TWI790735B - Auditing method - Google Patents

Abstract

An auditing method includes acquiring a structured data set according to a set of auditing indicators at a first point in time, performing data extraction, data cleaning and data transformation on the structured data set to generate a transformed structured data set, performing a cluster analysis on the transformed structured data set according to M clustering parameters to generate a clustering result, and acquiring an abnormal vendor candidate according to the clustering result.

Description

Audit method

The present invention relates to an auditing method, in particular to an auditing method used in a procurement cycle.

Purchasing is the behavior of an enterprise or organization to achieve its goals and obtain the correct quantity and quality of products or services at the correct time and at the lowest price. Procurement audit refers to abnormal transactions or frauds that occur during the inspection or verification of procurement.

Traditionally, procurement audits usually use random spot checks to manually check several purchase transactions, or manually check specific purchase transactions based on news, so as to find out whether there are any abnormal transactions. However, the procurement audit method of manual inspection is slow and cannot inspect all procurement transactions. In addition, the complexity of multi-dimensional data also increases the difficulty of manual inspection. Therefore, the traditional procurement audit is inefficient and difficult to audit problematic transactions.

An embodiment of the present invention provides an audit method, which includes obtaining a corresponding first structured data set according to a set of audit indicators at the first time, performing data extraction, cleaning and conversion on the first structured data to generate a converted first structure The converted first structured data set is grouped according to the M grouping variables to generate a first grouping result, and the first candidate abnormal supplier is obtained according to the first grouping result.

1: Audit system

10: Business database

12: Audit control device

14: Display device

16: Process robot system

200,900: audit method

C1 to C3,36: cluster

70:Dashboard

72 to 76: Sub-dashboard

800: K-average grouping method

S202 to S214, S802 to S810, S902 to S918: steps

Fig. 1 is a block diagram of an auditing system in an embodiment of the present invention.

Figure 2 is a flowchart of an audit method adopted by the audit system in Figure 1.

Figure 3 shows a graph based on the results of the first clustering.

Figure 4 shows another graph based on the results of the first clustering.

Figure 5 shows another graph based on the results of the first clustering.

Figure 6 shows another graph based on the results of the first clustering.

Figure 7 shows a dashboard based on the results of the first clustering.

Fig. 8 is a flow chart of a K-average grouping method adopted in the auditing method in Fig. 2.

Fig. 9 is a flow chart of another auditing method adopted by the auditing system in Fig. 1.

FIG. 1 is a block diagram of an audit system 1 in an embodiment of the present invention. The auditing system 1 is applicable to the auditing process of the procurement cycle in various industries. For example, the auditing system 1 can be applied to the auditing process of suppliers in the telecommunications industry. The audit system 1 can group the relevant data of all suppliers in advance to summarize the types of suppliers, and present the data types of suppliers in the form of charts or dashboards, so that auditors can clearly grasp the characteristics of suppliers and identify them Abnormal suppliers to conduct further audit work on abnormal suppliers. In addition, the audit system 1 can use machine learning (machine learning, ML) to detect changed data types to automatically identify potential abnormal suppliers and warn auditors, and use process robots (robotic process automation, RPA) to collect Unstructured data to save auditor time.

The audit system 1 includes a business database 10 , an audit control device 12 , a display device 14 and a process robot system 16 . The business database 10 is coupled to the audit control device 12 and the process robot system 16, and The audit control device 12 and the process robot system 16 are coupled to the display device 14 .

The business database 10 can be an enterprise resource planning (ERP) software running on a database server. For example, the database server may be a structured query language (SQL) server, and the ERP software may be SAP ERP software or Oracle ERP software. The business database 10 may include a purchase requisition (PR) database, a purchase order (PO) database, an employee database, a vendor database, and other purchasing cycle-related databases. The business database 10 can store procurement data in a structured data format. Structured data can have a fixed format, fixed fields, fixed order, and/or fixed data size. Structured data can be stored as excel files, csv files or other database-accessible files.

The display device 14 can be a display screen. Process robotic system 16 may be a device or tool for performing an automated process. The audit control device 12 can be a computer, and is used to implement the audit method to assist the auditors to identify abnormal suppliers.

FIG. 2 is a flowchart of an auditing method 200 adopted by the auditing system 1 . The audit method 200 includes steps S202 to S214. The steps S202 to S210 are used to group the supplier data to obtain the first candidate abnormal supplier according to the grouping results. The steps S212 to S214 are used to use the process robot system 16 to obtain the first candidate abnormal supplier. unstructured data to determine whether the first candidate abnormal supplier is abnormal. Any reasonable technical changes or step adjustments fall within the scope of the disclosure of the present invention. Steps S202 to S214 are described below: Step S202: the audit control device 12 obtains the corresponding first structured data set according to a set of audit indicators at the first time; Step S204: The audit control device 12 performs data extraction, cleaning and conversion on the first structured data to generate a converted first structured data set; Step S206: The audit control device 12 converts the converted first structured data set according to the M grouping variables. The structured data set is grouped to generate a first grouping result; step S208: the audit control device 12 generates visual data according to the first grouping result; step S210: the display device 14 displays the visual data, and the audit control device 12 obtains the first candidate Abnormal supplier; step S212: the process robot system 16 acquires unstructured data according to the first candidate abnormal supplier; step S214: the display device 14 displays the unstructured data, and the audit control device 12 receives an indication indicating whether the first candidate abnormal supplier Abnormal judgment result.

In step S202, the set of audit indicators may include split bill indicators, repeated payment indicators and other audit indicators. Order splitting is an abnormal transaction in which a single purchase item is divided into several purchase transactions in order to avoid the level of authority (LOA), and each purchase transaction is lower than the audited amount. Duplicate payment refers to the abnormal transaction in which the purchasing agency makes repeated payments to subcontractors for the same payment voucher. The audit control device 12 can obtain the related first structured data set from the purchase requisition database, procurement database, employee database, and vendor database of the business database 10 according to the order dismantling index. For example, the first structured data set may include bidders, bid winners, contractors, number of bids won, bid amount, bid items, bid winning date, latest bid winning date, and the contractor's information on the supplier's procurement case. The number of times undertaken, etc.

In step S206 , the audit control device 12 uses a grouping algorithm to group the converted first structured data set according to M grouping variables, where M is a positive integer greater than 1. The grouping algorithm can be K-means algorithm, which can divide the converted first structured data set into K groups, where K is greater than 1 positive integer. The M grouping variables can include the number of bids won by each supplier (Frequency), the average bid amount of each supplier (Monetary), the latest bid winning date of each supplier (Recency), and the number of times each contractor has handled the supplier's procurement case wait. The clustering algorithm can distinguish suppliers with similar bidding patterns in the multi-dimensional space into the same cluster. Figure 3 shows a graph based on the results of the first clustering. The audit control device 12 divides the converted first structured data set into clusters C1 to C3 according to the clustering variable 2 and the clustering variable 1, M=2, K=3. For example, the grouping variable 2 can be the average winning amount, and the grouping variable 1 can be the number of winning bids. Cluster C1 contains suppliers with low bid winning times and low average winning bid amount. Cluster C2 includes suppliers with high bid winning times and high average winning bid amount. Cluster C3 contains suppliers with a high number of winning bids and a low average winning bid amount. Although Fig. 3 shows an embodiment in which the converted first structured data set is divided into 3 groups according to 2 grouping variables, those skilled in the art can also divide the converted first structured data set according to other numbers of grouping variables according to the same spirit. The data set is divided into other numbers of clusters, for example, the converted first structured data set is divided into 10 clusters according to 4 clustering variables.

In step S208, the visualized data can be chart data or dashboard data. The visualization data can be the image data in Figs. 3 to 7. In step S210, the display device 14 displays the visualized data, as shown in FIGS. 3-7. The diagram of Fig. 3 has been explained in the previous paragraphs and will not be repeated here. Fig. 4 shows another graph drawn according to the results of the first grouping, wherein the horizontal axis may represent the number of winning bids, and the vertical axis may represent the average winning bid amount. Figure 4 shows that cluster 36 has won 8 bids, and the average bid amount is 100,000 yuan. Fig. 5 shows another graph drawn according to the result of the first grouping, wherein the horizontal axis may represent the cluster number, and the vertical axis may represent the number of suppliers. Figure 5 shows that cluster 36 contains nearly 280 suppliers, the largest cluster. FIG. 6 shows another graph drawn according to the results of the first clustering, wherein the horizontal axis can represent cluster numbers, and the vertical axis can represent clustering variables. The grouping variable can be one of the M grouping variables. For example, the clustering variable can be the number of winning bids, and Fig. 6 shows that cluster 0 has the highest number of winning bids. Fig. 7 shows a dashboard 70 based on the results of the first clustering. Dashboard 70 includes sub-dashboards 72 to 76, respectively representing the number of bids won, the average The winning bid amount and the latest bid winning date. Figure 7 shows that the number of winning bids for a particular cluster is 16.58, the average winning bid amount is 8.04, and the latest winning date is within 19.5. Since the characteristics of each cluster can be represented by the readings of the sub-dashboards 72 to 76 , auditors can understand the winning characteristics of suppliers of each cluster through the sub-dashboards 72 to 76 to determine one or more first candidate abnormal suppliers. In some embodiments, the dashboard 70 may also include another sub-dashboard, which presents the number of times of each contractor of a specific cluster, so the auditors can also understand the characteristics of the number of times of contractors of each cluster, so as to determine one or more Multiple first candidate exception suppliers.

In some embodiments, the auditor can see through the dashboard 70 that a certain cluster is characterized as multiple winning deals (for example, more than 10 winning deals) from the same supplier to the same procurement employee or the same procurement department, Therefore, it is determined that the supplier in the cluster is abnormal. In other embodiments, auditors can see through the dashboard 70 that a certain cluster is characterized by the fact that multiple successful bid transactions are concentrated in the same procurement employees, the same traders, and/or the same procurement items, and Most of the amount is close to the upper limit of the internal regulations, so it is determined that it may be a split transaction, and the supplier in this cluster is determined to be abnormal. In other embodiments, auditors can see through the dashboard 70 that a certain cluster is characterized by the fact that multiple successful bid transactions are concentrated in the same purchasing staff and the same dealers, but the purchased items are not from those dealers. The main business item, so it is determined that the supplier in this cluster is abnormal.

Auditors can set the suppliers in the abnormal cluster as the first candidate abnormal suppliers, and conduct audits on the first candidate abnormal suppliers. For example, the auditors can determine 3 abnormal clusters based on the characteristics of the suppliers in each cluster and the characteristics of the number of times each contractor undertakes, and set 20 suppliers in the 3 abnormal clusters as the first candidate abnormal suppliers. Since the audit control device 12 obtains the first abnormal supplier candidate only after grouping all relevant materials in the business database 10, the accuracy and efficiency of the audit will be greatly increased.

In step S212, the unstructured data can be free-form data without fixed format specification, such as including video or audio. Specifically, unstructured data may be purchase requisition attachments, purchase order attachments, image attachments, and/or other attachments in emails. The process robot system 16 will obtain the purchase requisition attachment, purchase order attachment, image attachment and/or other attachments of the relevant email of the first candidate abnormal supplier. In step S214, the display device 14 displays the purchase requisition attachment, purchase order attachment, image attachment and/or other attachments of the first candidate abnormal supplier acquired by the process robot system 16, and the auditor can directly judge the first candidate abnormal supplier based on the attachment Whether the quotient is abnormal, and the judgment result is input into the audit control device 12. Since the process robot system 16 can effectively save the auditor's time in collecting unstructured data, the audit speed will be greatly increased.

The auditing method 200 groups the relevant data of all suppliers in advance to summarize the supplier's type, and presents the supplier's data type in the form of a chart or a dashboard, so that the auditor can grasp the characteristics of the supplier at a glance and identify abnormalities Suppliers, to conduct further audit work on abnormal suppliers, and use the process robot system 16 to collect unstructured data to save the time of auditors, and greatly improve the accuracy and speed of audits.

FIG. 8 is a flow chart of the K-means grouping method 800 adopted in step S206 of the auditing method 200 . The K-means clustering method 800 includes steps S802 to S810 for dividing the N sample points in the transformed first structured data set into K clusters, where N is a positive integer greater than 1. Any reasonable technical changes or step adjustments fall within the scope of the disclosure of the present invention. Steps S802 to S810 are described below: Step S802: Randomly generate K group centers; Step S804: Calculate K Euclidean distances between each sample point and K group centers; Step S806: Group each sample point into a group according to K Euclidean distances, and group the sample points into the group with the smallest Euclidean distance; Step S808: Determine whether the groups of all sample points are no longer changing? If yes, end the method 800; if not, continue to step S810; step S810: calculate the sample mean value according to the samples of each group to update the group center of each group; continue to step S804.

The steps of the K-means clustering method 800 are described below with reference to FIG. 3 . In step S802, the audit control device 12 receives the value of K to randomly generate K cluster centers, and K is the number of clusters, which can be set by the auditor. In Figure 3, K=3, and the number of sample points is 36. Then, the audit control device 12 calculates 36 sample points (grouping variable 1, grouping variable 2) and 3 cluster centers to generate 108 (=36*3) Euclidean distances (step S804), and the 36 samples Point each cluster center assigned the nearest Kilid distance so as to group each sample point into a cluster (step S806). Since it is the first grouping, the audit control device 12 determines that the group of 36 sample points has changed (step S808), and recalculates the group hearts of the three clusters to update the group hearts (step S810), and repeats steps S804 to S806 . Until the audit control device 12 determines that the group of 36 sample points no longer changes (step S808 ).

FIG. 9 is a flowchart of another auditing method 900 adopted by the auditing system 1 . The auditing method 900 includes steps S902 to S918. Steps S902 to S914 are used to group the supplier data to obtain the second candidate abnormal supplier of grouping type change. Steps S916 and S918 are used to use the process robot system 16 to obtain the second candidate abnormal supplier Supplier's unstructured data to determine whether the second candidate abnormal supplier is abnormal. Any reasonable technical changes or step adjustments fall within the scope of the disclosure of the present invention. Steps S902 to S918 are described below: Step S902: The audit control device 12 acquires the corresponding the first structured data set; step S904: the audit control device 12 performs data extraction, cleaning and conversion on the first structured data to generate the converted first structured data set; step S906: the audit control device 12 according to M The grouping variable groups the converted first structured data set to generate the first grouping result; step S908: the audit control device 12 obtains the corresponding second structured data set according to the group audit index at a second time; step S910: audit The control device 12 extracts, cleans and converts the second structured data to generate a converted second structured data set; Step S912: the audit control device 12 converts the converted second structured data set according to the M grouping variables Carry out grouping to generate a second grouping result; Step S914: If the second candidate abnormal supplier is classified into the first type in the first grouping result and is classified into the second type in the second grouping result, then The audit control device 12 acquires the second candidate abnormal supplier; Step S916: the process robot system 16 acquires unstructured data according to the second candidate abnormal supplier; Step S918: the display device 14 displays the unstructured data, and the audit control device 12 receives Indicates the judgment result of whether the second candidate abnormal supplier is abnormal.

Steps S902 to S906 are the same as steps S202 to S206, and the description thereof will not be repeated here.

Steps S908 to S912 are similar to steps S902 to S906, the difference is that the second structured data set and the first structured data set are obtained at the second time and the first time respectively, so the sample points in the second structured data set and The sample points within the first structured dataset are different. second time and first time The time difference between them can be one month, half a year or other periods, that is, the audit control device 12 can periodically generate grouping results.

In step S914, if the type of the supplier changes, the audit control device 12 determines that the supplier is a second candidate abnormal supplier. For example, in Figure 3, if the second candidate abnormal supplier is classified into cluster C1 in the first grouping result (low bid winning times, low average winning bid amount), it is classified into cluster C1 in the second grouping result C2 (high bid winning times, high average bid winning amount), the audit control device 12 determines that the type of the second abnormal supplier candidate has changed, and obtains the second abnormal supplier candidate. In some embodiments, the audit control device 12 can issue a warning message for the second candidate abnormal supplier to be displayed on the display device 14 , and the auditor can check whether there is any abnormality through the dashboard 70 to enhance the accuracy of the audit. In some embodiments, the auditing method 900 can generate visual data according to the first grouping result after step S906, the display device 14 displays the visual data of the first grouping result, and the audit control device 12 corresponds to the visual data of the first grouping result. Get the first candidate abnormal supplier through chemical data. The second candidate exception provider and the first candidate exception provider can be the same or different. In other embodiments, the auditing method 900 can also generate visualized data according to the second grouping result after step S914, the display device 14 displays the visualized data of the second grouping result, and the auditing control device 12 corresponds to the second The visualization data of the grouping results obtains the third candidate abnormal supplier. The second candidate exception provider and the third candidate exception provider may be the same or different.

Steps S916 and S918 are similar to steps S212 and S214, and the description thereof will not be repeated here.

The auditing method 900 regularly groups the relevant data of all suppliers to detect suppliers whose status has changed, warns suppliers whose status has changed to notify auditors to check whether the suppliers whose status has changed is abnormal, and utilizes process robots The system 16 collects non-structural data to save the time of auditors and greatly improve the accuracy and speed of auditing.

The above descriptions are only preferred embodiments of the present invention, and all equivalent changes and modifications made according to the scope of the patent application of the present invention shall fall within the scope of the present invention.

200: Audit method

S202 to S214: Steps

Claims (10)

An auditing method for an auditing system. The auditing system includes a business database and an auditing control device. The auditing method includes: the auditing control device acquires a corresponding first A structured data set; the audit control device performs data extraction, cleaning and conversion on the first structured data to generate a converted first structured data set; the audit control device converts the converted first structured data set according to M grouping variables The first structured data set is grouped to generate a first grouping result, M is a positive integer greater than 1; and the audit control device obtains a first candidate abnormal supplier according to the first grouping result; wherein the audit control device It is a computer. The method as described in claim 1, wherein the audit system further includes a display device, and the audit control device obtains the first candidate abnormal supplier according to the first grouping result includes: the audit control device generates according to the first grouping result Visualized data; the display device displays the visualized data; and the audit control device acquires the first candidate abnormal supplier. The method as described in claim 1, wherein the audit system further includes a display device and a process robot system, and the audit method further includes: the process robot system obtains unstructured data according to the first candidate exception supplier; the display The device displays the unstructured data; and the audit control device receives a judgment result indicating whether the first candidate abnormal supplier is abnormal. The method as described in claim 1, further comprising: the audit control device obtains a corresponding second structured data set according to the set of audit indicators at a second time; the audit control device performs data processing on the second structured data extracting, cleaning and converting to generate a converted second structured data set; the audit control device groups the converted second structured data set according to the M grouping variables to generate a second grouping result; and if a second candidate abnormal supplier is classified into a first type in the first grouping result and is classified into a second type in the second grouping result, the audit control device obtains the first type For two candidate abnormal suppliers, the first type and the second type are different. The method of claim 4, wherein the second candidate exception provider is the same as the first candidate exception provider. The method of claim 4, wherein the second candidate anomaly provider and the first candidate anomaly provider are different. The method as described in claim 4, wherein the auditing system further includes a display device and a process robot system, and the auditing method further includes: the process robot system acquires unstructured data according to the second candidate abnormal supplier; the display The device displays the unstructured data; and the audit control device receives a judgment result indicating whether the second candidate abnormal supplier is abnormal. The method as described in claim 4, further comprising: the audit control device for the second candidate A warning message is issued for an abnormal supplier. The method as described in claim 1, wherein the M grouping variables include the number of bids won by each supplier, the average bid amount of each supplier, the latest bid winning date of each supplier, and the performance of each contractor on the procurement case of each supplier The number of times to undertake. The method as described in claim 1, wherein the audit control device grouping the converted first structured data set according to the M grouping variables to generate the first grouping result includes: the audit control device uses K average ( K-means) algorithm is used to group the converted first structured data set.

Images