Artificial intelligence and GIT.
Artificial intelligence has been an integral part of the solutions developed by GIT SAfor nearly eight years.
Already integrated into theIZICLICK software, it was then referred to asmachine learning. Today, the Geneva-based accounting software publisher wants to take it one step further.
PEPPERMINT and smart bill reading.
With this in mind, GIT SA has launched a research project in collaboration with the ICARE Research Institute in Sierre, with the support of the Innosuisse agency. The goal is ambitious: to enable the PEPPERMINT software to autonomously identify all relevant information on an invoice, capture it, transform it into accounting entries, and continuously improve itself through accumulated experience.
Invoices come in many different forms: paper, PDF, electronic format, or QR invoice. Their layout and content vary greatly, while the accounting system always requires the same key information: date, issuer, amount, VAT number, among others. Some data on an invoice, such as the number of international calls on a telecommunications invoice, is of no interest to accounting, unlike their total cost.
In the past, extracting this relevant information relied entirely on human intervention. Today, this work is increasingly automated. An invoice is scanned or imported into PEPPERMINT, which extracts the necessary data, fills in the corresponding fields, and automatically generates an accounting entry. However, the user still has a role in checking and validating the information. The wide variety of invoice formats can nevertheless complicate the automatic recognition of certain data, still requiring occasional human intervention to indicate their location.
This is precisely where artificial intelligence provides decisive added value. Each piece of data that is validated, corrected, or entered manually is integrated by the software through machine learning. This means that when a similar invoice is processed later, the system knows where to look for the missing information.
The research project aims to further improve data recognition rates. It draws on a combination of skills: in-depth knowledge of the software and cutting-edge expertise in accounting—provided by GIT SA—and specialized skills in artificial intelligence, provided by the ICARE Research Institute.
PEPPERMINT and confidentiality.
The main challenge of the project lies in ensuring data confidentiality. Each user contributes indirectly to training the software, simply by checking and validating the information entered. These lessons must be shared in order to improve the product, without transmitting the associated sensitive data. The project, which involves around ten people, should be available in early 2026.
