value into the conversion probability prediction model with competitor covariates, data preprocessing is performed on various features to eliminate the impact of dimensional differences on the training and prediction of the conversion probability prediction model with competitor covariates. Specifically, numerical features such as the number of channel touchpoints, the statistics of channel touchpoint time intervals, and the channel touchpoint interaction depth index are standardized using Z-scores; categorical features such as the distribution of channel touchpoint types are converted into numerical features using one-hot encoding; and the cumulative competitor reach pressure value is normalized to the mean based on the range, scaling the cumulative competitor reach pressure value to the range of 0 to 1. After data preprocessing, the touchpoint sequence feature vectors are comparable in all dimensions, making them suitable as input to the conversion probability prediction model with competitor covariates for joint modeling.

[0036] The conversion probability prediction model with competitor covariates incorporates the cumulative competitor reach pressure value as an independent covariate into the prediction process, so that the conversion probability output by the conversion probability prediction model with competitor covariates can reflect the conversion possibility under a specific competitor pressure level.

[0037] Furthermore, when the conversion probability prediction model with competitor covariates uses a logistic regression model, the input layer receives a concatenated feature vector. This concatenated feature vector consists of a pre-processed touchpoint sequence feature vector and cumulative competitor reach pressure values. The output layer of the conversion probability prediction model with competitor covariates outputs the predicted conversion probability value through a fully connected layer and a sigmoid activation function. The training samples include touchpoint sequence features and cumulative competitor reach pressure values ​​from historical customers who made purchases and those who did not, with 1 representing customers who made purchases and 0 representing those who did not. The conversion probability prediction model with competitor covariates is trained using a binary cross-entropy loss function. The parameters of the conversion probability prediction model with competitor covariates are iteratively optimized using a gradient descent algorithm to minimize the binary cross-entropy loss function value, thereby making the predicted conversion probability of the conversion probability prediction model with competitor covariates approximate the true label.

[0038] Furthermore, when the conversion probability prediction model with competitor covariates adopts the gradient boosting tree model, the input features of the model include the features of each dimension of the touchpoint sequence feature vector after data preprocessing and the cumulative competitor reach pressure value. The output of the model is the predicted conversion probability. The model iteratively constructs multiple decision trees, with each tree fitting the residuals of the preceding decision trees. Finally, the prediction results of all decision trees are summed to obtain the conversion probability. The training samples include touchpoint sequence features and cumulative competitor reach pressure value data for historical customers who made purchases and those who did not. Customers who made purchases are labeled 1, and those who did not are labeled 0. The model is trained using a logarithmic loss function. The gradient boosting algorithm iteratively optimizes the weights of the split nodes and leaf nodes of each decision tree to minimize the logarithmic loss function value, thereby improving the prediction accuracy of the model.

[0039] Step 5, calculate the pure causal incremental contribution value of the touchpoint: Compare the counterfactual conversion probability estimate with the actual conversion result, calculate the conversion probability difference after competitor control, and use the conversion probability difference as the pure causal incremental contribution value of the channel touchpoint. For customers who have already made a purchase, the conversion indicator value of the actual conversion result is 1. The pure causal incremental contribution value represents the decrease in conversion probability after removing a certain channel touchpoint. The formula for calculating the pure causal incremental contribution value is: ; in, Indicates channel touchpoints The pure causal incremental contribution value, Indicates the channel touchpoint index. Indicates removal of channel touchpoints The estimated counterfactual conversion probability value. A larger pure causal increment contribution value indicates a higher channel touchpoint. The higher the contribution to the pure incremental change of the conversion.

[0040] Furthermore, in order to normalize the sum of the pure causal incremental contribution values ​​of all channel touchpoints for the same customer to 1, the original pure causal incremental contribution values ​​of each channel touchpoint are proportionally adjusted.

[0041] The adjusted formula for calculating the pure causal increment contribution value is as follows: ; in, Indicates channel touchpoints The normalized pure causal increment contribution value, Indicates channel touchpoints The pure causal incremental contribution value, This indicates the total number of channel touchpoints in the customer channel touchpoint sequence. Indicates the summation index. Indicates channel touchpoints The pure causal incremental contribution value.

[0042] Step 6: Aggregate the pure causal incremental contribution value by channel identifier and competitor pressure level to generate channel value analysis results: Divide the cumulative competitor reach pressure value into three competitor pressure levels: low pressure, medium pressure, and high pressure. For example, the cumulative competitor reach pressure value in the top 33 percentile is classified as low pressure, the cumulative competitor reach pressure value in the 33% to 67th percentile is classified as medium pressure, and the cumulative competitor reach pressure value in the bottom 33 percentile is classified as high pressure.

[0043] The incremental contribution values ​​of pure causal relationships are aggregated in layers according to channel identifier and competitor pressure level, generating an incremental contribution distribution matrix and a channel competitor sensitivity ranking list. The incremental contribution distribution matrix is ​​a two-dimensional matrix structure. The row index of the incremental contribution distribution matrix is ​​the channel identifier, the column index is the competitor pressure level, and the matrix elements are the mean or sum of the pure causal incremental contribution values ​​of all channel touchpoints under the corresponding competitor pressure level.

[0044] Channel competitor sensitivity represents the degree of fluctuation in the net causal incremental contribution value of a channel as the level of competitor pressure changes. Channel competitor sensitivity is obtained by calculating the coefficient of variation of the net causal incremental contribution value of the same channel under different competitor pressure levels: ; in, Indicates channel Channel competitor sensitivity Indicates channel index, Indicates channel The standard deviation of the mean of the pure causal incremental contribution value under each competitor's pressure level. Indicates channel The arithmetic mean of the net causal incremental contribution values ​​under each competitor pressure level. The channel competitor sensitivity ranking list is arranged from highest to lowest channel competitor sensitivity.

[0045] The output includes the customer resource channel value analysis results, which include an incremental contribution distribution matrix and a ranking list of channel competitor sensitivity.

[0046] Furthermore, the customer resource channel value analysis results also include a statistical table of sample size for each channel under each competitor pressure level. This sample size statistical table is used to evaluate the statistical reliability of the analysis results of each cell in the incremental contribution distribution matrix. When the sample size of a certain channel under a certain competitor pressure level is lower than the preset minimum sample size threshold, the corresponding cell in the incremental contribution distribution matrix is ​​marked with an insufficient confidence indicator.

[0047] Furthermore, the minimum sample size threshold is set to 30 based on statistical significance requirements. When the sample size is less than 30, the variance of the statistical results is large, and there may be significant random fluctuations, which are insufficient to support reliable decision-making conclusions. Therefore, it is necessary to mark the results with insufficient confidence to remind users to interpret the analysis results of the corresponding cells of the incremental contribution distribution matrix with caution.

[0048] This implementation constructs a counterfactual contact sequence with missing contacts and infers changes in the probability of downstream channel contacts based on a channel causal relationship graph. This allows it to consider the collateral impact of a channel contact on downstream channel contacts when evaluating the contribution of a particular channel contact. Therefore, when the contribution of a channel contact originates from downstream channel contacts triggered by that contact, this mechanism can attribute part of the downstream channel contact's contribution to the upstream channel, thereby separating the channel's true incremental contribution from the passive absorption effect. This solves the problem of existing methods treating channel contacts as independent contributors while ignoring the causal transmission relationship between channels.

[0049] Meanwhile, this implementation calculates the cumulative competitor reach pressure value and incorporates it as a covariate into the conversion probability prediction model with competitor covariates. This allows the conversion probability prediction model with competitor covariates to separate the independent influence of competitor reach factors when predicting conversion probabilities. Therefore, in the counterfactual conversion probability estimation process, the competitor pressure level is controlled to be the same as the original channel touchpoint sequence, ensuring that the conversion probability difference only reflects the pure incremental effect of the company's own channel touchpoints. This solves the problem of existing methods overestimating the contribution of the company's own channels under high competitor pressure and underestimating the pure effect of the company's own channels under low competitor pressure, which is caused by not considering competitor factors.

[0050] Furthermore, by aggregating customer resource channel value analysis results in a tiered manner according to channel identifiers and competitor pressure levels, this implementation method generates an incremental contribution distribution matrix and a channel competitor sensitivity ranking list. This provides the insurance data platform with differentiated channel placement decision-making basis, enabling the selection of channels with corresponding competitor sensitivity characteristics for placement in market environments with different competitive intensities, thereby improving the consistency between customer resource acquisition placement decisions and actual results.

[0051] At least one embodiment of the present invention discloses a customer resource data analysis system for the insurance industry, comprising the following modules: The data acquisition and pressure value calculation module is used to acquire channel touchpoint sequence data of customers who have made transactions and competitor touchpoint signal data during the same period. It aligns the competitor touchpoint signals to each time interval of the channel touchpoint sequence according to time and calculates the interval competitor pressure value at the time of each channel touchpoint. The cumulative pressure value calculation module is used to calculate the cumulative competitive product reach pressure value of a customer at each stage of the conversion path by weighted summation based on the proximity of the reach time, according to the interval competitive product pressure value sequence. The counterfact sequence construction module is used to iterate through each channel touchpoint in the channel touchpoint sequence for each customer who has made a purchase, remove the target channel touchpoint from the original channel touchpoint sequence, and form a counterfact sequence. The counterfact conversion probability prediction module is used to infer the change in the probability of the existence of downstream channel touchpoints in the counterfact touchpoint sequence based on the channel causal relationship map. It inputs the counterfact touchpoint sequence and the cumulative competitor reach pressure value of the corresponding time period into the conversion probability prediction model with competitor covariates, and outputs the counterfact conversion probability estimate. The contribution value calculation module is used to compare the counterfactual conversion probability estimate with the actual conversion result and calculate the conversion probability difference as the pure causal incremental contribution value of the channel touchpoint. The analysis results generation module is used to aggregate pure causal incremental contribution values ​​in layers according to channel identifier and competitor pressure level, and generate channel value analysis results including an incremental contribution distribution matrix and a channel competitor sensitivity ranking list.

[0052] Based on the above methods and systems, the following examples are applied, with the specific application steps as follows: S1: Obtain channel touchpoint sequence data and competitor reach signal data, and calculate the competitive pressure value within the interval; Obtain complete channel touchpoint sequence data for customers who have already made a purchase, as well as competitor outreach signal data from the same period. Channel touchpoint sequence data includes the sequential channel touchpoints the customer interacted with before making a purchase, along with their timestamps. Competitor outreach signal data includes third-party advertising exposure records, competitor insurance app activity records, and insurance comparison platform access records.

[0053] Third-party ad exposure records are obtained through data interfaces with third-party ad monitoring platforms. These interfaces return the timestamps and ad identifiers of competitor ads that customers encountered within a specific time period. Competitor insurance app activity records are obtained through data interfaces with mobile device behavior data service providers. These interfaces return the launch time, dwell time, and interaction records of customers within competitor insurance apps. Insurance comparison platform access records are obtained through data cooperation channels with insurance comparison platforms. These interfaces return the timestamps of customer visits to insurance comparison platforms, the types of insurance products viewed, and the dwell time.

[0054] Competitor outreach signals are aligned chronologically to the time intervals of the channel touchpoint sequence, and the interval competitor pressure value is calculated for each channel touchpoint occurrence time. The interval competitor pressure value is obtained by weighting the number and intensity of competitor outreach signals within the time window of the channel touchpoint occurrence time, and is used to quantify the degree of competitor interference experienced by the customer at a specific channel touchpoint. The time window is set from 24 hours prior to the channel touchpoint occurrence time to the channel touchpoint occurrence time.

[0055] The competitive pressure value for a given time window is calculated as follows: All competitor outreach signals are statistically analyzed within the time window. Different intensity weights are assigned to different types of competitor outreach signals. Specifically, the intensity weight for third-party advertising exposure records is set to 0.3, the intensity weight for competitor insurance APP activity records is set to 0.5, and the intensity weight for insurance comparison platform access records is set to 0.7. The intensity weights of each competitor outreach signal are summed to obtain the competitive pressure value for that time window.

[0056] In March 20XX, an insurance company conducted a channel contribution analysis on its existing critical illness insurance customers. For example... Figure 2 As shown, this illustrates the level of competitor interference experienced by client C20240315 at different channel touchpoints. Before making a purchase, client C20240315 experienced multiple channel touchpoints, including social media advertising, website visits, telephone inquiries, and in-store consultations. Simultaneously, this client also received marketing information from several competing insurance companies during this period.

[0057] Table 1. Channel Touchpoint Sequence Data for Customer C20240315

[0058] Table 2 Competitor outreach data for customer C20240315

[0059] For touchpoint T002 (official website visit, 20XX-03-01 14:56), its time window is the 24-hour period from 20XX-02-28 14:56 to 20XX-03-01 14:56. Within this time window, there is a competitor's reach signal S001 (third-party ad exposure, weight 0.3). Therefore, the competitive pressure value for touchpoint T002 is: ; For touchpoint T004 (official website visit, 09:18 on 20XX-03-03), its time window is the 24-hour period from 09:18 on 20XX-03-02 to 09:18 on 20XX-03-03. Within this time window, there are competitor outreach signals S003 (price comparison platform visit, weight 0.7) and S004 (third-party ad exposure, weight 0.3). Therefore, the competitive pressure value for touchpoint T004 is as follows: ; Table 3. Calculation results of competitive pressure values ​​for each channel touchpoint.

[0060] S2: Calculate the cumulative competitor reach pressure value at each stage of the conversion path; Based on the competitor outreach signal sequence, the cumulative competitor outreach pressure value for each stage of the conversion path is calculated by weighting and accumulating the signals according to their proximity in time. For example... Figure 3 As shown, this illustrates the evolution of the cumulative competitor reach pressure value for a customer at each stage of the conversion path. This insurance product is a long-term life insurance product with a long decision-making cycle; therefore, the time decay coefficient is set to... .

[0061] For contact point T004, its cumulative competitor reach pressure value needs to be calculated. The occurrence time of contact point T004 is 20XX-03-03 09:18, i.e., the timestamp. The timeframe is 20XX-03-01 00:00. Competitor signals prior to T004 include S001, S002, S003, and S004.

[0062] Signal S001 occurred at 18:45 on 20XX-02-28, that is... Hour, intensity weight ; Signal S002 occurred at 20:30 on March 1, 20XX, that is... Hour, intensity weight ; Signal S003 occurred at 11:20 on March 2, 20XX. Hour, intensity weight ; Signal S004 occurred at 19:30 on March 2, 20XX. Hour, intensity weight ; The cumulative competitor reach pressure value for contact T004 is calculated as follows:

[0063]

[0064]

[0065] ; Table 4. Cumulative competitor reach pressure values ​​for each channel touchpoint

[0066] S3: Construct a counterfactual contact sequence with missing contact points; For the channel touchpoint sequence of customer C20240315, each channel touchpoint is traversed. The target channel touchpoint is removed from the original channel touchpoint sequence, while the temporal relationships and attribute information of other channel touchpoints are preserved, forming a counterfactual touchpoint sequence. After removing the target touchpoint, all touchpoints after that touchpoint are shifted forward on the timeline by a shift duration equal to the time interval between the removed touchpoint and its preceding touchpoint.

[0067] Taking the removal of touchpoint T003 (search engine advertisement) as an example, a counterfactual touchpoint sequence is constructed. In the original sequence, T003 occurs at 09:15 on March 3, 20XX, its preceding touchpoint is T002 (14:56 on March 1, 20XX), and its subsequent touchpoint is T004 (09:18 on March 3, 20XX). The time interval between T003 and T002 is 42.32 hours. After removing T003, T004 and subsequent touchpoints are shifted forward 42.32 hours on the timeline.

[0068] Table 5 Counterfactual contact sequence after removing contact T003

[0069] S4: Based on the channel causal relationship map, infer the probability changes of downstream touchpoints and predict the counterfactual conversion probability; Based on historical customer touchpoint sequence data, the insurance company constructed a channel causal relationship graph. Statistical analysis revealed that search engine advertising channels have a significant triggering effect on website visits.

[0070] Table 6 Key causal edges in the channel causal relationship graph

[0071] After removing touchpoint T003 (search engine ad), according to the channel causality graph, the conditional probability of search engine ad access to the official website is 0.68. Therefore, the probability of the existence of the official website access touchpoint T004 after T003 in the original sequence needs to be adjusted.

[0072] The probability adjustment factor for the existence of T004 is: ; The probability of T004 existing in the counterfactual sequence is: ; The feature vectors of the counterfactual touchpoint sequence and the cumulative competitor reach pressure value are input into a conversion probability prediction model with competitor covariates. This model uses logistic regression and has been trained on historical data.

[0073] Table 7. Eigenvectors of Counterfactual Contact Sequences

[0074] like Figure 9 As shown, the features of the original sequence and the counterfactual sequence are compared. After Z-score standardization and normalization, the feature vectors are input into a conversion probability prediction model with competitor covariates. The output is the estimated counterfactual conversion probability after removing touchpoint T003. .

[0075] S5: Calculate the pure causal incremental contribution value of the contact point; Customer C20240315 actually completed a transaction, with a conversion indicator value of 1. The counterfactual conversion probability estimate after removing touchpoint T003 is 0.73; therefore, the pure causal incremental contribution of touchpoint T003 is: ; The same method is used to calculate the pure causal increment contribution value of each other contact. The sum of the original contribution values ​​of each contact is... .

[0076] like Figure 4As shown, this displays the normalized causal contribution value of each channel touchpoint to customer conversion. The pure causal incremental contribution value of each touchpoint is normalized so that the sum of the normalized contribution values ​​of all touchpoints equals 1. The normalized pure causal incremental contribution value of touchpoint T003 is: ; Table 8. Pure Causal Incremental Contribution Value of Each Channel Touchpoint

[0077] S6: Aggregate pure causal incremental contribution values ​​by channel identifier and competitor pressure level to generate channel value analysis results; The insurance company conducted a channel contribution analysis on 500 critical illness insurance customers who made purchases in March 20XX. Based on the cumulative competitor reach pressure value at the time of each customer's final purchase, competitor pressure levels were categorized using the 33rd and 67th percentiles. A cumulative competitor reach pressure value below 1.2 was classified as low pressure, between 1.2 and 2.5 as medium pressure, and above 2.5 as high pressure.

[0078] like Figure 5 As shown, the distribution of average incremental contribution values ​​of each channel under low-pressure, medium-pressure, and high-pressure environments is displayed. The pure causal incremental contribution values ​​of all customer channel touchpoints are aggregated in layers according to channel identification and competitor pressure level, and the average contribution value of each channel under different competitor pressure levels is calculated.

[0079] Table 9 Incremental Contribution Distribution Matrix

[0080] Calculate the competitor sensitivity for each channel. Taking search engine advertising as an example, the average contribution values ​​of this channel at low, medium, and high pressure levels are 0.215, 0.183, and 0.142, respectively.

[0081] The arithmetic mean is: ; The standard deviation is: ; Channel competitor sensitivity is: ; Table 10. Ranking of Channel Competitor Sensitivity

[0082] like Figures 6-8 As shown, Figure 6 The sensitivity of each channel to changes in competitor pressure is displayed in order. Figure 7 Show the distribution relationship between the average contribution value of each channel and competitor sensitivity; Figure 8 Display the strength weight configuration of different types of competitor outreach signals; Based on channel value analysis, the insurance company found that social media and search engine advertising were highly sensitive to competitors, with their contribution significantly decreasing under high competitive pressure. In contrast, telephone inquiries and offline stores showed lower competitor sensitivity, with their contribution remaining relatively stable under varying competitive pressure. Therefore, in highly competitive market areas, the company should increase resource investment in telephone inquiries and offline stores, while in less competitive market areas, it can increase investment in social media and search engine advertising to maximize the efficiency of customer acquisition.

[0083] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for analyzing customer resource data in the insurance industry, characterized in that, Includes the following steps: Acquire channel touchpoint sequence data of customers who have made a purchase and competitor reach signal data during the same period. Align the competitor reach signals to each time interval of the channel touchpoint sequence according to time and calculate the interval competitor pressure value at the time of occurrence of each channel touchpoint. Based on the interval competitor pressure value sequence, the cumulative competitor reach pressure value of the customer at each stage of the conversion path is calculated by weighted summation according to the proximity of the reach time. For each customer's channel touchpoint sequence, iterate through each channel touchpoint in the sequence and remove the target channel touchpoint from the original sequence to form a counterfactual touchpoint sequence. Based on the channel causal relationship map, the probability change of the existence of downstream channel touchpoints in the counterfactual touchpoint sequence is inferred. The counterfactual touchpoint sequence and the cumulative competitor reach pressure value of the corresponding time period are jointly input into the conversion probability prediction model with competitor covariates, and the counterfactual conversion probability estimate is output. The counterfactual conversion probability estimate is compared with the actual conversion result, and the conversion probability difference is calculated as the pure causal incremental contribution value of the channel touchpoint. The pure causal incremental contribution values ​​are aggregated in layers according to channel identifier and competitor pressure level, generating channel value analysis results that include an incremental contribution distribution matrix and a ranking list of channel competitor sensitivity.

2. The customer resource data analysis method in the insurance industry according to claim 1, characterized in that, The competitor outreach signal data includes third-party advertising exposure records, competitor insurance app activity records, and insurance comparison platform access records. The interval competitor pressure value is calculated as follows: all competitor outreach signals within the time window of the channel touchpoint occurrence time are statistically analyzed, different intensity weights are assigned to different types of competitor outreach signals, and the intensity weights of each competitor outreach signal are summed to obtain the interval competitor pressure value. Among them, the intensity weight of third-party advertising exposure records is lower than that of competitor insurance app activity records, and the intensity weight of competitor insurance app activity records is lower than that of insurance comparison platform access records.

3. The customer resource data analysis method in the insurance industry according to claim 1, characterized in that, The cumulative competitor reach pressure value is obtained by exponentially decaying weighted summation of the competitor pressure value sequence in the interval. The decay weight is a negative exponential function of the natural constant. The exponent of the negative exponential function is the negative value of the product of the time decay coefficient and the time difference between the current channel touch point occurrence time and the competitor reach signal occurrence time.

4. The customer resource data analysis method in the insurance industry according to claim 3, characterized in that, The time decay coefficient is set differently according to the type of insurance product. A high time decay coefficient is set for short-term insurance products with short decision-making cycles, and a low time decay coefficient is set for long-term life insurance products with long decision-making cycles.

5. The customer resource data analysis method in the insurance industry according to claim 1, characterized in that, When constructing the counterfactual contact sequence, the time interval between adjacent channel contacts before and after the removed channel contact is adjusted, and the subsequent channel contacts of the removed channel contact are shifted forward on the time axis by the amount of shifting, which is the time interval between the removed channel contact and its preceding channel contact.

6. The customer resource data analysis method in the insurance industry according to claim 1, characterized in that, The channel causal relationship graph is a directed graph structure that describes the causal transmission relationship between different channel touchpoints. The nodes of the channel causal relationship graph represent channel types, the directed edges of the channel causal relationship graph represent the causal influence relationship between channels, and the weight of the directed edges of the channel causal relationship graph represents the conditional probability of the upstream channel triggering the occurrence of the downstream channel. The conditional probability is obtained by statistically analyzing the ratio of the number of customers whose downstream channel touchpoints appear in sequence after the upstream channel touchpoint appears to the total number of customers including the upstream channel touchpoint. Causal edges are established to connect channel pairs whose conditional probabilities exceed a preset threshold.

7. A method for analyzing customer resource data in the insurance industry according to claim 1 or 6, characterized in that, The change in the probability of the existence of downstream channel touchpoints in the counterfactual touchpoint sequence includes: after the target channel touchpoint is removed, searching for all outgoing edges with the channel corresponding to the target channel touchpoint as the source node in the channel causal relationship graph, obtaining the downstream channels pointed to by each outgoing edge and the corresponding conditional probability weights, marking the channel touchpoints belonging to downstream channels in the counterfactual touchpoint sequence as probabilistic existence states, the existence probability of the channel touchpoints in probabilistic existence states is equal to the original existence probability multiplied by a decay factor, the decay factor being 1 minus the conditional probability weight of the corresponding causal edge; in the subsequent conversion probability prediction, the channel touchpoints in probabilistic existence states participate in feature calculation by weighting according to their existence probability.

8. The customer resource data analysis method in the insurance industry according to claim 1, characterized in that, The conversion probability prediction model with competitor covariates takes the touchpoint sequence feature vector and the cumulative competitor reach pressure value as input, and outputs the conversion probability. The touchpoint sequence feature vector includes the number of channel touchpoints, the distribution of channel touchpoint types, the statistics of channel touchpoint time intervals, and the channel touchpoint interaction depth index. Before inputting the conversion probability prediction model with competitor covariates, the numerical features are standardized, the categorical features are converted using one-hot encoding, and the cumulative competitor reach pressure value is normalized.

9. A method for analyzing customer resource data in the insurance industry according to claim 1, characterized in that, The pure causal incremental contribution value is equal to the actual conversion identifier value minus the counterfactual conversion probability estimate; the pure causal incremental contribution value of each channel touchpoint is normalized, and the normalized pure causal incremental contribution value is equal to the pure causal incremental contribution value of that channel touchpoint divided by the sum of the pure causal incremental contribution values ​​of all channel touchpoints in the customer channel touchpoint sequence; the channel competitor sensitivity is obtained by calculating the coefficient of variation of the mean pure causal incremental contribution value of the same channel under different competitor pressure levels, and the coefficient of variation is equal to the standard deviation divided by the arithmetic mean.

10. A customer resource data analysis system for the insurance industry, characterized in that, The method for performing the steps in a customer resource data analysis method in the insurance industry as described in any one of claims 1-9 includes: The data acquisition and pressure value calculation module is used to acquire channel touchpoint sequence data of customers who have made transactions and competitor touchpoint signal data during the same period. It aligns the competitor touchpoint signals to each time interval of the channel touchpoint sequence according to time and calculates the interval competitor pressure value at the time of each channel touchpoint. The cumulative pressure value calculation module is used to calculate the cumulative competitive product reach pressure value of a customer at each stage of the conversion path by weighted summation based on the proximity of the reach time, according to the interval competitive product pressure value sequence. The counterfact sequence construction module is used to iterate through each channel touchpoint in the channel touchpoint sequence for each customer who has made a purchase, remove the target channel touchpoint from the original channel touchpoint sequence, and form a counterfact sequence. The counterfact conversion probability prediction module is used to infer the change in the probability of the existence of downstream channel touchpoints in the counterfact touchpoint sequence based on the channel causal relationship map. It inputs the counterfact touchpoint sequence and the cumulative competitor reach pressure value of the corresponding time period into the conversion probability prediction model with competitor covariates, and outputs the counterfact conversion probability estimate. The contribution value calculation module is used to compare the counterfactual conversion probability estimate with the actual conversion result and calculate the conversion probability difference as the pure causal incremental contribution value of the channel touchpoint. The analysis results generation module is used to aggregate pure causal incremental contribution values ​​in layers according to channel identifier and competitor pressure level, and generate channel value analysis results including an incremental contribution distribution matrix and a channel competitor sensitivity ranking list.

Images