Top Tips on choosing Intelligent Automation based on "fit for purpose"
How to choose the right Intelligent Automation tool for your Shared Services operations?
To evaluate a good fit, the important thing is to understand the difference in solutions. A fully autonomous intelligent system may not be the answer to every problem and it is certainly not the entry point for most enterprises. The solution is to design a system that solves the problem at hand, while taking into account the type of process, its maturity, dynamics, trust, existing skills, maintenance requirements, and data sets.
These nuances are getting lost in many of today’s conversations, however, as practitioners fixate on one solution or the other. And despite the current feeding frenzy, many consultants predict that, within a few years, RPA will fall back into the overall enterprise landscape as “just one of many” tools that support intelligent automation.
The first thing to consider, then, when evaluating IA's potential, is "purpose". This means identifying the objectives or desired outputs for which intelligent automation may be a solution. Identifying the problem, analyzing the ‘as is’ steps, recognizing where human intervention is creating blocks, and rethinking process flow … those are the logical starting points. Only then can you begin to consider which automation tool will deliver the right solution. Like many new technologies, it is best to start simply, learn, and progress to more complex use cases and processes.
Another vital point of differentiation is what a given type of automation technology’s capability, or purpose is. There is a general confusion around this that should be clarified.
There are two fundamentally different families of automation that can be described as follows:
Process specific automation tooling
- These tools provide automation in a very specific process area or knowledge domain. A good example here is optical character recognition/image recognition (OCR/IR). These tools are absolutely automation, but are more typically an “IT” solution and perform a specifically defined function that is typically not highly configurable by the user.
- RPA: These are server based, unattended process agnostic automation platforms that are configurable by the business user and can leverage the human interface for access to enterprise systems. These are best fit for processes where transactions have low levels of variation, business rules are easily encoded, and straight through processing output for a given population of transactions is 80% or better.
- RDA: This is an agent running in cooperation with, and simultaneous to, other desktop executables. The key difference to RPA is that RDA tools can be configured to stop and pause for cognitive contribution by the operator. They can also be configured to pre-fetch the decision support knowledge assets that are typically relegated to an operator’s domain. For processes with high levels of variation, where business rules are complex to code, and which don’t easily fit into an “if-then-else” scenario, RDA tools are a good fit.
Many RDA solutions emerged in customer engagement settings, where the need to enquire with users for additional or new information sources were the initial challenges the tools were built for. Now they often appeal to processes like purchasing, when requisitions are out of stock, to make alternative arrangements considering approvals necessary, cost difference etc. – all conditional steps typically provided by a human agent. These kinds of knowledge assets can be pre-fetched and presented to an operator for appropriate decision-making within seconds.
Inline Prescriptive Analytics
Analytics are key in terms of their impact on better decision making, particularly within the context of intelligent automation. The continuum of analytics, from least to most complex are:
- Discovery – What happened?
- Descriptive – Why did it happen?
- Predictive – What is likely to happen?
- Prescriptive – What should be done?
- Deductive – What would happen if?
The continuum is presented here with Prescriptive bolded due to two factors: its ability to provide process orchestration in a production environment; and the simple fact that machine learning and computational horsepower have made it a reality today.
Typical business analytics present a “rear view mirror” analysis. Predictive and Prescriptive offer a “windshield” view.
The ability to deliver near real-time (within 3-15 seconds) prescriptive analytics mean that they can be used in place of human judgment and decisioning. Prescriptive analytics tools require enormous amounts of curated data, however, and that is currently the limitation. Taking as an example a purchasing agent determining the right course of action where an item is out of stock, advanced analytics tools could ensure an automatic threshold of 95% confidence – based on characteristics and similarities to previous events – in determining that the next best step would be to substitute item x. This activity requires no pause for cognitive contribution by an operator, as steps are automatically taken. As each occurrence is either identified as correct or incorrect, the system gets smarter and more confident. When in-line prescriptive analytics are combined with an automation capability (whether process specific or agnostic) a cognitive automation is possible.
Considered broadly, these steps are more often than not sequential and complement each other: process specific automation provides not just specific outcomes but also the ability to observe a previously analog task in a digital fashion, and provides the “digital exhaust” for advanced analytics and machine learning. Robotic process automation combines easily described business rules, as provided by a rules engine, and the ability to leverage somewhat static decision criteria, to interrogate other systems real-time and pull out the condition criteria necessary to proceed with straight through processing on a highly confident basis. Robotic desktop automation tools apply where it is not so clear what the business rules are and there is a high level of interdependence or entanglement between business rules, which make it difficult to encode these into a rules engine. They provide an opportunity to concentrate the cognitive load of a processor by considering which information would be needed by an operator at a given ‘fork in the road’, to determine the next best step, and automate this activity. Finally, advanced prescriptive analytics help to fill in the gaps in both RPA and RDA processes to improve straight through processing for an ever greater population of transactions.
This article is an extract from SSON’s Intelligent Automation Global Market Report 2017 H1.
Download the full report here.
[The H2 version of this report will be published in December 2017]