Everywhere you turn, someone talks about Artificial Intelligence (AI) and Machine Learning. AI driven software is already changing business processes in small and large enterprises. Artificial Intelligence revolutionizes “manual” software workflows and leads to new customer experiences. In this article, we explore the use of AI and machine learning for document processing based on the example invoice processing.

Invoice processing

All companies around the world must process invoices. Invoices are what make the business world go round. Every year, companies issue over 18 billion invoices in Europe and America.

Today, invoice handling processes are time consuming and cumbersome. The European Commission estimates, that 80% to 90% of invoices (2017) are paper-based. In charge of the invoice processing are accounting departments and accounting service providers. And invoice processing is done manually or semiautomatically.

Let’s see how artificial intelligence and machine learning are revolutionizing this last, non-digitalized industry.

AI and Machine Learning for Invoice Processing

Business software vendors use Artificial Intelligence for invoice processing in several ways. AI based software supports:

  • Invoice Separation
  • Data Extraction
  • Eata Completion

Automated Invoice Separation

For paper invoices, digitalization starts with scanning. As a result, the scanner produces one document (most likely a pdf or TIFF). This document contains hundreds of pages and invoices. For further processing, users must seperate the documents manually for each individual invoice. This is time-consuming work and errors during the separation cause disruption within the proceeding process steps.

The machine-learning software Abacus Intelligence®receives one file with hundreds pages and invoices. Based on historical data, the AI realizes the start and the end of an invoice and creates for each invoice a separate file. There is no need to use barcode stickers or separate it manually.

Speaking in AI terms, the separation of invoices is a “classification” issue. You receive a document containing of multiple of invoices. You have to recognize where one invoice starts and ends. Abacus Intelligence® solves this problem the same way as human do. Abacus Intelligence® recognizes attributes at headerlevel at the first and on footerlevel on the last page for every document (supplier). This information is used to separate the scanned document in individual invoices and files.

Precise Invoice Data Extraction

For scanned paper invoices and for low value E-invoices data extraction is necessary. Old-fashioned OCR software extracts invoice information in an inefficient manner. Before you process an invoice for the first time, you must create a template manually. A template tells the software “where” to extract “what” data. You need to do this for every supplier and invoice layout. If a supplier changes the layout, you need to redo the template. OCR templating software is complex: Only well-trained employees can create good templates and template creation can be complex: The duration for the creation of a single templates varies from 10 Minutes to 1 hour. In sharp contrast, AI based software extracts information without the creation of templates. Abacus Intelligence® receives the OCR data and extracts the relevant invoice information:

  • Invoice number
  • Gross amount
  • Net amount
  • Tax percentage
  • Purchase order number

For each supplier invoice, the self-learning software creates a template-model automatically. This substitutes the manual model creation. Employees do not set up or optimize templates, the software does it in one second. With Abacus Intelligence®, you save template set-up time and maintenance costs.

Optical Character Recognition (OCR) creates for each invoice a document. Within a OCR document, it is hard to find the location, where the relevant data stored. To achieve this, training data for each individual invoice (suppliers) is required. Abacus Intelligence® connects to any accounting system and runs a reverse engineering. AI uses historical data such as company, chart of accounts, supplier and invoice layout to create a specific AI-model. There is no need for manual interaction, the model creation and adaption process is initiated automatically.

AI and Machine Learning for Data Completion

The ultimate goal of digitalization is to automate the full invoice processing workflow. Seperation and data extraction are the first and second step. Even if you receive an E-invoice, consiting all invoice data in the XML-structure, every company needs to “complete” the invoice with the appropriate accounting attributes. To date, accountants use their accounting knowledge to assign this attributes manually or search for it. Here, new technologies emerged to automate this process using artificial intelligence. Abacus Intelligence®completes each invoice with the following information:

  • General ledger
  • Cost center
  • Tax codes

Now the accountant use accounting knowledge to carry out checks, especially to check soft criteria and circumstances. In addition Abacus Intelligence®highlights any deviations form the past automatically for manual transaction. Speaking in AI terms, this is a “classification” exercise. Data from journals and historical invoices are used to train classifiers, individually for each business partner.

Concluding Remarks – Invoice Processing based on AI

To create AI models for separation, capturing, completion, Abacus Intelligence® connects with accounting software and trains models. The self-learning AI-software processes per day for thousands of companies for accounting service providers, shared service centers, enterprises and software vendors. The software is a state of the art web application and runs either as SaaS or on premise. Software vendors can integrate Abacus Intelligence using a comprehensive RESTful API.

Abacus Intelligence® has already changed the business of accounting service providers and shared service centers.