Homer Simpson ran for Sanitation Commissioner on a platform of, “Can’t someone else do it?” and he won. People love not having to do stuff. But automation is not always as instantly sound in practice as in theory. In an article for InformationWeek, Mark Dickson outlines why the return on investment (ROI) on automation will be underwhelming if you are not careful.
History Is the Best Teacher
Dickson uses an unusual but powerful example to deflate our automation ROI expectations: barcodes. For as much of a convenient godsend as UPCs and barcodes were, he cites the statistic that they only actually produced labor savings of 4.5 percent. Granted, 4.5 percent at scale is still a huge number, but still. There were several demonstrable reasons for the marginal increase, which Dickson too outlines. There is likewise reason to believe that many factors, some seen and others not so obvious, will affect the ROI of your automation efforts.
Dickson shares another example, this time from Netflix, who “offered a million-dollar prize to the first team that improved the accuracy of Netflix’s recommendation engine by 10%.” They were at first highly impressed with teams’ efforts to meet the goal, but there were ups and downs to the race:
Simple models were yielding significant advances — using just one algorithm, singular value decomposition (SVD), produced a 4 or 5% gain all by itself. As the first year of competition wore on, teams further improved their results, but the models became more complicated. The best team stood at 8.26% improvement after one year, but it would take two more years of collective work before a team finally could crack the 10% threshold.
Netflix awarded the prize, but the winning team’s resulting computer program is still just an academic curiosity. The winning team had to combine the output of dozens of algorithms to reach the goal, which would have been unwieldy to implement in production. Complexity can lose the race.
Thus, you must be careful not to automate for the sake of automation either. Dickson further concludes that an element of humanity and practicality must remain in all things that you seek to automate. People are inevitably turned off by processes that feel too cold and impersonal, such as in automated help desk settings.
Accounting for obscure variables, picking the right reasons to automate, and keeping humanity in the automation are not easy processes individually, let alone altogether. But you have to make it work. You can view the original article here: https://www.informationweek.com/a-contrarian-view-on-automation/a/d-id/1329700
