What Hyperautomation Is…
It goes by many names...Gartner calls it “Hyperautomation”. Forrester calls it “Digital Process Automation”. And IDC calls it “Intelligent Process Automation”.
According to Gartner’s report on the Top 10 Strategic Technology Trends for 2020, hyperautomation is “the combination of multiple machine learning, packaged software and automation tools to deliver work” and it “deals with the application of advanced technologies including AI and machine learning to increasingly automate processes and augment humans.”
Hyperautomation was at the top of Gartner’s list of strategic technology trends for 2020 and with good reason; as the adoption of automation is rapidly increasing, the technologies and tools available are simultaneously evolving and improving, setting the stage for a whole new era of automation – one that makes its potential in optimizing business processes and improving the lives of employees even greater.
So how does this fit in with accounts payable processing and why did we want to talk about it?
When a prospective client comes to us about our Accounts Payable (AP) automation solution the main questions we are asked is: ‘how good is your OCR scanning”? That is a great question, but it really is just the start of a whole process.
What you really want to know when investigating AP automation providers is whether their solution is “end-to-end automation” (or now called hyperautomation). So the question to ask is “what is your complete technology build” and in AP automation it must include:
• High accuracy OCR - converting images to text
• Smart data extraction - transforming the text into relevant data
• Human-in-the-loop (HITL) - 100% accuracy
• Machine learning - remembering the data and populating it into the applicable data fields
• Robotic process automation – mapping, coding and automating the right workflow for each document
So let’s break this down and look at all of these moving parts.
OCR – Optical Character Recognition. The technology that turns typed, printed or handwritten text into machine-encoded text.
Simply put, the technology is looking at an image or file and is able to identify what is on it — turning a picture into words. So is our OCR accurate?
Obviously, the accuracy of the conversion is important, and most OCR software like ours provides 98 to 99 percent accuracy, measured at the page level. This means that in a page of 1,000 characters, 980 to 990 characters will be accurate.
That accuracy level however will vary widely with the quality of the document being scanned. True PDF documents will get the best results and while our engine is sensitive enough to read low resolution scans and non-original documents, if there isn’t a 100% match the document is escalated to our validation queue to be ready for the data extraction process.
Smart data extraction. This next step is used to understand and process the text from the OCR to transform it into relevant data. This is critical to know because OCR by itself does not know what to do with the information it reads. This is where the “smart” in smart data extraction comes in. And it’s important to differentiate the two.
This functionality powers the next steps in the invoice processing workflow. Once data is extracted from unstructured content and then validated it then interprets the information and then applies it to the appropriate field. i.e. Put the invoice number from the PDF document into the right field in the application.
If at any time the validation of the fields extracted is incorrect, at expensemanager we send the invoice to our validation queue. By intercepting an invoice at this time the machine can be trained and this is critical for continuous learning.
Human-in-the-loop (HITL). Human-in-the-loop is an approach that we at expensemanager have championed for years. Human-in-the-loop machine learning can be thought of as attended machine learning applied to unstructured content. Machine learning in itself is very powerful for working with structured data, but in the accounts payable process there are many examples of unstructured data and reasons for the machine to not learn.
It may be a new supplier comes on board with an out of the norm invoice template, an existing supplier changes their invoice layout, there is a poor scan or a handwritten invoice, invoices with coffee stains!
Without the human-in-the-loop model every time an exception is received it will require manual data entry into the Finance system and because the machine is not trained low accuracy rates will continue. Without this critical part of the process, there will not be continuous learning.
In our opinion, that’s just not a complete service. Document exceptions are handled within our end-to-end process and this process is managed entirely by our team.
Machine learning. (ML) Utilising machine learning means the system will become more intelligent — smarter — over time, reducing the number of mistakes and increasing overall efficiencies. Remembering the data and populating it into the applicable data fields each time the data is recognized. e.g. This supplier has invoices with freight on it, so create a freight line as well.
Robotic processing. The final step. Robotic Process Automation reviews the now digital invoices against a company configuration and applies the business rules automatically. e.g Automatically approving the invoices that comply with the configured business rules and matched up with its purchase order. Or when an invoice and a purchase order amount do not match up and are outside of the tolerance range, then automatically route to another workflow.
You are looking for a system that includes every step, achieving the perfect harmony between the work of machine and that of people and gets smarter the more you use it.