How AI Enriches EDI Order Automation [Part 1]

In this two-part series, Esker’s Aurelien Coq discusses how AI allows companies to enrich EDI order automation by adding a functional layer to speed up and facilitate order processing.

AI and EDI: a brief history of time

Electronic Data Interchange(EDI) allows companies to exchange information electronically to replace traditional paper documents. EDI has been designed to automate exchanges such as purchase orders, delivery notes, invoices, etc. through the use of structured and standardised files. EDI is not a recent technology, as it has been widely used since the 1980s to exchange data between companies, directly from machine to machine — rather than from ERP system to ERP system — without human intervention and therefore without having to re-enter data.

Artificial Intelligence (AI) is actually an older concept than EDI: it was born in the 1950s, at the very beginning of computer science. The initial objective of AI was to create machines capable of imitating human behaviour and intelligence. Over the decades, AI has experienced ups and downs with a renewed interest since the 2010’s when major advances were made possible thanks to the combination of three elements: computers with high processing capacity, statistical and mathematical algorithms, and easy access to vast amounts of data. AI already has concrete applications in medicine and transportation, but can also be applied to document processing automation — specifically customer orders.

EDI enables sales order automation. However, AI can also be used in this field.
So, what are the differences and the best practices?

Order automation: EDI or AI?

With EDI, the customer and his supplier have to agree on a data format that will be used to place orders. It is the common language that will be used and understood by the machines on both sides. This common machine language is, in principle, one of the many industry standards (UN/EDIFACT, ANSI ASC X12, GS1 XML, TRADACOM, UBL, etc.) that each offer many variations. However, EDI systems do not have the capabilities and will simply not work even if the format difference is minimal. For EDI communications, trading partners must also define the technical protocol that will be used to transfer the data, often SFTP, AS2 or AS4. Each EDI connection between trading partners is a small IT project involving technical teams to connect the two computer systems, test the document flow and, above all, validate that everything works without friction. Indeed, the very philosophy of EDI is that once the flow has been implemented, you should no longer touch or even see the orders before they are integrated into the supplier’s ERP.

AI allows us to approach the automation problem from another angle: with deep learning technologies, we can train an algorithm with hundreds of thousands of real-world orders to create a neural network that is an expert at capturing data from sales orders. Basically, this amounts to concentrating very specific human intelligence in a small digital brain. This neural network is specialised in a very specific task: finding key information in an order (typically the order number), the desired delivery date, the ordered items, their quantities and prices, etc. … AI can automatically recognise more than 80% of the fields on new customer orders and reduce data entry by users. A second layer of AI-based on machine learning can further improve recognition by memorising user actions performed on unidentified or poorly recognised fields. The machine learns transparently, which progressively increases the automation rates. Of course, it takes a lot of effort — and a lot of data — to develop a platform that uses AI in this way. But when these technologies are directly incorporated into a SaaS order management solution, the benefit is immediate and from day one, the vast majority of the work can be done by the machine, allowing users to focus on more value-added tasks such as customer service.

An important difference between the two approaches is that we are not aiming at perfection every time with AI. The goal is not to have full automation with 100% of the orders processed only by the machines and integrated automatically with the ERP, but rather to have a combination of artificial and human intelligence. For example, 95% of the tasks will be performed by the machine and 5% by the users to manage specific cases, or maybe only visually validate the order and approve it in one click.

EDI and AI: Solving the business and technology equation

EDI is a relevant way of automating orders that are sometimes imposed by customers or legislation, leaving no choice as to whether to use it or not. But recent advances in artificial intelligence now allow businesses to achieve excellent automation rates without having to implement long and costly IT projects. Rather than having to define precisely which EDI format or protocol to use, it is simply a matter of exchanging an order in PDF format by email, which is easy and immediate for both trading partners. Rather than aiming at total EDI automation that will take months and will never be implemented across the entire order flow, AI offers automation rates close to those of EDI for a much lower implementation cost.

Beyond the implementation, I will discuss in Part 2 how AI facilitates the use and maintenance of an order automation solution by providing human intervention flexibility.

-Written by: Aurélien Coq