This post originally appeared on the freedz.io blog.

The invoice, a central document often neglected

Imagine… You are in 2020 in the finance department of a company receiving more than 50,000 invoices per year. All or almost all of these invoices are received in paper format, when they are received at all… Because of this, in the accounting department, a team of 4 people is dedicated exclusively to scanning the invoices and manually inputting data into the company’s ERP system.

Hard to believe? Especially when you know that the cost of processing a paper invoice received from a supplier is 15€. And yet, in 2020, more than 80% of invoice exchanges are still made on paper!

The path of a paper supplier invoice is often chaotic. This results in high costs for companies, time-consuming tasks for employees and regular conflicts with suppliers .

Freedz : a path towards 100% of e-invoicing

Freedz platform was created to address this problem by making customer-supplier relations more fluid. The solution allows a company to collect all of its supplier invoices in a natively electronic format. No matter how suppliers operate, Freedz offers a solution to collect their invoices in a format that can be directly integrated into the ERP system. Here, suppliers are responsible for their invoice data and the customer can automate the processing of invoices as soon as they are received.

“In order to offer an interesting and fast ROI to our users, we first chose to offer solutions that would allow them to receive e-invoices from their suppliers sending large volumes,” says Marc Vidal, Freedz Product Manager at Neovacom. This is why the Freedz platform concentrates different automated reception channels (EDI, API) and numerous e-invoicing formats. In some cases such as the long tail of low volume suppliers, the development of an automated connection is not interesting for the supplier nor economically viable for them. This is why Freedz has the capacity to receive invoices in PDF format. “But in order for Freedz to build the electronic format that can be automatically integrated by the customer, the supplier had to manually re-enter some information,” explains Marc. “So it could take 30 seconds to a minute to submit an invoice. When we noticed that some suppliers were using Freedz to enter more than 100 invoices/month, we realized that we needed to improve the experience for these users! “Neovacom started looking for a solution to automate the reading of the information contained in the invoices. This would make it possible to pre-populate the fields that were entered manually by the suppliers.

With Mindee, Freedz’s features capitalize on new AI capabilities

“Freedz was developed in SaaS with the latest generation of technology and architecture. It was therefore important for us to call upon an innovative company from the new French AI scene. “That’s where Mindee comes in.

“Mindee caught our attention because several players we knew well were already using their technology. It was their approach that appealed to us. Based on deep learning technologies, their invoice reading API is radically different from what exists on the market. The performance of their AI in terms of data reading is impressive. It’s real time with feedback in the second and a success rate of 95%! This greatly enhances the vendor experience on our platform. “Thanks to the addition of this technology, it only takes a few seconds to submit an invoice on Freedz!

How does it work?

So how do we achieve such performance? To overcome the limitations of historical technologies, Mindee’s deep learning researchers have been working on algorithms that seek to get closer to human behavior. “What we’re doing is computer vision through deep learning algorithms, not OCR. Believe it or not, traditional OCR solutions read every word on an invoice before looking for words that might match this or that information,” said Victor Briançon-Marjollet, COO at Mindee. “What we’ve been trying to do is to get closer to human behavior. When you read an invoice, you instinctively turn your attention to different areas depending on whether you want to read the amount including tax or the name of the supplier, for example. You draw from visual cues such as layout, font, colour and word arrangement the clues you need to unconsciously perform this task. This is what our AI also does. It will search for areas of interest where the desired information is likely to be found and then read it. Changing the comprehension paradigm of a document by reversing the steps of reading and locating information has allowed us to achieve unparalleled performances in terms of speed and accuracy. »

In addition, this technology is based on a neural network, which has certain other advantages. “Once a certain number of examples (40-50,000 invoices) have been passed on, the AI “generalizes”, i.e. builds its own “rules” that allow it to adapt itself to new cases. This is one of the great strengths of these neural networks: once in production, they also work on documents whose supplier, currency or even language are unknown to it. With traditional technologies, each new case required a new OCR setting to operate with the same performance, which is extremely cumbersome”.

Instead, Mindee sought to provide a technology that was accessible and simple to integrate for developers wishing to equip their platform with it. The integration of Mindee with Freedz was very quick because both our solutions were designed to be SaaS, open and with a variety of APIs,” said Marc. What’s more, their teams proved to be super reactive. »

An evolution that calls for more

“We are very pleased with this collaboration with Mindee,” says Dominique Fenain, President of Neovacom. “This is due to the excellent relationship with Mindee’s teams, but above all to the fact that their technology is an additional functional asset that they bring to our Freedz platform. The addition of this AI to our invoice submission portal improves the user experience for suppliers. It is a development that was also expected by our customers to make life easier for their suppliers. Because the whole purpose of Freedz is there, to improve the relationship between the two parties by making the exchange of information more fluid. Other developments will happen very soon! »

For his part, Jonathan Grandperrin, CEO of Mindee welcomes the alignment of visions between the two companies. “Our goal is to provide all developers who want to build software of tomorrow with powerful, robust document reading capabilities that can be easily integrated into their products. Freedz shares this vision of an open ecosystem in which software vendors and technology providers work together through standardized APIs to make the most of each other’s expertise. We have taken the time to capitalize on our previous products, particularly those dedicated to expense receipts, in order to provide an invoice reading engine that breaks the standards set by 30 years of OCR. We are delighted that a publisher like Freedz can show the full extent of what can be built by integrating our engines.