Thinking about Accounts Payable (AP) may not make your innovation heart skip a beat. Yet according to Basware, you would do well to take a good look at it. What does the company mean by this? We talked to a representative from the company to find out.
Accounts Payable will not currently be part of the core processes for many organizations. Processing invoices is usually seen as a necessary evil and an expense. This is not justified, Basware argues. Accounts Payable also provides a lot of data, if you organize it properly. This data, in turn, contains a lot of useful information. Not only can it help your organization operate more efficiently, it can also provide other interesting insights. Especially if you use a certain degree of ML and AI for this. In a conversation with Kris Vanhoutte, pre-sales consultant at Basware, we take a closer look at this.
The AP flow
The Basware platform deals with the entire journey within the source to pay world. It first selects the right supplier when placing the order. It then handles receiving the goods or services and finally processes the invoice. Vanhoutte calls this journey the AP flow. It consists first of all of receiving the invoice. This can take place on paper, something that is (fortunately) becoming less common. However, it can also be done via PDF, or even as an e-invoice. In the latter case, the invoices run from the supplier’s ERP system through networks such as Basware’s to the customer’s Accounts Payable system.
Merely ‘translating’ the invoices into data and an image isn’t enough for most organizations, though. Usually, the this output goes into the internal AP process. That’s where the real work begins for Basware, Vanhoutte points out. The company has developed a SaaS solution to get to work with the data from the invoices. After all, not every invoice needs to be treated in an identical way. Some have a PO, some don’t, some invoices are the same every month. The idea is to automate the processing of invoices as much as possible. That way you relieve AP staff of repetitive tasks. They can concentrate on more important tasks.
When the internal AP process is complete, we arrive at the third step in the AP flow. This is the final stage. Here the end result is generally linked to the ERP system or accounting.
Move away from manual filing and OCR with AI/ML
We all probably know the image of the trays on the desk of people dealing with invoices. Incoming, outgoing, paid, follow up, and so on. In the past, you had to do everything manually. This started with opening the envelopes that contained the often paper invoices. Then there was the manual keying of the data into a system, after which the entire internal AP process also continued manually. Automation was necessary to improve this.
One of the ways to eliminate the need to manually copy information from invoices is OCR. The standard version of OCR, however, is not very good, Vanhoutte says. Basware has made improvements, but it’s still not very good. This is why Basware has developed SmartPDF. After all, a great many invoices come in PDF form. 80 percent of them are machine readable. You don’t need to use OCR for that. Here too, however, they see that it is not obvious that everything is extracted properly. The layout of invoices, for example, is often ERP-related. That is why AI/ML is used for this. It checks what the invoice actually says using various variables. It then automatically transfers the correct data to the next step in the AP process.
From ML to AI
When we ask to what extent the above is AI, Vanhoutte immediately admits that within AP they are still mainly using ML. Among other things, unforeseen additional costs (transport, packaging, etc.) that appear on the invoice are automatically detected and processed. This is primarily a rule-based exercise. You could almost say it’s more like RPA than Machine Learning.
If you want to move further towards AI, you will have to largely abandon the rule-based approach and train the models to work smarter. As an example, Vanhoutte cites keeping a spend plan here. In general, these are very useful as they allow you to control invoices that are not related to orders, but still process them automatically. The crux is to recognize those types of invoices and preferably create such a spend plan for them immediately. Basware can provide this by recognizing patterns in historical invoice data.
With a rule-based ML approach, you can make a fine start with automated clear categorization of invoices. However, it would be much stronger if an AI could decide this on its own, without predefined rules, Vanhoutte points out. That’s what Basware is fully committed to at the moment, he says. The idea is that an AI will give suggestions for the coding of invoices, with a specific degree of certainty. This should create a lot more clarity in large organizations with many incoming invoices. In addition, it frees up employees to deal with other matters.
The training of the ML models that Basware is already deploying takes place on the basis of the data of the customer where the model has been deployed. Basware chose this approach because of the uniqueness of the customer data in terms of AP. There are, however, ideas to extend this and to train models on more general Basware data, he indicates when asked. Especially for something like categorization, it can certainly be interesting. Consider, for example, an organization that has standardized on Dell equipment, but in which someone nevertheless purchased a Mac. The accompanying invoice from Apple can be automatically recognized and processed as an IT invoice through that categorization, which gives the IT budget manager a complete overview of all costs automatically, without manual intervention.
The initial effect of automating processes like Accounts Payable is, of course, that you have to do much less manually. That in itself is an advantage, because you spend much less time on it. Time you can put into other efforts elsewhere in your organization.
According to Vanhoutte, however, you can do even more with the data you collect through this automation. This data gathered through ML/AI also gives you other insights, he says. It allows you to detect improvements in the process as a whole. If an automated process keeps getting stuck at the same point, it is a sign that something can and should be improved. In addition, you can now use the collected data to provide higher management with much easier insights. Think of financial reports, which you can now generate with a push of a button.
There is still a certain amount of human interaction in the above insights. However, Vanhoutte indicates that Basware will support these decisions even more in a next step. For example by proposing possible optimizations based on these insights. Their customers already have a relatively high level of automation in invoice processing, he says. With Basware’s solutions, thanks to ML and increasingly AI, you should be able to organize this even more efficiently, without losing control. In short, Basware still has plenty to work on within their already well-regarded solutions. We will therefore continue to follow the developments closely.