JP5525501B2 - Customer survival rate estimation system, customer survival rate estimation method, customer survival rate estimation processing server, and program for customer survival rate estimation processing server - Google Patents

Description

  The present invention relates to a technique for estimating the survival rate of a customer in the marketing field.

  In marketing analysis, it is necessary to select good customers who will continue to buy their own brands in the future when narrowing down the subject of direct mail and planning brand strategies.

  Conventionally, as a method for estimating a customer survival rate, for example, a method described in Patent Document 1 has been known.

  According to the method described in Patent Literature 1, it is possible to estimate the survival rate in consideration of the visit cycle of each individual, assuming that purchase and withdrawal occur randomly.

Makoto Abe, “Personal RF Analysis Based on Consumer Behavior Theory-Improvement of Pareto / NBD Model by Hierarchical Bayes”, Journal of the Japan Statistical Society, Vol. 37, No. 2, March 2008, pp. 239-259

  Since the customer survival rate estimation method described in Patent Document 1 is calculated based on purchase data, it cannot be applied to users who have not actually purchased. That is, for example, for a user who does not have a purchase history for a certain brand, the customer survival rate for that brand cannot be estimated.

  Therefore, it is necessary to estimate the survival rate applicable to users who have not made purchases.

  The present invention solves the above-described problems, and an object of the present invention is to provide a customer survival rate estimation system, method, and program that can be applied to users who have not actually purchased.

  The present invention uses Bayes' theorem to estimate including the survival rate (potential survival rate) of users who are not actually purchasing based on pre-purchase behavior (cart input, product browsing) data. In the following description, “pre-purchase behavior” is sometimes simply referred to as “pre-action”.

  The customer survival rate estimation system of the present invention receives a customer survival rate analysis request from an analyzer terminal operated by an analyst who analyzes customer survival rates, estimates a customer survival rate according to the analysis request, A customer survival rate estimation system including a customer survival rate estimation processing server that sends an estimated customer survival rate as a customer survival rate analysis result to the analyzer terminal, wherein the customer survival rate estimation processing server A user will make a purchase in the future based on the data stored in the log database when there is a customer survival rate analysis request from the analysis terminal and the log database storing data relating to purchase behavior and pre-purchase behavior A presence rate estimating means for calculating a customer survival rate indicating a probability and a pre-purchase behavior survival rate indicating a probability that the user will perform a pre-purchase behavior in the future; Based on the data stored in the service, the product of the probability of purchasing on the condition of pre-purchasing behavior of a user and the probability of purchasing on the condition of pre-purchasing behavior of a product is calculated as the purchase of all users and all products. By dividing the previous behavior by the probability that purchase will occur, the expected survival rate representing the purchase transition probability for a certain product of a certain user is obtained, and the expected survival rate corresponds to the user and product indicated by the expected survival rate The process of obtaining the estimated survival rate from the previous behavior by multiplying the survival rate before purchase calculated by the survival rate estimating means is executed for all users and all the behaviors before purchase, and is estimated from the obtained previous behavior. A potential survival rate estimation means for determining a final survival rate as a customer survival rate analysis result from the survival rate and the customer survival rate calculated by the survival rate estimation means. To.

(1) According to the invention described in claims 1 to 4, it can be applied to a user who does not actually make a purchase, and a potential survival rate of a user who does not make a purchase can be estimated.
(2) According to the invention described in claim 2, it is possible to provide a customer survival rate estimation method capable of estimating a potential survival rate of a user who has not made a purchase.
(3) According to the invention described in claim 3, it is possible to provide a customer survival rate estimation processing server capable of estimating a potential survival rate of a user who has not made a purchase.
(4) According to the invention described in claim 4, it is possible to provide a customer survival rate estimation processing server program capable of estimating the potential survival rate of a user who has not made a purchase.

The block diagram of the customer survival rate estimation system in the embodiment of this invention. The flowchart of the process which the survival rate estimation process part in the embodiment of this invention performs. The flowchart of the process which the potential survival rate estimation process part in the embodiment of this invention performs. The whole sequence diagram in the embodiment of the present invention. Explanatory drawing of the input and output data of the survival rate estimation process part in the embodiment of this invention. Explanatory drawing of the input and output data of the potential survival rate estimation process part in the embodiment of this invention. Explanatory drawing of parameter x, t, T used as the input of the survival rate estimation process part in the example of embodiment of this invention. The example of data used in the potential survival rate estimation processing unit in the embodiment of the present invention is represented, (a) is an explanatory diagram showing a data example of the number of days each user has purchased with each brand, (b) is each user Explanatory drawing which shows the example of data of the days which performed the previous action in each brand.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings, but the present invention is not limited to the following embodiments.

  Since the previous action is always required for the purchase, P (purchase) = P (previous action) P (purchase | previous action) is established for the survival rate = P (purchase) according to Bayes' theorem. Note that P (purchase | previous behavior) represents a conditional probability that the purchase occurs on the condition of the previous behavior.

  However, a user who has not made a purchase cannot calculate the survival rate because P (purchase | previous action) = 0 in the existing method, but in this embodiment, P (purchase | previous action) in another brand of the target user. ) And P (purchase | previous behavior) in the target brand of other users. Incorporating the expected value into the survival rate calculation instead of the actual measurement value makes it possible to estimate the survival rate of unpurchased people (probability of future purchases).

  FIG. 1 shows the overall configuration of the customer survival rate estimation system of the present invention. 2 is a flowchart of the survival rate estimation process performed by the survival rate estimation means of the present invention, FIG. 3 is a flowchart of the potential survival rate estimation process performed by the potential survival rate estimation means of the present invention, and FIG. It is a whole sequence diagram of a rate estimation processing server.

  In FIG. 1, reference numeral 100 denotes an analyzer terminal (computer terminal) operated by an analyst (for example, a marketer or product developer) who analyzes a customer survival rate, and 200 denotes a customer survival from the analyzer terminal 100. A customer survival rate estimation processing server that receives a rate analysis request, estimates a customer survival rate in response to the analysis request, and sends the estimated customer survival rate to the analyzer terminal 100 as a customer survival rate analysis result. The analyzer terminal 100 and the customer survival rate estimation processing server 200 are connected by, for example, a network (not shown).

  The customer survival rate estimation processing server 200 includes a log database 210 in which log data relating to purchase behavior and previous behavior of the user's products is stored, and the probability that the user will make purchases in the future based on the data stored in the log database 210. Based on the data stored in the log database 210, and the survival rate estimation processing unit 220 as a survival rate estimation means for calculating the customer survival rate and An expected survival rate representing a purchase transition probability for a certain product of a certain user is obtained, and an estimated survival rate from the previous behavior is calculated using the expected survival rate and the previous behavior survival rate calculated by the survival rate estimation processing unit 220. The final survival rate is calculated from the estimated survival rate from the previous behavior and the customer survival rate calculated by the survival rate estimation processing unit 220. And a potential survival estimation processing unit 230 as a potential survival estimation means for determining.

  The customer survival rate estimation processing server 200 is configured by, for example, a computer, and includes hardware resources of a normal computer, for example, ROM, RAM, CPU, input device, output device, communication interface, hard disk, recording medium, and driving device thereof. Yes.

  As a result of the cooperation between the hardware resource and the software resource (OS, application, etc.), the customer survival rate estimation processing server 200 has a log database 210, a survival rate estimation processing unit 220, and a potential survival rate as shown in FIG. A rate estimation processing unit 230 is implemented.

  It is assumed that the log database 210 is constructed in storage means / storage means such as a hard disk or RAM.

  Next, the operation of the system configured as described above will be specifically described. First, data stored in the log database 210 will be described. In the log database 210, data of parameters x, t, and T relating to purchase (= buy_) for each user, cart input (cart_) as the previous action, and product browsing (= view_) as the previous action (for example, (( 1), and the number of days when all users for each brand made purchases with a certain brand, and the number of days for which all users for each brand performed pre-actions (cart input, product browsing) with a brand. The data shown in (2) and (3) of FIG. 6), the number of days that a certain user for each user made purchases in any brand, and the number of days that a certain user for each user performed pre-actions in any brand. (For example, the data shown in (4) and (5) of FIG. 6).

  FIG. 5 shows an example of input data (1) and output data (6) (survival rates buy_life, cart_life, view_life for each purchase, cart input, and product browsing for each user) of the survival rate estimation processing unit 220. 6 is an example of input data (2), (3), (4), (5), (6) and output data (7) (survival rate for each user) of the potential survival rate estimation processing unit 230. Represents.

  The parameters x, t, and T are defined as shown in FIG. FIG. 7 shows the passage of time from the start of observation to the end of observation for each event of purchase, cart insertion, and product browsing, and * in the figure indicates that each event has occurred.

  The parameter x is the number of times (number of days) the event has occurred, and x = 3 in the example of FIG. The parameter t is the number of days elapsed from the first occurrence date of the event to the last occurrence date, and the parameter T is the number of days elapsed from the first occurrence date of the event to the observation end date.

  Next, details of the survival rate estimation processing unit 220 will be described. When a customer survival rate analysis request, for example, a brand (or EC site) to be compared is designated from the analysis terminal 100, the survival rate estimation processing unit 220 performs purchase, cart insertion, and product browsing shown in the flowchart of FIG. Each data acquisition process (steps S221, S224, S227), parameter estimation processing for each user regarding purchase, cart insertion, and product browsing (steps S222, S225, S228), purchase, cart insertion, and product browsing A survival rate calculation process (steps S223, S226, and S229) is executed. The specific processing is as follows.

<Data acquisition>
Process 0: Parameters x, t, and T relating to purchase, cart input, and product browsing are acquired for each user from the log database 210 (steps S221, S224, and S227). Specifically, for example, as in data (1) in FIG. 5, buy_x: number of days of purchase, buy_t: number of days elapsed from the first purchase date to the last purchase date, buy_T: from the first purchase date to the observation end date Elapsed days, cart_x: number of days of cart insertion, cart_t: number of days elapsed from first cart insertion date to last cart insertion date, cart_T: number of days elapsed from first cart insertion date to observation end date, view_x: product browsing Number of days, view_t: Elapsed days from first product viewing date to last product viewing date, view_T: Elapsed days from first product viewing date to observation end date.

  In this example, the previous action = cart input and product browsing is used, but other actions may be included. For example, when calculating the survival rate of the A brand, the above data is extracted by focusing on the purchase of the A brand and the previous behavior of the A brand (product browsing, cart insertion). The same applies to the determination of the survival rate for the a site (EC site).

<Parameter estimation for each user>
Here, a method of estimation by grid search is illustrated.

Process 1; User 1 is selected Process 2; Parameters x, t, and T are substituted into the survival rate likelihood function represented by Expression (1) (since it is a parameter for determining the survival rate of purchase in step S222, x → buy_x, t → buy_t, T → buy_T).

  Process 3: The purchase rate parameter λ and the withdrawal rate parameter μ in the formula (1) are changed from 0 to 1 for each division set by the service provider, and the value of each likelihood function is obtained.

  Process 4: λ and μ having the highest likelihood function are adopted as parameters of the user.

  Process 5: Change the user (add one ID) and return to Process 1. If the process is completed for all users, the process proceeds to process 6. In addition, existing methods relating to parameter estimation, such as a method of mathematically solving a parameter for which the differential value of the likelihood function is 0, can be applied.

<Purchase survival rate calculation>
Process 6: Substituting x, t, T, λ, and μ for each user into Expression (2) for calculating the survival rate, calculates the survival rate (step S223), and sets this as buy_life.

<Previous behavioral survival rate calculation>
Process 7: The same process is performed for carts (cart_x, cart_t, cart_T) (steps S225 and S226) and merchandise browsing (view_x, view_t, view_T) (steps S228 and S229) for the processes 1 to 6 and the calculated survival Let the rates be cart_life and view_life, respectively. An example of the calculated survival rate is the data (6) in FIGS. 5 and 6, where buy_life is the probability of future purchase (customer survival rate in the original sense), cart_life is the probability of future cart insertion, and view_life. Indicates the probability of future product browsing.

  Note that, when the process 2 is performed in step S225, the parameters x, t, and T to be substituted into the equation (1) are x → cart_x, t → cart_t, T → cart_T, and when processing 2 is performed in step S228, the parameters x, t, and T to be substituted into equation (1) are x → view_x, t → view_t, T → view_T

  Next, details of the potential survival rate estimation processing unit 230 will be described. The potential survival rate estimation processing unit 230 performs purchase transition probability estimation processing from cart insertion (step S231), purchase transition probability estimation processing from product browsing (step S232), and potential survival shown in the flowchart of FIG. The rate calculation process (step S233) is executed. The specific process is as follows.

<Estimated purchase transition probability (P (purchase | previous behavior)) of each user>
The following processes 8 to 12 are executed in steps S231 and S232.

  Process 8; pre-action = cart input.

  Process 9: User 1 is selected.

  Process 10: When it is going to estimate the survival rate of the A brand at the time of data acquisition in the process 0, the purchase survival rate = “the purchase transition probability of the A brand of the user 1” × the previous behavior survival rate. Here, the pre-behavioral survival rate has been calculated by the above-mentioned “Processing 7” (Steps S225, S226, S228, and S229 in FIG. 2) performed by the survival rate estimation processing unit 220. Is determined as an expected survival rate according to the following equation (3).

User 1's A Brand P (Purchase | Previous Action) = P in Brand A (Purchase | Previous Action) x User 1's P (Purchase | Previous Action) x k = {Number of Days All Users Purchased in A Brand / Number of days in which all users have made a pre-action with the A brand} × {Number of days in which the user 1 has made a purchase with any brand / Number of days in which the user 1 has made a pre-action with any brand} × k (3) )
Let Z be the expected survival rate expressed by this equation (3). K in equation (3) is a constant set by the service provider.

  Here, Expression (3) represents that the transition probability from the previous action is calculated based on the expected value, and the process will be described.

  Prior action is always required for purchase (P (previous action | purchase) = 1), so the survival rate = P (purchase) = P (previous action) P (purchase | previous action) is established by Bayes' theorem. To do. P (previous behavior) has already been calculated by the survival rate estimation process of FIG. P (purchase | previous behavior) = number of days of purchase / number of days of previous behavior, when X (1, A) / Y (1, A) in FIG. P (purchase | previous action).

  FIG. 8 (a) shows an example of data on the number of days each user made purchases in each brand, and FIG. 8 (b) shows an example of data on the number of days in which each user performed previous actions in each brand. The potential survival rate estimation processing unit 230 acquires from the data stored in the log database 210.

  However, for a user who has not made a purchase, X (1, A) is 0 as an actual measurement value. Therefore, an expected value of P (purchase | previous action) when the A brand and the category of the user 1 are assumed to be independent is obtained.

X (1, A) = N (., A) × N (1,.) / N (.,.)
Y (1, A) = M (., A) × M (1,.) / M (.,.)
Than,

  It becomes. Since the denominator of equation (4) is a constant, 1 / all users all brands P (purchase | previous behavior) is set to k, and equation (3) is obtained.

  Process 11: Multiply the expected survival rate Z of the expression (3) obtained in the process 10 by the previous action survival rate calculated by the survival rate estimation processing unit 220 (steps S226 and S229 in FIG. 2) corresponding to Z. And calculate the estimated survival rate from the previous action. For example, if previous action = cart input, cart_life × Z.

  Process 12: Change the user (add one ID), return to Process 9, and repeat the processes 10 to 11 for the user. If all users have been completed, the process returns to process 8 and the previous action is changed to product browsing, and processes 9 to 11 are repeatedly executed. If all users and all previous actions have been completed, the process proceeds to process 13.

<Potential survival rate calculation>
Process 13: For each user, out of the customer survival rate buy_life calculated by the survival rate estimation processing unit 220 and the estimated survival rate (previous behavior_life × Z) from the previous behavior obtained in the processing 11 and the processing 12 The larger one is output as the final survival rate (output_life), for example, as data (7) in FIG. 6 (step S233).

  In addition to the method of adopting the largest value, a method of employing an average value or a minimum value, a method of giving a weight to each survival rate, or the like may be employed.

  Process 14: When comparing the survival rates of a plurality of brands, the brand is changed, the process returns to Process 0, and the processes up to Process 13 are executed.

  The final survival rate output from the potential survival rate estimation processing unit 230 is transmitted from the customer survival rate estimation processing server 200 to the analyzer terminal 100 as an analysis result for the analysis request from the analyzer terminal 100. And displayed on, for example, a display of the terminal 100. Note that the same operation as described above is performed for the survival rate estimation at the EC site.

  The overall sequence of the above-described customer survival rate estimation processing server 200 will be described with reference to FIG. First, in the survival rate estimation process, purchase survival rate calculation (data (1)) is performed based on buy_x, buy_t, buy_T data, cart_x, cart_t, cart_T data, view_x, view_t, view_T data (data (1)) acquired from the log database 210. Step S223 in FIG. 2), cart insertion survival rate calculation (step S226 in FIG. 2), and product browsing survival rate calculation (step S229 in FIG. 2) are performed.

  Next, in the potential survival rate estimation process, the transition probability calculation source data (data (2), (3), (4), (5)) from the log database 210, the cart input survival rate calculation, the product browsing survival rate Based on the cart insertion survival rate data and the product browsing survival rate data (data (6)) calculated in the rate calculation, the cart insertion transition rate multiplication and the product browsing transition rate multiplication (multiplication of processing 11: step S231 in FIG. 3) S232) is performed.

  Next, the estimated survival rate from the previous action obtained by the multiplication (cart insertion survival rate × transition probability data, product browsing survival rate × transition probability data) and the purchase survival rate data obtained by the purchase survival rate calculation ( Potential survival rate calculation (step S233 in FIG. 3) is performed based on data (buy_life) of data (6).

  As described above, since the purchase transition probability of the user is obtained by an expected value (expected survival rate Z) instead of an actual measurement value, and the survival rate of the non-purchasing user is calculated by estimating the survival rate by Bayes' theorem, It is possible to estimate the potential survival rate of users who have not actually purchased.

  That is, in steps S231 and S232 performed by the potential survival rate estimation processing unit 230, as shown in Expression (4), data on the number of days each user made purchases in each brand and each user in each brand Since the expected value (Z) is estimated from the data on the number of days of the previous action and the expected value is incorporated in the survival rate calculation, the potential survival rate of the non-purchasing user can be estimated.

  Further, the present invention can be realized by configuring a part or all of the functions of each means in the customer survival rate estimation processing server of the present embodiment with a computer program and executing the program using the computer. It goes without saying that the procedure in the customer survival rate estimation method of the embodiment can be configured by a computer program and the program can be executed by the computer, and the computer can read the program for realizing the function on the computer. Recording media such as FD (Floppy (registered trademark) Disk), MO (Magneto-Optical disk), ROM (Read Only Memory), memory card, CD (Compact Disk) -ROM, DVD (Digital Ve) satile Disk) -ROM, CD-R, CD-RW, HDD, and recorded in a removable disk, or stored, it is possible or distribute. It is also possible to provide the above program through a network such as the Internet or electronic mail.

DESCRIPTION OF SYMBOLS 100 ... Analyst terminal 200 ... Customer survival rate estimation processing server 210 ... Log database 220 ... Survival rate estimation processing unit 230 ... Potential survival rate estimation processing unit

Claims (4)

Upon receiving a customer survival rate analysis request from an analyst terminal operated by an analyst who analyzes the customer survival rate, the customer survival rate is estimated according to the analysis request, and the estimated customer survival rate is analyzed as a customer survival rate As a result, a customer survival rate estimation system including a customer survival rate estimation processing server to be sent to the analyzer terminal,
The customer survival rate estimation processing server is:
A log database that stores data related to purchase behavior and pre-purchase behavior for users' products,
When there is a customer survival rate analysis request from the analysis terminal, based on the data stored in the log database, the customer survival rate indicating the probability that the user will make a purchase in the future, and the user will perform the pre-purchase behavior in the future A survival rate estimating means for calculating a pre-purchase behavioral survival rate indicating a probability,
Based on the data stored in the log database, the product of the probability of purchasing on the condition of pre-purchasing behavior of a user and the probability of purchasing on the condition of pre-purchasing behavior of a certain product is calculated for all users and all products. By dividing the pre-purchase behavior by the probability that purchase will occur, an expected survival rate representing the purchase transition probability for a certain product of a certain user is obtained, and the user and the product indicated by the expected survival rate in the expected survival rate A process for obtaining the estimated survival rate from the previous behavior by multiplying the survival rate before purchase calculated by the survival rate estimating means corresponding to the above is executed for all users, the behavior before purchase, and from the obtained previous behavior Potential survival rate estimation means for determining a final survival rate as a customer survival rate analysis result from the estimated survival rate of the customer and the customer survival rate calculated by the survival rate estimation means,
A customer survival rate estimation system characterized by comprising: Upon receiving a customer survival rate analysis request from an analyst terminal operated by an analyst who analyzes the customer survival rate, the customer survival rate is estimated according to the analysis request, and the estimated customer survival rate is analyzed as a customer survival rate As a result, a customer survival rate estimation method in a system including a customer survival rate estimation processing server to be sent to the analyzer terminal,
The analyzer terminal transmits a customer survival rate analysis request to a customer survival rate estimation processing server;
When the survival rate estimation means of the customer survival rate estimation processing server receives the customer survival rate analysis request, based on data in a log database in which data relating to purchase behavior and pre-purchase behavior for the user's product is stored A survival rate estimation step for calculating a customer survival rate indicating a probability that the user will make a purchase in the future and a pre-purchase behavior survival rate indicating a probability that the user will perform the purchase behavior in the future;
Based on the data stored in the log database, the potential survival rate estimation means of the customer survival rate estimation processing server sets the probability of purchase on the condition of a user's pre-purchase behavior and the pre-purchase behavior of a product. By dividing the product of the probability of purchasing as the pre-purchasing behavior of all users and all products by the probability of purchasing as a condition, the expected survival rate representing the purchase transition probability for a certain user is obtained, All users perform a process for multiplying the expected survival rate by the user indicated by the expected survival rate and the pre-purchase behavior survival rate calculated by the survival rate estimation means corresponding to the product to obtain the estimated survival rate from the previous behavior. Execute all pre-purchase behaviors, and calculate the final survival rate from the estimated survival rate from the determined previous behavior and the customer survival rate calculated by the survival rate estimation means. A potential survival estimating step of determining the rate analysis,
The customer survival rate estimation processing server transmits the determined customer survival rate analysis result to the analyzer terminal;
A customer survival rate estimation method characterized by comprising: A log database that stores data related to purchase behavior and pre-purchase behavior for users' products,
Based on the data stored in the log database, a survival rate estimation means for calculating a customer survival rate indicating a probability that the user will make a purchase in the future, and a pre-purchase behavior survival rate that the user will perform a purchase behavior in the future,
Based on the data stored in the log database, the product of the probability of purchasing on the condition of pre-purchasing behavior of a user and the probability of purchasing on the condition of pre-purchasing behavior of a certain product is calculated for all users and all products. By dividing the pre-purchase behavior by the probability that purchase will occur, an expected survival rate representing the purchase transition probability for a certain product of a certain user is obtained, and the user and the product indicated by the expected survival rate in the expected survival rate A process for obtaining the estimated survival rate from the previous behavior by multiplying the survival rate before purchase calculated by the survival rate estimating means corresponding to the above is executed for all users, the behavior before purchase, and from the obtained previous behavior Potential survival rate estimation means for determining a final survival rate from the estimated survival rate of the customer and the customer survival rate calculated by the survival rate estimation means,
A customer survival rate estimation processing server characterized by comprising: A program for a customer survival rate estimation processing server for causing a computer to function as each means according to claim 3.

Images