System and method for repeatable and interpretable divisive analysis

A computer system, comprising:

• a plurality of personal computer systems of customers and suppliers that are communicatively coupled to a data communication network;
• one or more enterprise resource planning (ERP) computer systems that are communicatively coupled to the data communication network;
• one or more third-party systems that are communicatively coupled to the data communication network;
• a spend management cloud service (SMCS) that is communicatively coupled to the data communication network and comprising one or more processors and one or more non-transitory computer-readable storage media coupled to the one or more processors and storing one or more sequences of instructions which, when executed using the one or more processors, cause the one or more processors to perform:
• uploading a plurality of CSV files of spend data from the one or more of the ERP systems and the one or more third party systems to the SMCS via a secure file transfer protocol server, the plurality of CSV files comprising one or more of master data and transactional data;
• causing web pages with particular content to be displayed by client applications at the plurality of personal computer systems of the customers and suppliers using an online interaction protocol;
• identifying, using a spend behavior cohort identification system, top-level spend behavior cohorts of spend transactions based on one or more spend behavior characteristics of spend transactions, wherein the spend behavior cohort identification system executes using one or more computer systems;
• recursively or iteratively dividing, using a divisive analysis system, a selected top-level spend behavior cohort resulting in a plurality of sub-level spend behavior cohorts arranged in a hierarchy, wherein one of the plurality of sub-level spend behavior cohorts comprises all spend transactions of the selected top-level spend behavior cohort having a selected categorical attribute value and does not comprise any spend transactions not having the selected categorical attribute value, and another of the plurality of sub-level spend behavior cohorts comprises all spend transactions of the top-level spend behavior cohort not having the selected categorical attribute value and does not comprise any spend transactions of the spend behavior cohort having the selected categorical attribute value, and comprising dividing the selected top-level spend behavior cohort according to a set of categorical attributes by determining the corresponding selected categorical attribute value that minimizes spend variance of spend transactions or maximizes spend amount homogeneity of the plurality of sub-level spend behavior cohorts with respect to the selected top-level spend behavior cohort;
• identifying, using an automatic spend risk identification system, a leaf spend behavior cohort of the plurality of sub-level spend behavior cohorts, arranged in the hierarchy, in response to determining the minimized spend variance of the spend transactions or the maximized spend amount homogeneity of the plurality of sub-level spend behavior cohorts with respect to the selected top-level spend behavior cohort;
• detecting, using the automatic spend risk identification system, a particular spend transaction that is a statistical outlier among spend transactions of the leaf spend behavior cohort of the plurality of sub-level spend behavior cohorts arranged in the hierarchy, wherein the automatic spend risk identification system executes using one or more computer systems;
• and
• causing a particular web page to be displayed by client applications at the personal computers of a particular customer using the online interaction protocol, the particular web page comprising a graphical table of risky spend transactions comprising at least one row identifying the particular expense spend transaction and comprising at least a textual characterization of why the particular spend transaction is risky.