The black box of AI is a thing of mystery and wonder to most people who don’t understand it, and they’re happy to leave it that way. But when a sizeable percentage of your task force is replaced by automated process bots, and most of those bots are there to improve the experience of your customers, the potential impact on client satisfaction and revenue is huge.

So while most people are still talking about the benefits of automation, more time should be spent considering the risks of leaving it unmonitored and unchecked – which we discovered when called in for a recent client project.


The right tool for the jobs

We spent three months with the RPA team of a company who’d aggressively scaled up their automation in the last year. Yet they’d not considered the governance of their new robot workforce. There aren’t any management books written about this new workforce, of course – no KPIs to motivate them and no organizational psychologist to understand why they’re not delivering. They work, or they fail. And when they fail they often do so unnoticed, causing a huge impact on customer experience and potentially the bottom line. The only way the company could tell when a robot slowed or ground to halt was by inspecting the logs through SQL queries. This was often done too late, invalidating SLA-style requirements.
So we built them a tool that could be used by both the local and offshore teams to monitor the company’s robotic processes and better determine problems in real time and, in doing so, learned several things companies should consider when implementing RPA.


Bots can’t think on their feet

Well, they don’t have feet. But you know what we mean. If the script isn’t written (or trained) to handle all the situations a human may encounter, it needs to be passed back to a human to complete. Accounting for all possibilities is a mammoth task; they’re not always obvious and the person doing the scripting needs skill and experience of the systems being interfaced. Anything unpredictable in an interface needs special attention. This could be a dialog box taking longer to appear, or displaying twice intermittently.
Often this special attention is scripted as a retry or deliberate slowdown, so a case takes longer compared to a human and with upgrades and changes to systems frequently happening, the script needs to change just as frequently.


Avoid vendor lock-in

Another problem to avoid is vendor lock-in. Some robotic applications use a proprietary scripting and toolset where simple development tasks such as peer review can only be done through the vendor’s product.


Seek to refine rather than replicate

The biggest issue we’ve seen is that a robot may implement a human process without any effort to improve or replace that process. Does a letter really need to be sent out? Why can’t we replace those two old systems with a single more efficient one? Ironically, RPA may actually impede change, or lead to higher IT costs associated with maintenance and disruption if the change is forced.
Robotic automation seems an easy and attractive way to reduce the cost of IT. But if the end beneficiaries are your customers and you’re not keeping your AI underlings in check, you could be doing more harm than good.