RPA (Robotic Process Automation) is the latest trend in digital transformation, with new entrants regularly joining the automation arms race leaders like UiPath, AutomationAnywhere, and BluePrism have pioneered. Lauded by many as an exponential boon to productivity, there is lots to consider when starting your data on an RPA journey, particularly, where the data originates. Let’s take a look at several use cases where paper-based input can create challenges that can make or break your RPA investment.
The situation: Michelle leapt from her chair at Central Bank’s loan processing department in a panic. “806! What does 806 mean? Has anyone ever heard of an 806 loan?” She showed a co-worker, “There, where the name should be, it just says ‘806’.” Equally surprised, her colleague shrugged. Under pressure to quickly process the application and render a decision back to the applicant, Michelle called the branch, explaining the mysterious application to Tom. “There’s no name. It just says 806.” After a pause, Tom let out a laugh, “Oh I know what happened. The applicant’s name is ‘Bob’ and the applications are hand-written. It must have read it in wrong from the scanner.”
What went wrong: Tom was partially correct. In a workflow like this, where applications are scanned, and an RPA bot uses OCR to parse the application’s fields populating the inputs into a database, there are possible several points of failure. Handwriting is very difficult for OCR engines to decipher. Often leaning on Machine Learning models to inform it, a B can easily be read as an 8, an O an 0, and so on.
Why quality matters: In cases like this, image quality is perhaps the most important factor for success or failure. A crisp, high resolution capture of each handwritten character presents the OCR software with the best chance to turn it into useable data for people like Michelle waiting at the other end and preventing Bob from just becoming a number. By increasing the contrast of edges within an image, automated sharpening on document scanners can make objects in the image appear ‘crisper’. This improves the appearance of the document for higher OCR read rates.
The situation: John, a senior analyst in the billing department, couldn’t understand how it happened. The new RPA bot was supposed to perform their monthly invoice reconciliation flawlessly. Last week, one of their biggest enterprise clients called to ask if there had been a mistake. “It’s not that we mind, but it’s usually us paying you, not the other way around,” the client’s accountant said. They had been issued a $300,000 credit on their monthly statement, exactly the opposite of what would have been expected, given the completion of a new equipment rollout at the client’s headquarters.
What went wrong: Scouring the database for the original, signed statement of work, John immediately saw the problem. On the total line, the amount that is remitted or billed to client, just along the line which read $300,000, was a dot. It was obvious to the human eye, but to a document scanner, OCR software, database and RPA bot that needed the information, it was read as a minus sign.
Why quality matters: When converting color documents to black and white the appearance of dots or marks, referred to as ‘noise’, can be caused by dust or lower paper quality like recycled paper. Noise removal algorithms can remove these dots to improve the appearance of a document, and minimize the impact on downstream technology that needs it.
The situation:The switch to the new intake forms at Memorial Hospital seemed to be going well. Patients loved the ease with which they could fill them out on their phones or on the terminals in the reception area. Behind the front desk, the completed form would print out, and the receptionist would request a physical signature from the patient. From there, the form was scanned in, where OCR software extracted the info and an RPA bot routed the data to the doctor, insurance provider, and national patient database, while initiating mailings of visit summaries and billing statements to the patient. About a week in, patients began calling, saying they hadn’t received a bill. Around the same time, the hospital saw an influx of returned mail, with addresses half completed, cities, states, and zip codes missing.
What went wrong: Surely this many people didn’t forget to include the second half of their addresses. A team went to work to analyze the forms, and realized the scanned, digital copies had black streaks running through the fields containing the missing parts of the address, resulting from dust in the scanner’s housing. The OCR software, though it knew where to look for the data, was unable to understand the data it was looking at, seeing a combination of letters and numbers with a line through part of them it could only assume was garbage. The RPA bot, trusting the data fed to it by the OCR software, continued its process accordingly.
Why quality matters: While this could have been fixed in some of the downstream technology (a bot performing field validation, for example), it is optimal to fix it at the point of capture to reduce the opportunity for error later. This is a common issue that streak filtering technology helps to address. Regular cleaning is needed to completely avoid this, but image enhancement technology can remove or reduce it when it occurs.
If considering an RPA solution for your business, congratulations. You are at the forefront of automation technology. To make the most of your investment and ensure the best chance at a timely ROI, the complementary technology must be up to par. RPA bots are only as good as the quality of the data they are fed. And data, especially paper-based, is only as good as the tool used to capture it. Kodak Perfect Page technology extracts optimal data from imperfect paper documents, correcting everything from dust dots and streaks to 8’s that are supposed to be B’s.