Business Decisions to Consider

Before starting any configuration or setup of the Spend Classification module, business decisions will need to be taken to assess how organization spend should be categorized.

This may require the introduction of a brand new category structure or the review and possible update of any existing set of category definitions used for processing spend transactions within your applications.

If you don't have any spend categories defined, you should review publicly available category definitions, such as UNSPSC, as the basis to develop your own category set.

If you have implemented the Oracle Fusion Cloud Procurement solution, you will already have a set of Purchasing categories configured. These can be used as-is within Spend Classification or you may opt to revise the category definitions, adding and deleting categories to meet your spend analysis needs.

It's recommended that an extensive consultation exercise be conducted with all the key stakeholders throughout the organization who have information on spend analysis activities. These existing stakeholders will need to be consulted to uncover any limitations or gaps in the current spend analysis results. Based upon their feedback, a draft proposal for the taxonomy to be used in Spend Classification can be circulated for review. If required, more than one taxonomy can be proposed.

Reviewing and refining the classification taxonomy may take a number of weeks to complete, depending on how many parts of the organization need to be included in the consultation.

For any revised or new taxonomy, it may not be practical to target thousands of category definitions. Managers within your organization may need to compromise their need for fine-grained insight of spending activity to avoid huge setup and ongoing administrative costs for the application.

After the taxonomy needs have been agreed with the business, set up in the application is a straightforward process supported through a spreadsheet.

Training Data Set

Spend Classification uses machine learning technology to process spend transactions and generate category predictions. The machine learning algorithms require training data to teach the engine how to identify similarities and patterns in the spend data that enable the tool to recognize and classify each transaction.

Training set data will need to be created for each category to be classified, with enough examples provided to illustrate the various types of spend that should be assigned to a category. At a minimum, thirty to forty sample spend records will be required for each spend category in order for the prediction engine to work properly.

The time and resource cost to build training set data should definitely be considered as a factor when establishing the set of spend categories that will be used to generate predictions in Spend Classification. It should be expected that a balance will need to be struck between the desired category granularity and the cost to craft the training data. It is certainly a consideration during an initial implementation to reduce the number of categories, as the tool will allow these to be iterated to greater detail over time.