Increase organizational productivity through automated invoice processing
By LexRock AI | Published on February 17, 2022
Problem and objectives
Problem
- LexRock AI had to find a way to reduce invoice processing time for the firm Objectif Lune, an organization that provides services to over 25,000 customers in more than 80 countries. This task usually takes two to eight hours to complete manually.
Objectives
- Automate the invoice standardization process;
- Decrease invoice processing time;
- Enable organizations such as Objectif Lune to increase their organizational growth.
Proposed solution
Solution
LexRock AI is a Montréal-based artificial intelligence (AI) company specializing in scanning, editing, analyzing and extracting information from documents. It was given the mandate to help Objectif Lune, a company specialized in transactional documents, automate the invoice standardization process for its customers.
Principle
LexRock AI created a language comprehension technology that reads the structure of an invoice and automates the recognition of its contents. The algorithm that was developed reads images, text, tables, and any other type of information that an invoice may contain. The company then created a programming interface that could be included in Objectif Lune’s system, thus making it easier to use. The model is capable of recognizing information in all the company’s available languages: English, French, German, Dutch, Italian, Portuguese and Spanish.
Role of AI
AI was key to the project’s implementation. Using an artificial intelligence solution enabled them to establish a growth strategy through a significant reduction in the time allotted for each document.
The techniques used by LexRock AI consisted of:
- Natural language processing, including the following techniques: LSTM, BI-LSTM, RNN, conditional random fields, transformers (e.g. BERT, GPT-2, GPT-3, RoBERTa, FlauBERT, DistillBERT), K-means algorithm; . Vision in the context of text detection: FCN, Fast-RCNN and DGCNN;
- Conversational models/algorithms: Dual intent-entity classifier, Conversations, ResponseSelector.
Impacts
Main outcomes
Main benefit
Invoice processing time decreased from 2-8 hours to only 5 minutes.
Impacts on Objectif Lune
The algorithm created by LexRock AI now supports Objectif Lune’s long-term development strategy. The time savings are significant and allow the company to seek out new customers while strengthening its niche position in the market. Being accustomed to working with large companies that manufacture printing equipment, Objectif Lune can now offer these organizations a digital transformation solution. The organization now uses the tool to map data internally.
Impacts sur LexRock AI
Le projet a permis à LexRock AI de perfectionner certaines techniques en IA, telle que les algorithmes de pairage. Le résultat de cette collaboration fut une augmentation de la maturité technologique de l’entreprise et une ouverture vers un nouveau domaine niche, soit celui de la vision en IA. L’entreprise est maintenant en mesure d’appliquer son expertise en reconnaissance d’image dans le cadre de documents non structurés.
Generation and dissemination of new knowledge
Collaboration with research institutes such as the Institut de valorisation des données (IVADO).
Challenges takled
One of the major challenges of the project was to locate the different parts of the documents to be analyzed. LexRock Ai developed a tool capable of identifying the different parts of a PDF document. Data cleansing had to be done. The many formats of the information contained in the invoices such as text, images and tables meant that LexRock AI had to use multiple deep learning algorithms.
Data privacy quickly became an important ethical issue in the development of the project. Personal information, such as client names and addresses, was anonymized to maintain client confidentiality.
Conditions for success
Team mobilised
Internal resources
- Automatic natural language processing (NLP) specialists;
- Software developers;
- SaaS specialists.
External resources
- Objectif Lune’s data mapping team;
- President of Objectif Lune’s French subsidiary;
- Researchers from the Institut de valorisation des données (IVADO).
Funding received
Funding received from Prompt AI
Project stages
Objectives and strategy drawn up between the two companies;
Definition of algorithms to be used for the different types of available invoices;
Creation of data mapping technology.
Project stages - to come
The next step will be using an SaaS solution that LexRock AI will provide in the next phases of the project.
Context of the solution
Customers
LexRock AI’s products and services are intended for companies that manually enter data into documents. The companies in their target market are mainly financial, legal and insurance companies.