Featuring research by UNC Tax Research Scholar and UNC Kenan-Flagler Business School Professor of Accounting Eva Labro and her colleagues on predictive analytics.
The growing use of predictive analytics to analyze manufacturing plant data has increased factory efficiency while re-shaping the employment relationship among both managers and factory floor employees, according to a new study from Eva Labro, Professor of Management Accounting, UNC Kenan-Flagler Business School; Mark Lang, Professor of Accounting, UNC Kenan-Flagler Business School; and Jim Omartian, Assistant Professor of Accounting, University of Michigan’s Ross School of Business.
The research paper, Predictive Analytics and the Changing Manufacturing Employment Relationship: Plant Level Evidence from the Census Data, is the first to provide large-sample evidence of the impact of predictive data analytics on a representative sample of U.S. manufacturing plants.
In the study, the authors examined a sample of more than 25,000 plants using proprietary data from the U.S. Census for the years 2010 and 2015. They found that the use of predictive analytics has increased substantially over that period, with the greatest deployment in younger plants and professionally managed firms with more educated workforces operating in stable industries. Family-owned firms were found to rely less on predictive analytics, consistent with lower managerial sophistication and fewer resources. Where adopted, the technology is re-shaping the manufacturing landscape.
For the paper, the authors examines three primary questions:
The report notes that the rapid decline in the cost of sensors and machine-to-machine communication through the “internet of things,” along with the reduction in the price of storage, has led to an enormous growth in the volume of real-time data available to factory managers. This has fueled the expanded application – and the increased accuracy – of the predictive analytics tools.
By gathering and processing this real-time data plants are now better able to identify the likelihood of future outcomes based on historical data and to anticipate and optimize manufacturing cycle times, equipment and labor utilization, demand spikes, supply chain efficiency, product quality, and defect rates, among other factors. The result has been a significant improvement in operating efficiency, with performance targets that are more accurate, long-term oriented and linked to company-wide objectives.
Centralization wins out
Prior to the publication of the research, it was an open question as to how the increased application of predictive analytics would impact management decision making. On the one hand, the availability of detailed, real-time plant-level data argued for delegation to the local level where this information could be quickly put to work. On the other, because the information provided is quantitative and easily communicated to headquarters, it lends itself to a top-down approach.
In determining the level of delegation, the study looked at the plant’s discretionary authority over decisions like human resources (hiring, promotions and pay), marketing (advertising, pricing and new product introduction) and capital expenditures. In the event, the authors found that companies had in general used the predictive analytics data to take more control of these functions, further centralizing decision making and reducing delegation to plant managers, lowering the need for managers at the plant level.
Results for marketing provide support for the centralization approach, as predictive analytics allows headquarters to better forecast product demand and adjust advertising, pricing and product mix without relying on local expertise. A similar outcome was found for human resources, suggesting that analytics supported centralization by facilitating the ability to accurately predict local staffing needs and centralize hiring, promotion and pay.
“For headquarters, this data acts as a substitute for the private informational advantage of the plant manager, reducing the need to delegate decision making,” says Labro. “This creates significant efficiencies for the enterprise and improves business outcomes. It also has a profound impact on both compensation practices and the nature of the plant-level workforce.”
Impact on compensation
While the authors hypothesized that the centralization of decision-making would have a negative impact on performance-based compensation at the plant level, they found this not to be the case. Management did elect to use predictive analytics broadly in the design and implementation of performance targets, however, it resulted in the establishment of more rigorous, quantitative performance goals. The added forecasting ability supported by predictive analytics was also associated with a shift in plant level focus from short-term to long-term targets.
With these goals in place, performance incentives were strengthened as bonuses and career outcomes were tied more closely to meeting the targets. Promotions, too, were more closely linked to performance metrics. The study also identified a positive correlation between the analytics and the propensity to fire or reassign managers for underperformance, as well as a likelihood that the overall need for managers (but not staff) had been reduced.
“Predictive analytics gives the company better insight into performance and allows management to set more precise goals at the plant level. Incentives can be more fine-tuned,” says Labro. “At the same time, it also creates a closer link between underperformance and termination and accelerates the termination process.”
What it means for rank-and-file workers
The impact of the growing use of predictive analytics may be most keenly felt on the factory floor, as rank-and-file employees make up the vast majority of the manufacturing workforce and represent a significant component of the U.S. economy. While the study found no major impact on the size of the manufacturing workforce, it points out that the composition of that workforce may be changing.
The authors note that if the complexity associated with predictive analytics necessitates a more skilled workforce you would expect to see a greater proportion of full-time, specialized workers on set schedules that can be planned well ahead of time. However, they find that decisions directed by the analytics lead to more part-time workers being hired to fill in on specific tasks on an “as needed” basis, with many working irregular hours, not unlike the “gig economy” model prevalent in parts of the service industry.
“We’re seeing what you might call the ‘uberization’ of the manufacturing workforce as one result of the deployment of predictive analytics,” says Labro. “Rather than a full time, highly specialized workforce you see a movement towards labor that is increasingly commoditized, temporary, cross-trained and employed on a flexible schedule.”
Here to stay
The authors conclude that predictive analytics is an evolving force in manufacturing and is likely to grow in importance going forward as technology facilitates greater use of data for decision-making, optimization and performance management.
“A number of the conclusions from our research are counter-intuitive – in particular the compensation practices for plant management that has essentially seen much of its authority migrate to headquarters,” says Labro. “But most interesting is that given a choice between using data to support a distributed, localized approach or to consolidate control at the headquarters level, headquarter management chose centralization. So far this has worked out well from the perspective of the enterprise as efficiency improved. It will be interesting to see how it plays out over time.”
To view the paper in its entirety, visit the webpage.