Invoice Processing Automation with ML Transaction Classification

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional invoice processing systems rely on manual entry of additional information, such as general ledger codes, which increases processing time and reduces accuracy, as they only recognize text data from invoice documents without automatically classifying transactions.

Innovation Solution

A method utilizing a machine learning network to extract text data from invoices, retrieve vendor and recipient characteristics, and predict associated transaction classifiers, such as general ledger codes, to automate the classification process, thereby generating a comprehensive invoice summary document.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If manual entry of additional information is used, then completeness of invoice processing is improved, but processing time increases

Engineering Contradiction:
Improvecompleteness of invoice processingVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system automatically retrieves vendor characteristics data and transaction classifier data, then uses a machine learning network to predict transaction classifiers without requiring manual intervention. The system serves itself by autonomously completing the classification task that would otherwise require human input, thereby maintaining completeness while reducing processing time.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

Vendor characteristics data and transaction classifier data are retrieved and prepared in advance before the actual invoice processing occurs. The machine learning network is pre-trained and ready to predict transaction classifiers. This preliminary preparation eliminates the need for manual data gathering during processing, reducing overall processing time while ensuring all necessary information is available for complete processing.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If manual entry of additional information is used, then accuracy of invoice processing is improved, but processing efficiency decreases

Engineering Contradiction:
Improveaccuracy of invoice processingVSAvoidprocessing efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The manual mechanical process of retrieving data and classifying transactions is replaced with an automated machine learning system. The ML network processes vendor characteristics, recipient characteristics, and transaction details to predict appropriate transaction classifiers automatically. This substitution maintains accuracy through intelligent prediction while dramatically improving processing efficiency by eliminating manual operations.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

A machine learning network acts as an intermediary between the extracted text data and the final invoice summary document. The ML network processes the raw data, applies learned patterns from training, and outputs predicted transaction classifiers. This intermediary layer ensures accurate classification by leveraging learned relationships while automating the process to improve overall efficiency.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Speed

If automatic text recognition is used, then processing speed is improved, but completeness of data extraction decreases

Engineering Contradiction:
Improveprocessing speedVSAvoidcompleteness of data extraction
Core Design Contradiction:
SpeedVSLoss of information

Solution Approach 1:

The system performs multiple functions within a single automated workflow: it extracts text data from invoice images using OCR, retrieves additional vendor and transaction classifier data from databases, uses a machine learning network to predict transaction classifiers, and assembles the complete invoice summary document. This multi-functional approach ensures both speed through automation and completeness by gathering all necessary data types systematically.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The invoice processing system is divided into distinct functional segments: text extraction module, data retrieval module, machine learning prediction module, and document assembly module. Each segment handles a specific task and outputs data for the next segment. This segmentation allows each component to be optimized for its specific function while working together to produce a complete and accurate invoice summary document efficiently.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP4407540A1Invoice processing
Publication Date: 2024.07.31 SAGE GLOBAL SERVICES LTD
  • EP4407540A1 patent drawingFigure 1
  • EP4407540A1 patent drawingFigure 2
  • EP4407540A1 patent drawingFigure 3

AI summary

Disclosed herein is a method of generating an invoice summary document from an invoice received by a recipient from a vendor. The invoice summary document for processing by an invoice processing system. The method comprises: receiving an electronic invoice image file, extracting text data from the electronic invoice image file, processing the extracted text data to identify: vendor identity and transaction details data, using the vendor identity to retrieve vendor characteristics data, inputting to a ML network: the vendor characteristics data; recipient characteristics data, transaction details data, and transaction classifier data associated with transaction classifiers potentially associated with the one or more transactions, outputting from the ML network estimates of which of the transaction classifiers are associated with the one or more transactions, and assembling the invoice document using the transaction details data and the transaction classifiers output from the ML network.