How RPA is evolving with AI: in five slides

About a month ago, I published an article on how to scale RPA in six slides; it garnered a lot of interest. Not, I hasten to add because of any writing skills on my part, but because it was considered to be concise and clear.

So, I'm back...

With the recent release of Gartner's first Magic Quadrant on RPA*, all 18 vendors covered are expecting an up-tick in business; that's the power of Gartner.

But let's face it, in terms of the core platforms and their ability to automate the boring, mundane and repetitive work that office employees hate; to allow these employees to focus on higher-value and more fulfilling tasks?, all 18 have this capability to a lesser or greater extent.

Also, given that the adoption of RPA in the Fortune 500 companies around the world is approaching 100%, (UiPath alone has 60% of this community as customers), what people are now interested in is how RPA can be combined with artificial intelligence (AI) to provide additional benefits to their automation programmes and help them on their digital transformation journeys - a journey, by the way, that has no end point.

The problem with AI is that it's complicated and it's fast-moving. Therefore, I've tried to simplify it into four areas. Each one is represented by an 'understanding' provided by AI tools that have be included with RPA to gain the maximum benefits from both types of technology.

I offer you five slides on what's important, and why.

Some of the earliest adopters of RPA were the outsourcing companies, as they saw that automation could reduce their cost to serve their customers in a way that labour arbitrage was increasingly failing to do. However, they had a problem: it was extremely unlikely that they had access to their clients' applications - the systems and technology the customers were using to run their business. Instead, the business process outsourcing (BPO) firms had to access their clients systems over (generally) Citrix.

The BPO firms could certainly use coding and macros to automate many of the tasks they'd been given by their customers but because all they had to work with was a bitmap - a picture - every time the customer updated, patched or otherwise changed their systems, it immediately broke every automation that the BPO company had built and they had to start again. This was time consuming and expensive.

Along comes RPA in 2015, or thereabouts, and these platforms used anchors and other elements to allow the automation to survive any changes made to the underlying systems so they didn't break every time.

Wind forward four years and now the best RPA platforms are using computer vision (an AI tool) so that the system understands every element on every screen in exactly the same way as a human does. This allows RPA customers (BPO providers as well) to create an automation safe in the knowledge that it doesn't matter what changes are made, the robot will be able to 'see' and 'understand' what it's looking at.

Paper will not die.

Every company is still deluged with documents, files, invoices, purchase orders, curricula vitae and other sundry bits of paper. The idea of a paperless office to most is a distant dream.

This reality has previously been addressed with scanning technology: after all, if you can digitise the information on paper, you can use a robot to handle it. No? No.

Scanned documents are only part of what is needed. In order for the system to 'understand' what it's looking at and then allocate that document to the right robot or person to handle, the system needs to use several AI capabilities: named entity recognition, sentiment analysis, intelligent optical character recognition, natural language understanding, translations, machine learning and so on.

RPA vendors have been working on the intelligent OCR technologies with commercial operations like Abbyy and are increasingly utilising the developments in the other areas created by the likes of Microsoft and Google. What's interesting about these latter two is that they have open sourced most of the elements required for document understanding. This means they are free to use and so eventually all software will use them.

For the RPA companies, it means that they're using the best possible technology to allow customers to handle the unstructured data sitting in their piles of paper.

Which processes should be automated? I'm in agreement with Automation Anywhere on this when they said that "any process that can be automated, will be." But where do you start?

Identifying which processes to automate first and the order in which to do them has heretofore been the preserve of the automation centre of excellence (COE) in combination with the subject matter experts in each business unit or process area. Now, RPA vendors are increasingly seeing the value of process mapping technology from the likes of Celonis and Minit, which Lean six sigma and process improvement teams use to identify process flows, bottlenecks, exceptions and so on. The purpose is to illustrate the optimum path§ through any process to maximise efficiency.

RPA vendors like UiPath are working on changing the output of the Celonis activity from a picture of a process map to a XAML script; a XAML script that becomes a robot. So in simple terms, what we're developing is self-building robots.

The system will watch what the human user does, identify the the optimum path where there is repetitive activity and then create a robot to do it; automatically. This is the ultimate endgame when it comes to ease of use for this type of technology.

In addition, there is a second part to process understanding: understanding what happens when (inevitably) processes change.

All processes change over time. Business rules change; technology gets patched or replaced, business priorities alter. This breaks robots. So RPA vendors are now looking at machine learning (ML) to capture any increase in exceptions - a sure sign that something has changed - and then the automation platform will reconfigure the robot to meet the new needs of the process. These are self-healing robots; and are still some way away.

Ultimately, robots will be controlled by voice.

Let's say that you work in a bank and your robot doesn't understand what to do with a particular document or payment; it would flag it up as an exception to the human user and today, either the user will deal with this manually or go back to the RPA developer to change the robot so that it knows what to do in future.

That's not what will happen soon.

If the system and robots have conversational understanding, you will be able to communicate with the robot in natural language: in the case of the bank employee, all they will have to say is "OK robot, if you see that sort of document, stick it in the Wells account." The robot, of course, has to understand that 'Wells' in this case means Wells Fargo; and not holes in the ground with water at the bottom or a town in Somerset in the UK (which are also both 'Wells').

Conversational understanding therefore needs the process or industry specific ontologies to allow NLP and natural language understanding to function. In UiPath's case, we're working with a number of vendors in this space including Kore.ai and Humley.

Slide 5 is different; it's not about an AI technology at all but the corollary of what happens when the previous four: visual, document, process and conversational understanding are fully implemented into RPA platforms. They will disappear.

It seems counter-intuitive that all of this work will lead to the disappearance of RPA and AI but I think this is exactly what's going to happen. However, it won't disappear because it's not getting used: it will disappear because it will be used everywhere!

Bill Gates in the early eighties envisioned a time when every household and every desk would have a computer. Today, we envision a time when every office worker will have a robot; to take over the work they don't want to do (the boring and repetitive stuff) and to help them to do the work they do want to do (augmenting them and increasing their efficiency - and happiness).

At this stage, RPA and AI will be considered as natural as working itself and will effectively become invisible.

So there we go, five slides that you can reference when someone asks you how RPA and AI are coming together. As usual, I'd really value your feedback in the comments section below.

* The Gartner RPA MQ is available free of charge from our site.

? The Forrester research study into RPA's impact on employee engagement and happiness is also free from UiPath.

§ The optimum path is sometimes called the 'happy path', which I hate. As one of our public sector customers said to me the other day, the idea of government having a happy path for notification of bereavement doesn't bear thinking about...