BLOG

RPA: Moving Rapidly Past Mere Automation

opt_wp_admin

Author: Sekhar Prakash, Head of Delivery and IT Infrastructure

Ask not what RPA can do for you… Ask what you can do with RPA!

In an office not too far away from you, a manager queried his company’s RPA: “What will be my operational cost for this quarter?”

A few seconds later, RPA said breathlessly, “With a Level 4 storm expected to hit the coast in September, the load on the networks is likely to increase from the average of 78% to 163% for four days. With generator runtime of 18% and utility supply at 82%, operational cost for the territory will be $28.1 Mn, a 22% growth QoQ and 36% from the same quarter last year.”

If that sounds like Jarvis that only Iron Man can have, you may not be familiar with the rapid evolution of robotic process automation  (RPAs).

RPAs have indeed come a long way. As Bill Lydon, editor of Automation.com, put it, the rules of enterprise architecture have been rewritten in the past few years. Once, for a long while – decades and decades, in fact – automation meant plant-level reconfiguration. Automation meant putting a timer on an on-off switch; or using a sensor to stop a machine from overheating; or robotic arms that installed dashboards and steering wheels. The term robotics conjured up mechanical limbs and the whirr of motors that moved them.

In retrospect, that’s a very narrow idea of what robots can be. There are other kinds as well, bots without limbs or motors but go about their tasks quietly and efficiently anyway.

Bots made of lines and lines of code.

RPA got Bots

RPAs are software systems made of bots, each of which can fulfil a specific task. A group of tasks, acted upon in concert, complete a process. The first-gen RPAs typically automate processes that involve a lot of repetitive, low-intelligence tasks where there is a high chance of error (typically due to monotony or fatigue). In fact, this is still the primary function of RPAs. But recent strides in artificial intelligence, machine learning and shared infrastructure has let us create more intelligent RPAs. Systems that don’t just read, but can also think and help you make better decisions.

The range of abilities of an RPA depend on the bots that it can press into service. The earlier RPAs consisted of simple bots who could scrape screens, scan characters, run reports, match entries in different databases and reconcile data from different formats. They did what they had been programmed to do, nothing more, nothing less, and what they were programmed for were only such tasks.

In fact, the way RPAs have become more complex bears a lot of similarities to the evolution of mankind’s collective intellect. The earliest man, for instance, first learnt to stand on two feet, then run. Then he learnt to use his fingers to fashion crude weapons, and out of those crude weapons came better ones. As his understanding of physics, chemistry and metallurgy grew, he developed even finer tools. In the past 150 years, we have gone from dismissing the idea of flight for mankind as impossible, to sending manned missions to other planets.

Getting better everyday

A McKinsey report in 2017 warned organizations about the risks of adopting RPA, when it was crude and limited. That is no more the case. In the last year, the era of hype, which typically accompanies any paradigm shift, is over because RPAs are now showing results. A conservative estimate is that at least 4 out of ten organizations will have switched to RPAs by 2020.

The very structure of RPAs, the modular skeleton of bots, is responsible for how quickly RPAs are becoming essential to today’s organizations. RPAs add value by being intelligent. The tremendous strides in artificial intelligence, natural language understanding and machine learning has led to bots who can colour outside the dotted lines and make sense of unstructured contexts just as easily and accurately as in pre-defined cases. Bots can be trained to deal with unforeseen circumstances and variations, and take decisions on their own to navigate around such issues.

For instance, making sense of free-form descriptive text has always needed a human mind at work to separate the irrelevant from the relevant. With an RPA, however, a bot can use NLP to scan hundreds of thousands of entries and highlight common themes that can then be addressed proactively. You could even extend the functionality further and use the RPA to trigger these corrective actions without human agency.

Learning to get along

At another level, this capacity to learn will also reduce incompatibility with existing systems and the need to customize an implementation.

In fact, that’s where earlier (what we’d like to call premature) implementations of RPAs probably failed. In many cases, bots failed to work together to make sense of a collective system of data. In others, bots failed to adapt to upgrades to existing systems. They could not integrate.

But with self-learning bots built into purpose-built RPAs, life becomes so much easier for in-house administrators. For one, there is a fair amount of standardisation and commonality in the data, structure and applications employed in an industry. The RPA then adapts to specific changes it encounters, keeping its core intact, and then keeps learning from every byte of data passing through it. The day is not far off when the example we started this article with is the norm, not the exception.