Cognitive Automation: The Next Panacea or Evil Job Bandit?

Published July 16, 2019

Become A Contributor

Written by: Greg Council
Author Image

Greg Council

Greg Council is Vice President of Product Management at Parascript, responsible for market vision and product strategy. Greg has over 20 years of experience in solution development and marketing within the information management market. This includes search, content management and data capture for both on premise solutions and SaaS. To contact Greg and Parascript, please email: info@parascript.com.

Read More
While it’s impossible to predict exactly what the future holds as more enterprises and service providers adopt cognitive automation to improve many of their manual processes, reading the tea leaves—or better yet—looking at actual early use cases suggests it’s both: job thievery and, if not a panacea, some fairly drastic improvements in efficiency and quality results. 
 
Automation: Levels of Impact  
 
Let’s first take a look at the automation landscape as a whole to understand the level of impact based upon types of tasks involved and the automation used. Harvard Business Review (HBR) provides a useful summary article that explains how to deconstruct work into tasks that can be automated. In the article, the authors describe assessing work by three characteristics: repetitive vs. variable work, dependent vs. interdependent work, as well as physical work vs. mental work.  
 
There is also another set of constructs focused on the nature of inputs and outputs. It is quite easy to imagine a decision-making process that is complex, using data that is not. Take predicting the weather. Most data involved in weather forecasts is highly structured and very granular. However, the process of taking this data and using it to create reliable models is decidedly not straightforward.  
 
Conversely, processes exist that result in more-straightforward decisions, but using data that is diverse and highly variable. In the case presented by the HBR article, a credit analyst looks for common signals that can determine creditworthiness while the input data can be more complex, involving bank statements, credit rating reports and even social media.  
 
It is appropriate to use a model that takes into consideration the nature of inputs and outputs onto which we can overlay the three characteristics identified in the article. Using this approach, we can assess the impact of automation based upon the nature of work itself.  
 
Basic Automation with High-level Impact 
 
The most basic level of automation and arguably the one with the highest level of impact so far is applied to rote, highly repeatable and low variance tasks. Think of activities in the IT back-office such as regular back-ups of data, automated provisioning of software resources (such as email accounts and CRM access) and almost anything associated with cloud “DevOps.”  
 
These activities can be highly automated due to the nature of the work and low probability of exceptions to workflows. The inputs can be highly structured with very little variability while outputs are often binary; the result is either a successful completion or an exception. These tasks are very independent with interactions typically only with application interfaces. There is very little mental effort required.  
 
Simple scripts that implement rules, which govern how to execute a task, automate most of these types of tasks. As mundane as they are, prior to applying automation, human staff were required to perform these tasks. Therefore, the ability to achieve high-levels of unattended automation has had a significant impact on staff focused on these activities. However, are all activities just as exposed?  
 
The domain of rote tasks with limited input and output complexity represents the low-hanging fruit of automatable tasks. Few human-centric tasks have achieved unattended automation.  
 
Cognitive Automation and the Difference It Makes 
 
Automation, focused on more-complex processes, is typically the realm of “cognitive automation,” where more advanced machine learning is employed to supplement or replace the rules-based approach used to automate rote processes.  
 
Take, for example, customer service. Customer service involves interactions with customers and other staff internal to the organization. Chatbots are often presented as an example of complex activities where machine learning and natural language processing are put together to enable automation of certain areas of customer engagement whether it is to answer a question about a product, provide updates on account information, or to handle a support issue.  
 
The level of variance of input and output associated with these activities can be enormous so machine learning—with its ability to analyze massive data sets—is employed to solve the problem. However, can a chatbot replace a customer service or technical support staff? The answer, at least today, is decidedly “no.” The true effectiveness of a chatbot—to determine the customer’s mood, adjust the interaction and understand when to escalate to a higher authority—is limited. As a result, chatbots are often used as an assistive technology to enable support or service representatives to manage a larger workload queue and to focus on high-value activities.  
 
Unattended and Assistive Automation Trends 
 
Ultimately, the trend is toward automation at the individual task level especially where more-complex processes are involved since these represent the bulk of activities within an organization. Automation will be a mix of unattended and assistive depending upon the nature of the task. In the example of an accountant, we might see more unattended automation to handle input data in order to get it into a format that assistive processes can utilize. Here the accountant is directly involved in reviewing machine learning-based analysis to arrive at a conclusion. 
 
The Everest Group has identified the latter as an increasing trend among automation vendors that see the limitations, both functionally and market-wise, of pursuing a pure unattended automation goal.  
 
A recent Everest Group blog has identified examples of “one robot per employee” where the emphasis is on automating discrete tasks of knowledge workers instead of attempting to apply automation to an entire workflow. The article cites examples such as automatically changing passwords based upon IT policies and automated logging into multiple systems. Other automation efforts for discrete tasks include meeting-setting in which a virtual agent coordinates identification of different attendee availability.  
 
Another HBR article cites two different types of “collaboration” with AI. This includes humans assisting machines to handle exceptions and train the systems in the first place and machines assisting humans such as in computer aided design and prototyping. 
 
Cognitive Automation and Our Jobs 
 
Ultimately, the impact of cognitive automation to jobs—either positive or negative—is based upon several factors regarding the nature of the work. Even though there are foreboding articles and studies on the looming massive job loss due to automation, we are still far away from wholesale replacement of most jobs due to the complexity involved.  
 
The characteristics of the inputs and outputs of most processes along with the level of complexity of the discrete tasks necessary in a majority of work done today by humans makes replacement unfeasible. In addition, we must consider another critical factor: the higher the value or risk (financial, political, etc.) associated with a particular type of work, the less likely we are to see immediate replacement of workers in favor of machines. There are multiple reasons why a flight crew is still in the cockpit, for example, even though much of the actual work of flying a plane has been automated. 

You May Also Like…

Steven D. Levitt: it’s all about incentives

If you walk past a bookstore or watch 60 Minutes you have likely heard of the wonderful works by the academic economist Steven D. Levitt, Freakonomics and Superfreakonomics. Levitt teamed with his journalist collaborator Stephen...

SIG|ORG Spotlight Content