Will cognitive automation spell the end of outsourcing?

I recently spoke at the Deloitte Shared Services and Outsourcing Executive Forum at Deloitte University. The audience’s interest level in automation strategies was high. For the most part, Deloitte doesn’t outsource directly, but only provides consulting services to its clients about outsourcing and shared services strategies, so the event and this essay seem appropriate venues to discuss the impact of “cognitive automation” technologies on the outsourcing industry.

I use quotation marks around the term “cognitive automation” because though I am told that that is the preferred term these days at Deloitte, I am still getting used to it. I like the term because it suggests a broad approach that goes beyond current automation-oriented technologies and methods. But unfortunately, there is little consensus on what term to use in describing technologies that perform tasks previously done by humans. As I announced at the forum, I have little love for the term “robotic process automation” (RPA) that has been used to describe tools that automate highly structured administrative tasks. This proclamation was somewhat unfortunate, given that (as I found out later) one of the conference participants had a hand in creating the term, but she didn’t seem to mind my criticism of it.

The fact is, however, that the process automation employed thus far has been quite robotic. That is, it has been applied to repetitive, structured work that can be expressed in logical rules and workflows. Some have referred to this type of work as “swivel chair processes” because it typically involves interfacing with multiple systems. They might include, for example, replacing a lost ATM card in a bank or automatically rebooting a network device in a data center.

Doing these types of tasks with a machine may not be earth-shattering, but the early adopters of RPA do seem to have generally achieved some impressive returns on investment (ROI). In a series of case studies done by the Outsourcing Unit at the London School of Economics, the RPA implementations yielded rapid (and high) ROIs. In one telecom company that automated 160 different processes, the ROI ranged between 650 and 800 percent.¹ Because RPA is relatively easy to configure and implement in a particular application, it typically produces benefits rapidly. However, since the processes it helps automate are typically small and administrative in nature, automating even large numbers of them is not necessarily transformational for the organization that undertakes them. Most of the internal people whose jobs these systems automate are redeployed to other, more interesting types of work.

They may be transformational in a negative sense for services outsourcers, however. While there don’t seem to be any overall numbers available, many of the RPA projects I’ve learned about have involved reductions in headcount—but headcount owned by offshore outsourcing firms. It’s a logical trend; the processes that were structured and codifiable enough to move to services outsourcers are the same ones that will be automated. Many outsourcers have realized this and now offer either their own proprietary RPA solutions or “white-labeled” versions from other vendors.

The automated work may be highly structured at the moment, but it will not always be so. Deloitte’s preferred term, “cognitive automation,” suggests correctly that automation software will become more cognitive and less robotic over time. Several vendors that once offered only structured automation software are now claiming some degree of cognitive capability, from the ability to learn to the ability to interact in human language.² Business processes involving a high degree of knowledge and expertise, from radiology to legal document review, can already be at least partially automated. Much IT programming work will eventually be automated as well, according to no less an authority than Bill Gates—who speculated in an online chat that it is “safe for now,” but perhaps not in the future.³

This will affect not just a few companies within an industry, but also global employment and economic trends. Outsourcing, for example, is one of the biggest sources of employment in India. The industry has been a major factor in the rise of the middle class in that country. It’s not obvious what kinds of activities will replace outsourcing for these workers. And in countries like the United States, it’s likely that increased use of cognitive automation technologies will bring back at least some jobs from offshoring locations. Almost all automation systems require some degree of human configuration, oversight, and maintenance.

However quickly this transition will happen, it’s becoming clear that in the future, outsourcers will have to do more than provide labor arbitrage. They will have to provide a variety of cognitive automation technologies to the marketplace—both their own and those of other firms. Their remaining people will have to be process and technology experts, and intervene when the automated system can’t do the job. They’ll have to become skilled at work design and the appropriate division of labor between humans and machines. The fact that there will still be humans involved in almost every process suggests that “augmentation” may be a better term than automation.

These capabilities are so different from those of present-day outsourcers that it calls into question the very name of the industry. Will there still be “outsourcing” as we know it today? When a machine does a company’s work—regardless of who owns that machine—is it being outsourced, automated, or augmented? Perhaps “business process outsourcing” will become “cognitive process oversight.” Perhaps “offshore outsourcing” will transform into “offshore augmentation.” All that’s clear is that some smart machines are coming, and they promise a very different outsourcing industry.