Robotic Process Automation (RPA) can reduce the burden of repetitive, simple tasks on employees. RPA uses software robots (Bots, for short) to automate boring data entry tasks such as screen-scraping data from an open application window and entering that data into a database or another business application such as ERP or HR.
The fundamental value of a bot lies in its programmed ability to run completely unattended and never need a break. We can refer to this operation as straight through processing.
Bots rely on predictable and reliable data inputs. They work best for processes where there is no deviation from the game plan. That is why bots are ideal for basic data entry tasks or for manning a self-service chat.
To run as intended and deliver the promise of straight through processing, a bot must in essence enter into a Service Level Agreement (SLA) with the source of the data. If anything deviates from the game plan, the bot will stop, and the process will grind to a halt. A human operator must then intervene, which of course devalues the utility of the bot.
While one should expect a bit of exception handling to occur in many processes, it doesn’t take a lot of exceptions to ruin the return on investment that was based on replacing human labor costs.
Enter the document
RPA adopters are eager to expand the use of RPA to automate more complex tasks that currently require humans to open and read documents. For example, a PDF invoice arrives via email and a human must open it, read it, and enter selected data into the accounting software. A more complex example would be mortgage loan underwriting, where dozens of different documents in various formats are submitted and data contained inside those documents must be organized, verified and entered into the loan origination system.
At this point, RPA leaves the safe and happy shire of data predictability and reliability and enters the Mordor of document capture. Anyone with experience at this knows how hard it can be to achieve, let alone maintain, true straight through processing of data from unreliable and unpredictable documents.
There are several document capture options available for this exotic use of RPA. Let’s call this “Cognitive RPA”. How do you choose the right tool for the job?
Let’s examine two different process scenarios.
Despite the promise of seamless digital invoicing between companies, most invoices today are still submitted as individual PDF files and there is no standard format for the data layout. Each vendor has its own invoice format. Essential data such as company name, invoice number, part numbers, taxes and purchase amounts can occur anywhere on these disparate invoices. Remember, the bot doesn’t care about any of this. It has an SLA with the document capture software and the data had better be good.
Template-based capture solutions are the most common for invoice capture. This combines our old friend OCR with predetermined invoice templates. However, given the variability inherent to most supply chains, this may get you only to 50% straight through processing. Bots will stop until humans read the invoices and manually enter the data. Since an enterprise RPA solution can cost nearly as much as human labor, this doesn’t make any sense financially.
To achieve anywhere close to lights-out automation with a bot, you will need very sophisticated intelligent document processing (IDP) software to meet the SLA. It must be capable of extracting the correct data from anywhere on the invoice, do it at speed, and make intelligent decisions on the fly using machine learning to determine the best data extraction modality for each document.
Mortgage loan origination
When one applies for a mortgage, the lender opens a loan file for the applicant. It’s easier to think of this using the familiar file / folder metaphor. The applicant must submit documents (or “files”) that will go into their folder to be analyzed by the lender, who will make a decision on whether or not to offer the mortgage.
For a typical USA mortgage application, the loan file can swell to over 100 different documents in a variety of formats: scanned images, fax images, JPGs from a smartphone camera, PDFs both digital and image-based, spreadsheets, emails and more. There is no data layout consistency. Some images will be low resolution and even contain handwriting.
The amount of human intervention involved to analyze the data contained in a single loan file can run into days. RPA to the rescue? With the document mess we’ve described here, only if the most sophisticated IDP software is deployed on the front-end to meet the data SLA and provide steady throughput.
Cognitive RPA is only as good as your IDP
RPA is an amazing advance towards a future of truly automated repetitive processes. There are so many document-intensive processes yet to be tamed by RPA. When you select an Intelligent Document Processing solution to make your RPA “Cognitive”, choose wisely. There are no shortcuts and there are no easy solutions. Go with a vendor with deep experience and the best science behind the software. Now is the time to consult an expert who can assist your evaluation process.