Trust, talent and targets—executives must address all three challenges to exploit the full potential of machine-reengineering. To alleviate these challenges and gain the most benefit from machine-reengineering, our preliminary research in this area suggests a series of steps, tailored to the specific needs of individual organizations.
Overcome the trust paradox with experiments
More than ever, executives should create an environment that enables data-driven change. In such an analytical culture, employees should be expected to challenge, increase or decrease the influence of algorithms. Edmunds.com CEO Avi Steinlauf cautions: “You can’t just turn it all over to a machine that will do your thinking for you. Tools can help, but you need smart people thinking about things in curious and intuitive ways.”6
Half of our respondents’ organizations now use prescriptive or predictive analytics, driven by machine-learning algorithms. Some of these algorithms are undoubtedly more effective than others. Underperforming algorithms should be taken out of production, while high performers should be shared and adapted across silos. In our research, a third of respondents use machine learning in only one production system. Companies should consider applications for successful algorithms in other production systems.
Scout for newly necessary skills
Many leading learning researchers are pioneering the machine-reengineering skills that organizations will need to master tomorrow. For example, MIT’s Cynthia Rudin is exploring how to build simpler and easier-to-interpret machine-learning models—insights that employees will ultimately need to make algorithms easier to verify.7
Partnering with researchers through one-to-one arrangements or via associations and consortia will help executives develop a greater understanding of emerging technical skill needs. (Machine-learning vendors can also spot emerging machine-collaboration skill needs.) Any problems that arise in a workforce’s interactions with algorithmic processes also indicate new skills to develop.
Take human aversion to algorithms: Studies have shown that people are far less forgiving of forecasting algorithms that err than of erring humans. But researchers at the University of Pennsylvania have found that if people are permitted to modify algorithms even a little bit, they are likely to continue to use them. This finding points to a newly needed management skill: Sniffing out how, and how much, staff can modify algorithms without degrading their performance.8
Prioritize projects, optimize tactics
Our respondents identified several factors executives should consider when prioritizing machine-reengineering projects, including the potential financial impact and the speed of the process change. Also important? The learning delta: the difference between an algorithm’s and a human’s capacity to learn in the context of big data. What can a machine learn that can then help people improve performance?
For decision makers, there’s a trade-off—assessing the value created by machine-reengineering processes against the acceptable investment costs and areas where customer needs and competitive conditions require faster process change. Ultimately, executives must consider their goal: Is it process optimization or business model change? Or is it to test a process that would truly disrupt the industry? Our research has identified companies that have successfully used machine-reengineering to pursue all these goals—some by applying algorithms to data that previously collected data, others by creating new data sets and APIs (see figure below).