phonlamaiphoto - stock.adobe.com
If Forrester's Craig Le Clair is right, invisible robots are poised to have a far bigger impact on jobs and automation than real physical robots. His group predicts they will automate well over $100 billion in work in the next couple of years. In the meantime, these invisible robots are shaping an automation revolution that could be more impactful than the Industrial Revolution with new kinds of winners, losers and lots of unexpected consequences.
But what exactly are these invisible robots? One of the most prominent examples are the bots created by a new breed of software called robotic process automation (RPA). On the surface, these RPA bots just copy and paste things in, well, a robotic way, which is boring stuff. But their rote work is having major consequences as it opens the gates for business leaders to think about new ways to automate processes.
The next Salesforce?
After I attended a media day hosted by the RPA vendor Automation Anywhere, I was curious to see what an invisible robot factory might look like. To be honest, it did not look like much. The headquarters was a couple of suites in a much larger complex of one-story office buildings spread out across the outskirts of San Jose. It looked like people had just moved in, and indeed the offices had recently spilled over to an adjacent suite -- evidence that this $6.8 billion startup is in one of the fastest growing industries in software.
Some RPA insiders, like Phil Fersht, founder of Horses for Sources, thinks robotic process automation is overhyped and overvalued. But I keep wondering if perhaps this is just the beginning of something much bigger, with companies like Automation Anywhere or one of its competitors becoming the next Salesforce one day.
It is plausible since RPA shows a path for integrating AI and automation into business processes precisely because the invisible RPA bots allow messy things to work together.
The value of simplicity
Where it gets interesting is that these RPA bots are basically building the infrastructure for all the other pieces to fit together such as AI, CRM, ERP and even documents. They believe in the long-heralded walled-garden approach in which enterprises choose one best-of-breed infrastructure platform like Salesforce, SAP or Oracle and build everything on top of that.
History has shown that messy sometimes makes more sense. The internet did not develop from something clean and organized -- it flourished on top of TCP: a messy, inefficient and bloated protocol. Indeed, back in the early days of the internet, telecom engineers were working on an organized protocol stack called open systems interconnection that was engineered to be highly efficient. But then TCP came along as the inelegant alternative that happened to work and, more important, made it possible to add new devices that no one had planned on in the beginning.
Completing the automation loop
Now for some practical stuff: The big news I heard at the Automation Anywhere event was the announcement of its process mining tool, Discovery Bot. Process mining tools pursue different approaches for deducing the recipes and variations that organizations follow to run their business. It's like reverse engineering a recipe from an ingredient list on a package -- you might get an overall idea of how to make something from reading the ingredient list, but it may not taste just right.
One approach to process mining looks at the logs generated on the back ends of applications like ERP systems to deduce the steps in a business process. Another approach looks over people's shoulders as they do things, which is what Automation Anywhere does.
But the real magic is that Discovery Bot automatically generates a prototype bot based on what it sees to create an instant but fragile replica: It will run but may break easily. However, this provides the template that an expert could turn into something that is more resilient. It's not just reformulating the recipe from the ingredients, but rather asking people how it compares to the original recipe. Over time, this will help generate something that is closer to the original process.
Also, the tools can estimate how much effort will be required to build this more resilient bot and how much it will save the company. This way, an automation team can prioritize which bots to build first.
In 2019, Automation Anywhere's competitor UiPath bought a process mining company for tracking what the company does, and a portal for prioritizing which bots to build first. They aren't automatically generating the prototype bots just yet.
Navigating the RPA bot programming hype
One of the promises of these new tools is that they let anyone build a bot -- but it's not quite as straightforward as the hype may lead you to imagine. The Automation Anywhere training team took a group of media people through the paces of building a simple RPA bot to copy data from different styles of invoices into a spreadsheet. Surprisingly, only two out of 10 of us were able to build a bot that was 100% accurate.
In retrospect, I spent a lot of effort trying to sort out subtle invoice formatting problems, such as a dash in the invoice number from one vendor's document caused by the way the program interpreted the text box layout. The key to making these work well requires an attention to detail. It would most likely work well for people who deal with numbers and are used to double checking that they add up correctly. But as our cohort clearly showed, not everyone demonstrates the required level of nitpicking.
Adding intelligence to documents
The exercise also demonstrated the mechanics of intelligent document processing. Whereas RPA automates the cutting and pasting of data between applications, intelligent document processing does this for documents. This is important because every company has a slight variation on the way they format common documents, implement things like invoice numbers or describe their terms. Intelligent document processing provides a way to map these variations consistently.
Enterprises have traditionally used optical character recognition (OCR) technology to capture rote text from documents. Intelligent document processing takes this to the next level by interpreting the context of data within the captured text.
Harvey Spencer, founder of HSA Associates, estimated that the market for OCR was about $4 billion in 2018, but intelligent document processing could be worth about $30 billion and is growing at about 80% per year. This is already much bigger than RPA -- perhaps because businesses have a greater need to process documents than automate apps.
HSA calls the series of services associated Capture 2.0. Much like intelligent processing automation
(IPA) adds AI to RPA, Spencer envisions Capture 2.0 working with a series of microservices for automated understanding, classification and extraction of corrected valid data. Different algorithms might be employed that perform the equivalent of double key entry where 100% accuracy is required.
Paving the last mile for AI
A key promise of RPA bots lies in paving the last mile of infrastructure required to deploy AI. There are two aspects to this: One lies in providing a framework for plugging AI modules or skills into existing business processes -- the second is to make it easy to automate or enhance decision-making. In the first case, this would make it easier to drop in better AI skills for things like fraud detection, reading documents or estimating the cost to fix a damaged car from a picture. In the second case, decision engines would extend the capabilities and applicability of business rules management systems that enterprises use today.
All the major RPA vendors have rolled out an AI infrastructure piece and developed marketplaces that sell dozens of AI modules. These are also being embraced by large AI vendors to help make their tools more useful to the enterprise. For example, Automation Anywhere is working with Microsoft and Google to help sell its AI tools and services into enterprises. The RPA infrastructure makes it easy to drop in its AI modules or services without having to do a lot of technical work to fit these into existing applications.
The decision orchestration piece is still rather young, and for the moment, a key aspect of this is support for automated machine learning, which can automatically generate the best AI component for a given type of decision.
Only time will tell if RPA is but a fad or represents a genuine shift in how businesses think about automation. There is an effort underway to eliminate the idea of robots entirely through IPA or even hyperautomation. After all, the tech industry is constantly debating what to call the next big thing. But in the meantime, RPA will enable the invisible robots to make visible the kinds of things people are already doing and help CIOs to build a better process.