Three major trends are reshaping the industry: a major rethinking of outsourcing vs. moving in-house, advanced handwriting recognition becoming mainstream and the increased need and reliance upon data science. This article explores these trends, as well as what they mean to enterprises and their service providers.
Rethinking the Benefits of Outsourcing
Just recently, HFS research published an article on the acceleration of insourcing operations that service providers currently provide. Why is this? One of the primary reasons is the renewed interest in automation. Also, it is the perspective that with automation, reliance upon manual labor is reduced, the outsourced version of which is still the primary business model of many service providers both large and small.
Historically, if an organization wanted to rid itself of low-value, but necessary tasks or processes, the best option was always to outsource these functions to a service provider that could provide the same capability at less cost through economies of scale. With automation, there is the promise of handing over the work to “bots” that can be deployed anywhere and whose costs are not sensitive to typical wage arbitration. A bot costs the same whether it is deployed onshore or in low-cost regions. HFS calls this “going straight to digital.”
Service providers are impacted by increasing expectations to improve efficiencies through the inclusion of automation within their current outsourced operations combined with the desire to participate in those cost savings. Enterprises, on the other hand, need to improve their own understanding and wherewithal with respect to these automation technologies.
Forms Automation: Handwriting Recognition Redux
While there is a continued decline in the use of paper-based processes, there is still a significant amount of paper flying around, mostly due to the need to support a wide variety of transactional processes that include a number of stakeholders where the ability to apply standardized technology is limited. Simply put, while the use of “e-forms” instead of the paper variety can significantly improve a process, the ability to deploy this technology consistently and broadly enough to improve a process is limited to specific areas. While handwriting recognition has been around for almost two decades now, it has been limited to specific, constrained applications where enterprises use well-defined data.
With increases in low-cost computing power combined with deep learning algorithms, handwriting recognition is going through a renaissance, of sorts. Using deep learning and a significant amount of data, it is possible to expand the use cases beyond specific form data types such as dates and amounts to more expansive fields; this includes multi-line entries that are common within the healthcare and finance industries.
For example, patient descriptions regarding a condition can be transcribed more comprehensively and claims adjudication, which includes handwritten correspondence, can be more effectively automated through the ability to process larger amounts of unstructured handwriting. Both service providers and enterprises benefit from this expansion of document automation capabilities to this most complex and very common form of unstructured data.
Data Science Goes Mainstream
While a lot of excitement and activity surrounds automation of all sorts, the ability to apply automation successfully depends heavily on the nature of the tasks involved. Many processes can benefit from simpler rules-based automation, which has been the focus of robotic process automation.
When considering more-complex processes that require the use of difficult unstructured information, the ability to achieve a meaningful of automation via scripted rules is almost nonexistent. This is largely because unstructured data has a significant amount of variance that requires a significant amount of data to analyze and on which to build rules. However, the amount of data is so varied and significant that most humans cannot analyze it to construct rules that can appropriately manage it all. This is where machine learning comes into play. What many organizations still do not understand are the fundamental prerequisites to applying machine learning to complex unstructured data tasks: namely, large curated input sample data.
Currently, the domain of large tech companies such as Google, Facebook and Amazon, the process of collecting, curating and tagging input data is expanding to include enterprises that seek to use and benefit from automation powered by machine learning. Yet most organizations do not employ data scientists to deal with data problems such as bias, so they are increasingly relying upon specialist software vendors and service providers to provide this critical component.
Even while organizations will accept the benefits of “black box automation” preferring to focus on outcomes, they still need to educate themselves on statistical and data science concepts so that they can truly understand where the line should be drawn between the definition of a successful automation project and one that doesn’t live up to expectations.
Greg Council is Vice President of Marketing and Product Management at Parascript. Greg has over 20 years of experience in solution development and marketing within the information management market. This includes search, content management and data capture for both on premise solutions and SaaS.