Automation is nothing new, but RPA requires an entirely different mindset.

Automation has been used to reduce labor costs and speed time-to-market for over a hundred years. But unlike most forms – think assembly lines for car production, which were first put into use by Ford Motor Company in 1913 – robotic process automation performs tasks in the background, out of our view, and many of our decisions are made on the foundation of the software “bots” outputs. 

There are certain degrees of risk with any type of automation. One unique to RPA is that of errors in the RPA algorithms themselves. A 24/7 bot works much faster than its human equivalent, and without someone—someone human—checking its work, there could be dramatic and catastrophic results. 

Imagine a scenario in which a bot is tasked with performing the calculations required to determine whether or not to offer a loan or mortgage to an applicant. Now imagine that due to a single small algorithm error, the bot accidentally says “yes” to one in a hundred applicants who should receive a “no.”

While that’s a failure rate of only 1 percent (and a success rate of 99 percent, of course) that bot is also able to process thousands of applications per minute. Thus, after only a few minutes, hundreds of high-risk loans might be approved. Just think of the enormous impact a hundred bots using the same faulty algorithm would have on a financial institution.

There are many ways to mitigate that risk. The obvious is to have a human checking the work of a bot. Even though that would require human intervention, only a small sample would need to be checked to determine if there is an issue. The most future-forward way is to have another layer of RPA on top of (and wholly separate from) the implementation that approves or denies the loan application. We may not be quite there yet, but we soon will be.

The best way, however, lies in how those bots are rolled out to begin with. Just because you have the capability to roll out hundreds of bots doesn’t mean you should do it right off the bat.

We recommend implementing small pilot projects—20 bots, for example, for customer service. After a brief period of testing, measuring customer service stats and processing metrics, you’ll be able to determine if tweaks are required, or that your initial deployment of bots is delivering to expectations, and that you’re ready to roll out increasingly more until you have a full RPA implementation.

It’s true that RPA can be a game-changer in terms of cost reduction, enhanced efficiency, and increased productivity. But any RPA implementation has to be a full business operations transformation, and that needs to be designed and directed by a human hand.