Haven’t we all been promised a paperless office for years? And although there have been various attempts and solutions, the latest answer to the paper problem appears to be machine learning and robotic process automation (RPA).

Once you have identified paper as the problem in your organization, the first step towards automating your back-office functions is to understand how you can convert inbound streams from paper to a data file, which may include taking advantage of the following:

• Electronic Data Interchange (EDI) computer to computer data transfer
• XML data transfer - sending party provides a structured data file
• Portal for web form creation or data file upload
• Email with an attached file
• File share transfer, such as Google Drive, Dropbox, Share Point, etc.
   

If you can leverage the above, you have come a long way in addressing the issue of paper documents being delivered via physical mail. The trouble is, most of the above options require the sending party to make changes. EDI and XML require an IT project to complete and maintain the transfer. A portal will require duplication of effort by the sending party and a willingness to connect to a multitude of portals.

Emails are by far the easiest method of sending and receiving files. After all, who doesn’t have access to an email account these days? Files are easily attached to email too. A file share can be useful, but how do you know if the documents have been received and processed effectively?

What the above process has done is change the mindset of the receiving and sending parties to adopt electronic transfers, so we are starting the digital journey and beginning to solve some of your processing inefficiencies!

How Can Businesses Take Advantage of Robotic Process Automation?

So, now you have the data - unless, of course, the files sent are images. If someone has scanned a document to create an image (pdf/tiff/png, etc.), then you don’t have data, and you need to convert the image (picture) into data using Optical Character Recognition (OCR)

The challenge here is although OCR has been around since 1913 (Optophone), and was commercialized in the 1930s, OCR technology has undergone many variations and developments in the last 70 years or so. OCR is reliant on the quality of the document being scanned and the scanning quality.

We’ve all seen ‘B’s’ converted to ‘8’s’ and ‘1’s’ to ‘L’s’, among other errors. These OCR errors require powerful algorithms to try and clean up the error or to pass them over for human intervention. So, it’s clear that an efficient and automated process needs to avoid paper and OCR.

We are now left with the inbound data file, which is where AI and robotics can come into its own and can start to transform and optimize your digital transformation and automate your process.

Computer programs and robotics rely on logic and data to function, an image file is simply a picture that a program can do nothing with. If you provide a program with the data from an application, invoice or patient, we can then automate that logic.

As you receive data and do not need to rely on OCR, you can’t misread the characters, so gone are the days of translation errors, and we have accurate data to work with. The next questions are, why do I need the data and what do I need to do with it to complete the transaction/task?

These processing steps/rules are what an AI and robotics solution can manage and automate. Taking an accounts payable process as an example, if we address those suppliers who issue the highest volume of invoices first, we will ensure an early and rapid reduction in manual processing. Get them to send a data pdf via email, it’s so simple, and there is no cost to the sender.

The process will look something like this:

• A supplier issues an email with attachments or file transfer of a data file (pdf, MS Office Doc, other data stream).
• Receiving party collects the file from an email account or file share and because the data is already present, it deploys a program to read and extract the data, with 100% accuracy.
• Suppliers will send their layout of an invoice with critical fields in different locations and with different formats such as: Date formats; Number lengths and including a prefix or suffix; Units of measure; Delivery addresses.
   

Once the program has the data and understands what the receiving system requires, the extracted data can be validated and, where required, enriched to ensure the receiving system can automate the upload and then process the transaction.

This process may include validating the data with a purchase order (PO) and or a receipt note. This is where the interesting stuff happens. Let’s assume there is an error, the price is incorrect, or the PO cannot be found, the AI/robotics can trigger an action to automatically query the price or the validity of the PO and action the response.

If there is a logical set of business rules that can be deployed to manage a business process, this can be programmed, and the formula can be as complex as required, but where a human makes an intuitive decision, it’s harder to program.

AI uses the programmed intelligence, and the robot can fetch and carry data based on the applicable logic and it’s a symbiotic relationship. But, it all must deliver the required outcome and accommodate the multitude of variances and scenarios that are found in the real world, such as 8/12/2018, now is that the 8th December 2018 or the 12th August 2018? If the sender had an address in the USA, then August would be the right answer.

AI and robotics can process your data very effectively, but any robotic process needs to receive accurate data in a format that it can read, the downstream process must be understood and supported by the AI and RPA process to deliver value and effective business outcomes.

Steve Britton is a Business Process Digitisation and Automation Solution Specialist with e-invoicing network CloudTrade