Line Item: Extracted Automatically

What do all these positions have in common?

All of these business documents are about information about the Purchase2Pay (procurement view) or Order2Cash (sales view) process. To be more precise, additional information related to the purchase of a single product. The logical grouping of the individual items is driven by administrative aspects (e.g. time interval or delivery).

What are the typical use cases behind line item recording?

For this reason, e.g. a 3-way comparison (order, invoice, delivery) for the automated release of invoices and their dark posting is often a complex task. In addition to the 3-way comparison, a wide range of customer inquiries are an issue when recording: split bookings with tax consultants, automated account assignment, approval of cost estimates, automatic approval of offers, automated acceptance of orders, price comparison with central regulators, automatic order creation, damage case processing from insurance companies, prediction of price developments in construction through to the extended acquisition of information about products (derivation of ingredients in the food sector).

What does a good line item capture need?

Capturing a well-structured table and its individual rows is not rocket science. However, line items are very often not well structured or represent a complex group structure. Information about products can vary greatly depending on the industry, descriptions of products and their order or invoice information often have multiple lines or cannot be clearly assigned to columns. For this reason, a technology is needed that can recognize optical structures, masters a specific vocabulary and is capable of learning in both areas. With the combination of NLP, deep learning and special AI architectures, the automation of position data becomes manageable.

In addition, it requires data or examples. So the combination of a document and the structured data to be extracted from which the AI ​​can learn. Most AIs are ready2use. This means that they are usually pre-trained and have already learned many line item structures in advance and can also grasp them out-of-the-box. If the recording quality is not sufficient, this can be improved through data feedback and a human-in-the-loop approach.

With the BLU DELTA AI, you have the option of having line items or individual items extracted from your documents on the one hand, and you can also send us direct data feedback on the other. In this way, you can record corrections in your software in a non-transparent and automated manner and send them to the BLU DELTA AI. As a result, it regularly learns in the background and automatically makes an improved AI available again. BLU DELTA follows a shared intelligence concept. This means that the AI ​​not only learns from your corrections, but from the corrections of all BLU DELTA customers!

Thanks to our learning position data acquisition, many customers of an ERP manufacturer were able to massively reduce the manual entry of position data. All previous test customers then decided to migrate to BLU DELTA AI.

The acquisition of position data is also a challenge for you and you would like to automate this? Then let's take a look at your problem together and we'll tell you whether we can help you with our solution.