How robotic is your process ?
Bumps in execution
Consider the most obvious issue: That many processes in the real world encounter ambiguous or erroneous or simply unanticipated “bumps” in their execution. The “robots” of RPA only do what they have been told to do by developers. We know that developers often can’t anticipate all the potential bumps, and it would be beneficial to engineer into the “bot” a “breakdown lane” for exception handling—a place that offers the chance for humans or other bots to triage and repair the interrupted transaction.
Beyond the exception-handling case, there are processes in the business that are inherently nonlinear, particularly in areas such as product design and development, sales and/or sourcing negotiations, interactive customer interaction, and M&A—the list is long and important. The further you look beyond the world of highly structured business environments, the more you find an automation application relying on human communications, decisions, and relationships.
What’s needed
To break out of the structured process world, RPA will need to address the full range of cognitive computing capabilities. It will need to master the conundrum of how to set up the kind of “clean” data environment that enables “AI” tools such as machine learning and deep neural networks to successfully identify significant patterns, reveal important insights, or prepare recommendations for human decision making. It will need to master the metadata-based integrations with the multiple external systems that animate the enterprise computing environment. Making RPA cognitive begs the question: At what point do application engineers decide that an RPA bot might be trusted to make a decision on its own, without human input or oversight? But if human interaction continues to be required, what kind of human-bot communications will prove effective? Voice? Text? A purpose-specific alert system?
Ready for prime time
Beyond this basic set of considerations for cognitive RPA applications, a huge question looms: When will we be comfortable enabling the machine itself to design the process? Will RPA allow the robot to create the application from the ground up?
Kurzweil’s blood-stream-resident nanobots, which integrate the latest in medical research and clinical practice to identify weaknesses or failures inside the human body and make decisions about what kinds of enhanced nutrients, medical procedures, or repair/replace operations are necessary for optimal health, are due to be in production (according to Kurzweil) by 2030. That timing is just about right for Kurzweil, who will turn 82 that year. However, the more immediate question might be: Will cognitive RPA be ready for prime time by then?