The Deception of Full Automation

Oct 27~28, 2021

Automation is an idea that some people either against it because they fear it’s going to take away their jobs, or support it to the point that they want to “automate everything”. Both are incorrect and shows how ignorant those people are, however, the latter is even more so and could be harmful to the organization.

First of all, it’s pointless to say automate everything, or full automation, or 100% automation, if you don’t provide a clear definition of everything, full, 100% first. Take invoice processing for example, which is a common RPA practice. Most solution starts from reading data from a spreadsheet, then navigate to certain application to perform tasks such as validate invoice, match invoice information against data in other systems, record invoice as well as supplementary data in accounting system, etc. Without a clear definition of Full Automation, we need to answer the following questions first:

1. How do we start the automation? Manually by double clicking on an icon on your desktop, or it’s scheduled to run automatically whenever there is available data?
2. Where is the data in the initial spreadsheet from? If it’s prepared manually, can we automate that too?
3. Once the automation program started, is it really bulletproof to the point that no human intervention is ever required?
4. Is the last step really the last step, can we extend the automation further to include monthly consolidation, general ledger posting, etc.?

Secondly, even though I was opposing “cost-benefit analysis”, but this process is unavoidable in making automation decisions. If we want to automate the above four steps, we need to first evaluate feasibility and necessity, then cost and benefit if we answered yes to both feasibility and necessity question.

Thirdly, it’s harmful to talk about “full automation” under an ambiguous concept, especially in front of those people who are already concerned that automation will take away their jobs.

Number four, it’s possible to automate human decisions as long as those decisions can be modeled into decision trees, however, in real life, it's either too difficult to do so, or too volatile to maintain. Maintenance is also part of the manual work that often neglected by “full automation” proponents.

Finally, it’s unfair to enforce so-called “automation percentage” KPI to teams such as Finance Department. Not only achieving “100% automation” would mean disposal of the entire finance team, if it’s even remotely feasible, but also it makes no sense by talking about automation percentage of finance when it heavily relies on the data generated by front-end, middle-tier and back-end business divisions, especially when those data are often of great volume, sometimes poor data quality, and not yet conciliated with the standards of financial statements.

The point is, rather than talk about full automation, 100% automation, or automate everything, let’s be grounded and take serious look at the process that we want to automate, or have already partially automated, and find out where we can bring the greatest value to the processing by either automation, or improving the current manual process by many traditional techniques, such as train our people properly, remove bottlenecks, etc.