Everyone talks about AI (and more appropriately machine learning) as being the silver bullet to the challenge of achieving highly efficient business processes. And on the surface, the excitement and perspectives appear appropriate. After all, machine learning has the ability to see things in data that humans might never notice and automatically perform tasks with consistency, all day long.
But in all of the excitement over machine learning, some basic fundamentals of processes are being overlooked. One is the fact that processes, especially complex ones that deal with complex data, will never achieve 100% automation, so a human is always needed to monitor and step in when the machine needs help. Many in the automation world call this “human in the loop”. Yet while all of the attention is on increasing efficiency from automated tasks, there is still a lot of process friction when humans get involved.
Take for instance the process of evaluating the results automated by a system. Do reviewers have to move their mouse around and perform a number of tasks to review the information and make corrections or does the software enable very quick review via a highly efficient layout and reviewer aids? Often times a “human in the loop” mentality leads to outright neglect of the user experience and efficiency of the human-oriented processes making things cumbersome at best. This is especially true for processes that involve complex document-based information such as remittance reconciliation within healthcare. In that process, complex remittance documents often referred to as Explanation of Payment or EOPs must be examined and compared with the initial claims and balanced using contracted adjustments on a service-by-service basis. As you can imagine, this process involves a lot of searching for data and data entry into revenue cycle management systems. If a “human in the loop” approach is taken, most of the focus is on achieving high levels of touchless automation with anything leftover being handled by exception. Unfortunately, due to data complexity, touchless automation may never go over 60% leaving 40% of data to be reviewed and potentially manually entered.
But there is another approach that can improve the efficiency of the review process. We tend to think of it as “assistive automation”. Instead of focusing solely on “lights-out” automation of data entry, we combine process reengineering of the review and data entry process itself. Instead of relying upon a traditional data review and entry approach that would display page after page of data requiring the user to perform a lot of actions, we add capabilities to make review and data entry much more efficient. These process enhancements can include a “shepherded” approach to display of data that helps the staff focus on only what is required while also providing aids through contextual data. If a field needs to be reviewed, the user might be aided by not only focusing their attention on that specific area of a document but also by displaying other information such as the matching claim data and then indicating what the problem might be. Through time-motion studies of the process, we might find that simply keying the data instead of using a mouse to point to it can reduce the overall time required by 30%. The addition of context along with a shepherded workflow can take a review process that might take 2-3 minutes down to less than 20 seconds. That may not seem like a lot until you add up all of the review tasks that are required.
It may be hard to grasp, but overall process efficiency and significant improvements don’t need to rely as much on advanced ML-based AI as many believe. The most successful automation projects take a holistic approach, not at just the steps that can be automated, but the steps that require us humans. At Parascript, we do spend a lot of time working on novel ML-based technologies meant to reliably automate as much as possible, but we also sweat the small stuff that is not about automation but taken together creates the most efficient, reliable, and effective processes.