Summary: Let’s start by clarifying the difference between RPA (Robotic Process Automation) and IA (Intelligent Automation). Then we’ll show why AI/ML inside Intelligent Automation is the secret sauce that really makes this work.
As we’ve tracked the adoption of AI/ML across industries the application that always seems to lead is RPA (Robotic Process Automation), sometimes also referred to as IA (Intelligent Automation). The problem with giving credit for AI/ML adoption to RPA is that RPA historically has meant robotic process simplification. That is, processes were indeed automated but it was all rules based; no real AI/ML in sight.
As frequently happens in our profession it’s a problem of definition and naming. If you built your platform brand on the “RPA” name and added AI/ML features you’re hesitant to let go of the name. If you’re a new entrant to this platform war and you started with AI/ML capabilities you differentiated and rebranded by labeling yourself “IA – Intelligent Automation”.
So for the sake of clarity let’s agree that Intelligent Automation (IA) rests on the foundation of RPA to which real AI/ML features like computer vision, NLP, and predictive analytics have been added.
Of course just because we say so doesn’t make the confusing naming of smart RPA go away. There are still review and rating agencies wanting to call this “RPA 4.0”. You’ll just have to be on the lookout to be sure there’s real AI/ML inside and not just a label change.
Conversely, platforms that label themselves IA, Intelligent Automation most likely do have AI/ML inside, but caveat emptor, do your homework.
This graphic from a 2017 Everest Group report does a good job of contrasting the difference (although here Everest labels IA as artificial intelligence (AI)).
So Why Is IA so Popular?
IA platforms are a relatively simple and organized way for companies to access the power of AI/ML without getting lost in the technical weeds while keeping their eye on the prize which is cost reduction, quality improvement, and better customer experience. And they do this by hiding the AI/ML as utilities within the platform.
Mostly, this also means that your platform vendor is looking out for and constantly improving the AI/ML components so your company’s improvements don’t require much attention from your in-house data science team.
Projects using IA platforms also rely on teams of mixed skill business users with some participation from your data science team. The blended skills implementation team is by now a well-recognized best practice for implementing AI/ML.
And finally, IA Platforms generally encourage projects that are small bites rather than trying to take on huge mission critical processes in a single project. That latter approach might have been popular in the more Wild West days of AI/ML adoption when everyone was focusing on home runs. But it’s much easier to manage for project risk and for ROI when the goals are somewhat constrained.
So What Do Typical IA Projects Look Like
Since we are talking about reasonably constrained processes you’re going to see a lot of examples in other industries that may equally apply to yours. For example, everyone takes in invoices or orders or inquiries from customers. You might call these ‘content heavy’ processes where the action and process output are reasonably simple but have required a human to evaluate the ‘input content’ in order to create the correct ‘output action’.
What should be evident here is that the input is unstructured data, in the form of text, voice, or image and that for IA to work that input must be captured digitally. Once they are, then NLP and image processing AI/ML techniques can be applied directly. And if the outcomes are more probability driven than rules driven, then predictive analytics can be used to promote the best alternative among several options.
In the Front Office
A combination of computer vision and NLP can read, interpret, and extract key data from text, emails, letters, and images. Handling incoming insurance claim forms with accompanying photos, or collecting supporting docs for mortgage or bank loans are also good examples.
The first line of customer support is increasingly becoming chatbots to handle customer enquiries based on voice, text, or even image input. This same first line support strategy could just as easily apply to employee queries of IT or HR.
Predictive models can guide ‘best answer’ responses to customer and even internal questions.
Back Office
Any process that involves documents or even electronic correspondence from outside the company, for example invoices, that might typically be scanned is an effective target. Machine vision is much more effective that previous OCR technology and the IA process can bring together different elements of unstructured data to determine which method of further processing is most efficient.
In financial processes that are highly regulated, input data must be cross checked among documents and also evaluated for compliance to regulations which can now be automated. Accuracy, error, and fraud checks can be built into to process. Predictive analytics for example is at the core of identifying potentially fraudulent events and referring them to appropriate experts for further examination.
Collecting information from hundreds of documents across the company and in different formats, for example to assemble reports or narratives and table for financial reports can be automated. Law firms and legal departments are finding automated discovery to be a huge time saver.
Projects to improve the quality and decrease the cost of these processes are common to most businesses. Companies that are most well advanced in their ‘digital journey’ are using Intelligent Automation (with AI/ML inside) to capture these savings.
Other articles by Bill Vorhies.
About the author: Bill is Contributing Editor for Data Science Central. Bill is also President & Chief Data Scientist at Data-Magnum and has practiced as a data scientist since 2001. His articles have been read more than 2.5 million times.
Bill@DataScienceCentral.com or Bill@Data-Magnum.com
http://www.datasciencecentral.com/xn/detail/6448529:BlogPost:1001917
