One option was to develop a self-learning product matching engine that would necessarily have to be highly adaptive, flexible and geared toward self-improvement due to the complex requirements involved:
- Combinations of information from different systems (product ID, material information, order history, etc.)
- Use of customer-specific terminology
- Identification of synonyms, acronyms, ambiguous terms, etc.
- Ability to understand the relative significance of potentially important words in certain positions
The solution had to make the process more efficient and save time without adding to its complexity. This is why ONE LOGIC opted to take a different approach: rather than raw data, it used tables from the ERP system (with PO numbers, PO items, etc.) as its basis. These historical orders had already been integrated into the system and could be used as a training data set.
Each order contains open text fields, fields with predefined products and/or information and categories with special commodity group codes that have to be completed. If the entries are not able to be allocated to a ready-made category, the Data Product Platform ONE DATA breaks down the order text into single words, then uses statistical methods, mapping, collaborative filters, etc. to determine their possible meaning, and calculates the likelihood that they refer to a certain commodity group, category, item number, etc. In the next step, this information is sent back to the department that oversees incoming orders for a final (manual) check.
By bringing ONE DATA into play and breaking down orders into single words, it was possible to automate a time-consuming manual process and convert paper orders directly into entries in the system. The process is quicker, more efficient and more secure, especially as the customer can check the order again at the end without having to spend a great deal of time on it.