RPA and ROI in Customer Service Centers: When Bicycles Fly?

By Chris Surdak, JD
Senior IRPA AI Contributor

Some of the most spectacular stories about ROI in the world of RPA come from the automation of customer service processes. There are tales, perhaps myths, of giant organizations using RPA in customer service centers generating full ROI within days, if not hours, of implementation. These service centers previously employed thousands of human workers who answered phones and questions. Post-implementation, these same centers maintained a few supervisors who oversaw the work of dozens of bots, performing the same job quickly, quietly, efficiently, effectively, and most important, cheaply.

Now, let’s get real. The likelihood of any of these stories being real is about as likely as finding E.T. hiding in your closet eating Reese’s Pieces. The more fantastic the claim, the less likely it is to be provably true. Accepting the reality that these unicorn stories are fictional does not discount the value RPA brings to the table. On the contrary, our desire for mythical intervention in how our organizations currently operate is a manifestation of our deep-seeded understanding that we can and should perform so much better, yet we are not certain of how to achieve such a breakout performance.

Here’s a clue: Better tools never beat better approaches. If we truly want better results, we need better verbs (actions) not better nouns (things). Further, outstanding, disruptive, revolutionary outcomes are created by the right combination of tools and approaches — rather than one or the other.

ROI versus ROV: Chasing the right goal

The dominant driver in the hundreds of RPA deployments I’ve worked on through the years is ROI via cost savings. The emphasis is nearly always on doing the same for less rather than doing more. This naturally stems from the efficiency mentality underpinning the Lean movement, which unfortunately often morphs into a scarcity, zero-sum-game mentality when misapplied.

While achieving ROI may be inherently good, not all returns are created equal.

  • Michael: Did you explain school to him (E.T.)?
  • Elliott: How do you explain school to higher intelligence?
  • Michael: Maybe he’s not that smart. Maybe he’s like a worker bee who only knows how to push buttons or something.        

While it may be true that a penny saved is a penny earned, it stands to reason that a penny earned on an additional dime of revenue gives the recipient many more options than simply saving a penny; those other nine pennies of cost-covering revenue bring a wealth of operational options with them.

Rather than focusing solely on ROI, I suggest focusing on return on value (ROV). While cost savings is pretty one dimensional, value comes in a variety flavors and textures. Pleasing a customer who only values cost is a binary affair; you’re either the cheapest or you’re not. Conversely, customers concerned with value may be served in a wide variety of ways — with more opportunities to increase revenue — as opposed to just cutting costs — which necessarily decrease revenue. The fairy tales first appeared around 2016, just as I was coming up to speed with RPA. They spoke of near limitless potential with this new technology and its ability to communicate across space and time with a mere finger, speak clairvoyantly through others, and make metaphorical bicycles fly. If E.T. was the extra-terrestrial of 1980’s movie fame, RPA was E.T. the extravagantly transformational of the 2010s.

  • Elliot: He’s a man from outer space and we’re taking him to his spaceship.
  • Greg: Well, can’t he just beam up?
  • Elliot: This is REALITY, Greg.                   

Now, five years on, we have a more pragmatic understanding of RPA and what it is, and is not, truly capable of doing. And while RPA may not have us flying across the sky on our Huffy, it can certainly bring tremendous value to business processes — even the mundane operations used by the typical Customer Service Center.

How may I help you?

So, what does all this have to do with RPA in call centers? Many have attempted to fully automate their customer service processes with RPA. This is the nouns/tools approach: Keep the same old processes and approaches, but use RPA instead of people to do the work at lower cost. The assumption is that humans and bots are interchangeable, and bots represent better, faster, cheaper versions of people. Replace people with bots in your business processes and voila, instant cost savings and process improvement.

  • Gertie: Here he is (E.T.).
  • Mary: Here’s who?
  • Gertie: The man from the moon. But I think you’ve killed him already.

As with most things, this absolute view is both incorrect and potentially damaging — as many organizations can now attest. The fact is, humans and bots have opposite, and complimentary, strengths and weaknesses.

  • Humans are great at cognition, pattern recognition, and problem-solving. We’re terrible at completing repetitive tasks quickly according to a set of immutable rules.
  • Bots are great at completing repetitive tasks quickly according to a set of immutable rules. Cognitively, bots are as dumb as a box of rocks — or at least no smarter than the coder who defines and develops them.

This critical distinction must be acknowledged and accepted. Henry Ford notwithstanding, humans are terrible at repetitive tasks. On average, we bore easily, lose focus, and tire quickly. Over a century, studies have shown that people will inject variety — also known as errors — into otherwise stable processes simply to break the monotony. Anyone who works with large-scale call or customer service centers knows people burn out and leave these jobs quickly, and many such departments see employee turnover rates at or above 100% annually. The associated cost of high turnover is one reason so many companies hope to automate this portion of their business.

Conversely, bots start out with an IQ of zero and only increase from there with great effort, and at significant cost. The number of if/then statements required to lift a bot’s IQ by a single point increases geometrically, if not exponentially, and over time, the rules are likely to change far more frequently than most companies realize.

Such business requirements follow the old 80/20 rule, whereby 20% of the requirements drive 80% of development effort and cost. When John F. Kennedy said, “We will land a man on the moon before the end of this decade, and bring him safely back to Earth,” his three requirements drove nearly all of NASA’s budget for the Apollo moon program.

What does Apollo have to do with RPA bots? Not all requirements are the same nor are their resulting solutions. If you intend to fully replace humans with bots, you must replicate all the tasks they perform — both physical and cognitive. Bots can automate physical tasks with ease, but they perform cognitive tasks poorly.

The cybernetic workforce

As a result, the best answer for meeting your automation goals is augmentation, rather than substitution. Full automation of an end-to-end process belongs to the world of enterprise software packages; monstrosities of complexity and nuance that cost tens or hundreds of millions of dollars and years if not decades to deploy. In contrast, Intelligent Automation is all about task-oriented point solutions that maximize the strengths of RPA to enhance the inherent abilities of humans. In short, let the bots do what they do best and let the people do what they do best.

  • Scientist: You said it has the ability to manipulate its own environment?
  • Michael: He’s smart. He communicates through Elliott.
  • Scientist: Elliott thinks its thoughts.
  • Michael: No, Elliott . . . Elliott feels his feelings.           

What does a cybernetic workforce look like? Business processes are decomposed into their constituent parts — tasks, rules, inputs, and outputs. Repetitive tasks, and those which require a significant number of common steps, are assigned to bots. Tasks driven by hard-set, easily articulated rules are also owned by bots. Bots work best where inputs and outputs must meet definite criteria to be correct and where quality can be easily defined by quantitative criteria.

Conversely, any tasks or rules qualitative in nature, which require judgment or perception to properly apply, must be assigned to humans. Arbitrary or conditional rules are also the domain of humans. Inputs that must be conditionally evaluated before use, in which the evaluation criteria contain multiple conditions, and change with context, should also belong to humans — if only because creation and maintenance of such code is difficult and expensive. The same goes for correct outputs; if they have a lot of interdependent criteria and are qualitative and context sensitive, leave it to people to decide.

This blending of humans and bots is critical to operational, and financial, success with RPA. Replacing humans with RPA is a fool’s errand. Enhancement and leverage are the path to success with RPA.

The human factor in RPA adoption

Not enough digital ink is spent on the issue of sociological RPA adoption. In nearly every RPA implementation I worked on over the last seven years, there was palpable fear of this technology amongst those who would be using — or replaced by — it. If an employer is chasing a replacement strategy, such fear is completely understandable — and justified. Conversely, if one is following an augmentation strategy, users tend to welcome RPA with open arms.

Tell 100 call center workers that 50% of them are about to be replaced with bots, and you should expect to breed some digital Luddites who will resist your efforts at every turn. Instead, tell the same workers you will be eliminating 80% of the job they hate, and making their work easier, and you will likely see a strong embrace of this new initiative. Over time, will 100 workers shrink to 40 or so? Of course! But this should occur gradually through natural attrition rather than a pronouncement beforehand.

  • Elliott: You could be happy here; I could take care of you. I wouldn’t let anybody hurt you. We could grow up together, E.T.                

This also demands active management of the attrition process. Most people resist change. As you bring about change through automation, some people will embrace it, some will resist, and most will passively accept it in 20-20-60 fashion. Achieving a cybernetic organization requires you to identify and support the 20% who embrace the change, remove the 20% who resist it, and actively nurture the 60% in the middle who are willing to accept and adapt to change.

Finding the 50% reduction you hope for in your human workforce is relatively easy. They will self-identify if you allow them to do so. Manage this correctly, and you’re likely to succeed. Fail to actively manage your workforce transformation, and your chances of success are slim to none — with “slim” actively looking for work on Glassdoor and LinkedIn!

A practical example

So, what does effective automation look like for a call center agent? How do you optimize the use of RPA to maximize immediate value for your customers and your business? And how do you set yourself up for active adoption by your staff, rather than active resistance? Consider these pointers:

  • Leverage both unattended and attended bots. Unattended bots are those that work individually without direct oversight from humans. Attended bots are those working under the supervision of, and preferably in concert with, humans. In automating call center processes, unattended bots should be used for initial triage (e.g., collecting customer data and determining the nature of the call), input and output processing (i.e., when and if context is quantitative), and in pre-processing transaction data to facilitate the call. These are tasks an unattended bot can perform quickly, with a high probability of success, and near perfect repetition. These are setup tasks for the tasks that follow.
  • Where direct interaction between customers and representatives is preferred, use attended bots to facilitate interaction between humans. Attended bots can quickly call up multiple customer records, availability of stock, rules and resources for resolving issues, and other look-up or process activities that benefit the transaction. The goal here is to speed up necessary, but non-value-added, activities performed by the service representative, so they can focus on their interaction with the customer. Ideally, these transactional automations are embedded in a process map or script the service representative follows as they serve the customer — performing incremental tasks to serve the script rapidly, consistently, and accurately every time.
  • Attended bot scripts should start out simple, automating the 20% of tasks that address 80% of use cases. Over time, incremental tasks can and should be added to automation scripts, so they encompass an ever-growing percentage of the process’s total workload. Note that humans are never fully replaced in this approach. Rather, their time and attention are constantly focused down to those tasks best served by humans while the bots capture more and more of the “rules and clicks” workload. Ideally, process cycle times will drop precipitously as customer satisfaction inversely increases.
  • Continue the iteration process over time, using increasing productivity to absorb staff turnover. While this lengthens the time required to achieve full ROI, it vastly increases the likelihood that ROI targets are actually realized. Note too that this approach belies the common belief that bots require little to no support once deployed. Effective bots should be constantly evolving rather than static. Contrary to popular mythology, a static bot is often a failed bot, but constant evolution is empirical evidence of bots making a difference to your business.


In call center automation, slow and steady is a far more effective approach than a big bang. While the timeline for achieving ROI may be longer, the likelihood of actually realizing ROI is far greater. Recognizing that your existing workforce may feel threatened, rather than liberated, by RPA is critical to organizational adoption and must be proactively managed from the beginning. A key step to achieving this goal is seeking symbiosis — rather than substitution — between humans and bots. Performed effectively, RPA adoption might help your customers discover that when they phone you, they’re phoning home.

Featured Content

Latest News


Are you ready to take the first step and learn more about RPA?

Contact Us