How to Make Robots Intelligent?

As many people are aware, Robotic Process Automation (RPA) is a wonderful tool for automating rules-based processes. But, despite the cleverness of the software, those aren’t intelligent robots yet. Effectively, they are dumb. By that we mean that they will do exactly what you tell them to do, again and again and again. Which is perfect when you want 100% compliance and all your processes are rules-based and repeatable. But quite often this is not the case, and those processes will require complex decisions to be made or have data inputs that are unstructured or particularly variable. At this point intelligent robots are necessary. In order to achieve this, adding other technologies for some help can be very effective.

Much has been made of the evolution of RPA to include ‘intelligence’, and the current trend for ‘Intelligent Automation’ is driven by the need to extend the capabilities of robots to cope with the common process challenges of ambiguity and complexity. Now, we all know that none of these technologies are truly intelligent (in the sense that the human brain is) but the addition of certain types of Artificial Intelligence (AI) into the mix of automated tools means that the robots can now cope with many of these challenges.

When to use intelligent Robots?

One of the most common uses of AI to support Robotic Process Automation is to manage the diverse range of documents and data that often make up the inputs to many processes. Robots need structured and consistent data if they are to function properly, so when the incoming data looks slightly different each time and/or includes natural language, then AI can be brought in to cope with those challenges.

For example, if the robots are expected to process invoices, then those invoices are likely to be highly variable: they may have the address in the top right-hand corner or on the top left; the date may be in the UK or US format; there may be a VAT line or not; and a long etc. In the past, the way to cope with this variant would be to create a different template for every different type of invoice, and this would quickly become unwieldy and difficult to manage. The same would be true for Remittance Advices, Good Received Notes, First Notice Of Loss forms – basically any document that could be sent in from multiple sources, and therefore outside of the control of the receiving company.

Here at Roboyo, we have been working with a leading German telecoms company to help them test a solution that would automatically read the information of incoming forms, in this case proposals, and automatically enter it into their system-of-record so that it could subsequently be processed by robots.

We used ABBYY’s intelligent character recognition system for the Proof of Concept. ABBYY is an established ICR software solution that integrates well with a number of RPA vendors, including UIPath with whom they have a strategic partnership. We were provided with a relatively small number of documents, just 9, to ‘train’ the system on, and then a further 10 to test how well it had learnt.

Intelligent Robots learn fast

We found that, even with such a small training set, the software was able to extract the relevant information with a high degree of accuracy. As you might expect, this was particularly true, where the test proposals were from the same vendor as the proposals it had been trained on. But even with different vendors, where the format of the proposal was different, the software still performed well.

The clever thing about ICR systems is that they can learn on the job. So, when the system didn’t initially recognise a piece of data, a human would record that data instead. Then, with the next example document it would then be able to find and recognise that data automatically. The human interaction is highly assisted, so it can be done quickly and efficiently. Furthermore, it didn’t need help from an AI expert: these systems can be trained by anyone familiar with the process.

Benefits of implementing intelligent Robots

The telecoms company will be able to benefit significantly from this addition of AI capability. Firstly, it has opened up new processes that can be automated through RPA, and many of these processes can now be automated end-to-end. Secondly, there is minimum manual effort involved in extracting the data from the documents, which also means it can be done much quicker, and, potentially, 24 hours a day.

So, providing robots with a degree of intelligence, even if it is artificial, is certainly a way to maximise investments in RPA, as well as delivering benefits in their own right. If you want to know more about how AI can super-charge your robots, please do get in touch.


Tags: Abby, Artificial intelligence