Processing invoices is a process that every company has. If you have a lot of suppliers and a lot of invoices, the costs of such a process can be quite high. A major retailer asked us to take a critical look at the invoicing process and come up with an RPA solution. This is the story of a few weeks of work that saves the customer 4,000 working hours (and 100 trees) each year.
Companies still have many questions about RPA. People have all kinds of associations with 'robotics', which have absolutely nothing to do with RPA. You can talk about it endlessly, but showing it works better. Only when people really see a robot clicking through screens, they see what a software robot does and how different RPA is from 'classical coding' and integrating systems into the backend. By quickly showing tangible results, you create enthusiasm. And then you see that an organization will automatically start thinking about the possibilities. All kinds of applications emerge where RPA can add even more value.
The way to do this is by executing a Proof of Concept (PoC). A 'pilot project' of four to six weeks. This shows in practice what RPA is, what you can do with RPA, what it delivers for the organization and what possibilities there are to automate even further.
The customer I'm talking about already knew exactly which process we had to tackle: manually processing 600,000 supplier invoices per year. Employees first collected these from e-mail, then printed them and incorporated them manually into the invoice system, after which they were stored and kept in storage for seven years. That was a thorn in the customer's side, and rightly so. The question they were faced with was: "How do we get the right data from all those different invoices? We found the answer in UiPath's machine learning applications.
We divided the process into three functional steps, which we automated separately: retrieving data, classifying the invoice and securing the data. A first software robot retrieves the invoices from the mailbox and passes them on to an application with artificial intelligence. It analyses them and retrieves all important data. These are then entered into the invoice system by a second robot. In this way we replace the labour-intensive, error-prone and 'tedious' manual process.
In this project we will not only use RPA, but also Artificial Intelligence (AI). That sounds much more complicated than it is, because this technology is nowadays available as Software as a Service (SaaS). So you don't have to develop or install it yourself. Recognizing invoices is also a problem that is more or less the same for all companies. That's why you don't have to train the model yourself. Previous users of this software have already done that for you. So you can start processing invoices right away.
I found it a nice challenge to look for the right solution for this client. The offering is very large and so are the differences in price and functionality. So I made an extensive analysis: how good are models? Are they good for retraining? How does the application integrate with existing systems? Can we build in human validation in a user-friendly way? This was a very informative process for me.
For this PoC, we ended up using UiPath document understanding in combination with UiPath software robots for invoice processing. But we may end up making other choices for the production environment.
The goal at RPA is usually to let processes run completely autonomously ('unattended'). This is not always practically feasible. In this project we therefore combined attended and unattended applications of RPA and AI. For example, different suppliers use different terms and numbers: order number, invoice number, payment reference and debtor number. If the AI is not sure what the meaning of a number is, the robot asks people to check it. The human feedback then trains the algorithm again, increasing the confidence levels of the AI and reducing the need for human intervention. This is important, because we use RPA to automate recurring, tedious tasks. We don't want employees to have to enter the invoice numbers of a particular supplier manually all the time because the robot doesn't recognize the invoices. To give the algorithm room to learn for itself, we did not create any invoice templates and did not write any code. The robots that read the e-mail and enter the data into the systems immediately worked completely autonomously.
An important goal in this project was to have zero impact on IT. Or rather: zero impact of IT on us. Of course, we also work with a backlog and a good development and testing process in RPA projects, but we can implement changes much faster than 'traditional' IT teams, who are often hampered by system boundaries and technical hurdles when adding value. At RPA, we want to work 'end to end' on the entire process. And we want to move forward quickly, without worrying about dependencies. Of course, we kept in touch with IT, but no effort was required from them other than making a computer with a remote desktop available. That was enough to run our robots.
With AI Computer Vision, an UiPath product, we were able to very easily set up communication with desktop apps from that workstation. So we didn't need any backend links or data connections. UiPath's low code development capabilities allowed us to get this up and running quickly.
Meanwhile, the development of the project has been completed and the entire process is ready to go into production. Tests have shown that the robot is already able to process 350,000 invoices per year, which represents a saving of 4,000 working hours. That was the basis of the business case and the revenues far outweigh the costs.
However, once the robot is in production, we expect much more profit. Take, for example, the printing of the invoices. That is no longer necessary now. That not only saves a lot of money, but also about 100 trees a year. And how about data storage and storing all that paper for 7 years, with all the susceptibility to errors, space and peripheral issues that come with it?
Employees can spend the saved time with solving invoice issues. Once they've figured out what's wrong with a particular invoice, the company can change its own process or contact the supplier to ask for adjustments. This work is now often left lying around.
Furthermore, data on all invoices is stored in a form that is easy to analyze. This provides a deep insight into operating costs, supplier relationships and cash flows within the company. This makes it easier to precisely manage and steer the company.
But actually, that's not the purpose of a PoC. The goal of a PoC is to generate enthusiasm and to lay a foundation for further automation. Well, it has succeeded. The results of the project were received enthusiastically at all levels. The management was of course happy with the quick result, the lower costs and increased control. On the work floor, people were especially happy with the elimination of boring, repetitive work and getting more time to work on improving the processes. And, of course, we are now going to look further into the possibilities of RPA and AI with this customer.
This company knew exactly what it wanted to do with RPA and why. Most companies haven't gotten that far yet. Customers often come to us with the idea of 'doing something with RPA'. We then get to work and, together with the customer, map out all the processes (of a department). Then we make a 'heatmap', on which we classify all processes according to complexity and potential profit. We are of course looking for the process with the lowest complexity and the highest profit. We choose that process for our PoC.
When you ask customers about their goals, you often hear phrases like "we want to grow from 2 million turnover to 3 million" or "we want to become 15% more efficient". I see RPA as a way to translate these abstract goals into the reality of a company. A PoC with intelligent process automation works very concretely with one process and gets everything out of it. You immediately see the impact on your business goals afterwards.
And once you get started, you see more and more possibilities. By continuing to analyze your processes and meanwhile automating your processes step by step, you discover more and more use cases for RPA. By applying RPA, you also collect more and more data, which gives you insight into further applications and improvements. This 'snowball effect' can make that first PoC project the start of the total transformation of your company.