by Richard Harris,Content Director for Works Applications America, Inc.
We all have a preconceived notion of what Robotic Process Automation (RPA) is, much like I did. My goal with this case study will be to help you dispel some of those misconceptions and introduce you to some potential benefits of an automated work flow.
WHAT IS ROBOTIC PROCESS AUTOMATION AS MOST OF US UNDERSTAND IT TODAY?
It’s a software-based robot worker, just as you had imagined. But the thing to know here is the idea of how these robots are trained at their tasks. Instead of a programmer coming in to code each step of their day using code-based instructions, these robots can be trained using demonstrative steps. We can simply show the robot what the task is, then it learns and mimics the work, much like training a human worker.
But the promise of a perfect robot worker is not quite achievable…yet.
There is a missing piece to the puzzle to get the system to function properly. We cannot simply hand off our mundane, repetitivetasks to a computer and expect it to perform flawlessly with 100% accuracy, instantly savingyou time and your company money.
Sure, we can train the system on tasks, and this is a must. But what happens when you get an exception in the work flow. For instance, let’s think about invoicing. What happens when you get an invoice from a new vendor, or your current vendor decides to change their invoice format? Well, this is an exception that your robot worker cannot deal with, because it cannot think. It can only complete a given task.
Let’s look at this invoicing example in a bit more detail. Nowadays, there are systemsthat use Optical Character Recognition (OCR) technology. Think of it as a smart scanner. The scanner can look at a paper invoice and extract the information from the scan. It then takes the important information such as vendor name, address, cost, date, etc. and parses that into a database. That’s great when you see the same exact invoice from the same exact vendor. Allof the invoice information is in the same place.
It’s the same process that a human worker would do for manual data entry. You take an invoice, type the information into fields in your ERP software then store that information in your database. And, remember, we are also training our robot coworker at the same time so it can take on the typing task that we no longer have time for. So, the robot worker now knows how to deal with that same invoice from that same vendor where all of the information is in the same place time after time. It’s accurate here.
And we can let the robot worker run on its own, right? Sure, we can if we know that it will see the same exact information time andagain. But this is the real world. We have many vendors sending us many different invoices in different formats. There is no standardization. Now there’s a problem. Familiar invoices only make up about 30% of what a companyreceives from its vendors.
Your new, and expensive, RPA system needs help. It needs to be re-trained on how to deal with this instance of a new invoice format. So, you go back, get a human worker to sit down with the system and go through a step by step process to re-train the system to know how to deal with this iteration. This sounds a bit ineffective. Well just how ineffective is this?
So, let’s look at how this affects your work flow. You have a system that is roughly 30% accurate due to your RPA investment, but this system needs constant manual intervention to fix problems along the way. Your automated system, especially if you have other systems relying upon them to perform their tasks, puts your whole AP operation on hold each time you need to re-train the robots on a new exception. That’s not exactly efficient. But why does this happen? What is going on under the hood?
Automated systems today using current OCR technology are still dependent upon what’s called static data matching. For example, your system is looking for data in the same place on invoices. Then it sends that data to the same place on the server that it has been taught to. It is inflexible. So, how do we address this problem? With Artificial Intelligence, and more specifically, machine learning.
I’m sure you are familiar with the idea of Artificial Intelligence. This is a system that not only learns tasks, but it is now capable of making suggestions. Remember the work flow where we had the robot observing a user’s tasks? Well now the robot is doing more. It is doing its data entry tasks, pulling information from invoices, but it is also dealing with the pesky exceptions as well. Here is how.
THE KEY IS REPETITION.
Let’s say we load thousands of invoices into a system, then let it make up its own mind about what an invoice is based on analgorithm, what features it has, and what information should be where. Now we have a system of machine learning. Now we can ask it to look at a new invoice, and from its previous experiences it can figure out what information is what. And, to take it a step further, if the system is still stumped, it will then interact
with a human, presenting a suggestion. It can narrow things down on its own, so we simply teach it a particular instance of a new invoice, and it will know for next time. Done. No re- training necessary.
So, the system that we looked at without AI, remember that was using a static data matching methodology? An AI powered system, uses a dynamic data matching method. Over time, as the system sees more and more invoices, the accuracy rate increases. This is due to the system’s ability to learn and suggest.
Now let’s look at a few examples of how particular entities can benefit from AI technology. The pharmaceutical industry has begun using Shared Service Centers (SSC) where invoices from multiple subsidiaries can be funneled to one location for processing. They see tens of thousands of invoices a month from hundreds of different vendors across several locations.
As I’m sure you can imagine, there will be daily instances of exceptions on multiple invoices, requiring human intervention to re-train the system. This slows down the operation and, if growth is experienced, will require an expansion of the labor force. With an AI enabled system, exceptions and growth are not things that will hamper work flow. And they shouldn’t be.
Here is another example. The US federal government, specifically the Office of Management and Budget is mandating
that vendors shift to a digital e-invoicing methodology by 2018. This affects more than 80 agencies who are already struggling with tight budgets and a restricted workforce. Add to that a drastic increase in the number of contractor invoices flowing in, and you have a real problem on the horizon.
These agencies need a new way to deal with this growth and increase in invoices, while being able to maintain their current workforce and payment targets. They need a platform that utilizes AI and machine learning to reduce worker’s tasks and their systems reliance on re-training.
RPA is not intelligent. It can observe and repeat, but systems can’t deal with new information without AI. The automation comes from machine learning technology. Once implemented and trained, an RPA system with AI is incredibly accurate, approaching 100% with minimal human interaction.