Jeff Bezos Speaks to Amazon’s Ergonomic Job Rotation Initiative





The perception of the value ergonomics brings to Corporate America’s bottom line just received a shot in the arm … by one of industry’s most innovative thinkers. On April 15, 2021 Jeff Bezos in his Letter to Shareholders provided an overview of Amazon’s 2021 plans to leverage job rotation in their operations. Bezos stated that he is an inventor; and that invention is what he does best, creating the most value for Amazon. He then outlined upcoming innovations in safety supporting the company’s initiative to be "the Earth’s Safest Place to Work". One of Bezos’s priorities for 2021 is to work alongside Ops to stubbornly and relentlessly invent the solutions needed to meet this objective.


Bezos noted that 40% of Amazon’s work-related injuries are MSDs (musculoskeletal disorders), and that these MSDs are most likely to occur during an employee’s first six months at Amazon. He stated the need to invent solutions to reduce MSDs for new employees, many of whom might be working in a manual labor job for the first time. Bezos mentioned the success of Amazon’s WorkingWell program which was launched at 350 sites across Europe and North America last year. He spoke to how coaching small groups of employees on proper body mechanics not only reduces workplace injuries but has a positive impact on employee’s lives outside of work.


Now at this point in the Letter to Shareholders, we were beginning to lose interest. As engineering-based ergonomists, we would rather identify root causes and develop & implement effective solutions than reactively teach employees how to better withstand the physical rigors of work (i.e., coaching to “keep the load close”, “bend at the knees when lifting”, etc.). But our interest picked up in the next paragraph when Bezos mentioned Amazon’s efforts to develop automated staffing schedules utilizing sophisticated algorithms. He said the objective is to rotate employees among jobs such that they use different muscle-tendon groups, thereby protecting employees from MSD risks. Bezos went on to state that this new technology is central to a job rotation program the company is rolling out throughout 2021. What new technology could Jeff Bezos be referring to?


We speculate that he is not talking about applying 15-25 year old single-task ergonomic evaluation models (i.e., STRAIN Index, NIOSH Lifting Guide, RULA, etc.). Nor do we believe he is talking about applying consultant-developed risk assessment tools with High, Moderate, and Low semi-quantitative input categories; and then simply rotating employees from jobs with high/red ratings into jobs with low/green or moderate/yellow ratings. Neither approach would comprise a “sophisticated algorithm” nor fit any reasonable definition of “new technology”. We speculate that Amazon’s 2021 job rotation program plan involves software algorithms and multi-task ergonomic evaluation models.


(Note that the information below is purely speculation, and not the stated views of Jeff Bezos or Amazon safety/ergonomics staff.)


Job rotation has recently received negative criticism from academic researchers and consultants alike. The current consensus is that job rotation offers only limited, or even no benefit in reducing MSD risks. We do not fully agree with this “consensus”. Here’s why …


First, notable SLRs (Systematic Literature Reviews) on the topic (Padula et al. 2017) mostly reviewed studies completed prior to the advent of modern multi-task ergonomic evaluation models (circa 2016-17). These multi-task evaluation models are relatively new and include FFT, Fatigue Failure Theory (Gallagher & Schall. 2017; Gallagher et al. 2018; Hani et al. 2019) and the RCRA, Recommended Cumulative Recovery Allowance (Gibson & Potvin. 2016). Practitioners are only beginning to harness the full potential of multi-task ergonomic evaluation models. Most existing single-task evaluation models and consultant-developed risk assessment tools have no natural mechanism to model the physical exposure of employees rotating into different jobs, nor do they have the capability to calculate a cumulative shift exposure dose. This is largely due to single-task models relying upon average or worst-case input values and being designed to evaluate mono-task jobs (i.e., they assume the worker performs a simplistic action/motion repeatedly with little variation). After multi-task models become more widely adopted and their application matures/advances, there will no doubt be subsequent case studies where multi-task models serve as the basis for the design and management of job rotation plans.


Second, the SLRs we have reviewed excluded studies demonstrating the benefit of job rotation in reducing MSD risks, particularly at moderate to low physical exposure levels (Mossa et al. 2016). It is possible that these studies did not match the keywords used by the researchers in their SLR. Whether intentional or not, potentially relevant studies were excluded from the conversation.


Third, we believe that many ergonomic researchers have a bias against job rotation. They simply don’t like the idea of companies using job rotation as a solution in leu of actually making job improvements to reduce MSD risks. A recent paper (Mehdizadeh et al, 2020) conducted a theoretical analysis of three hypothetical jobs. One of the three jobs had very high lifting demands. Subsequent analysis of hypothetical job rotation paths all resulted in unacceptable injury risks for any rotation scenario involving the high risk job. Although this paper makes some very insightful points, the central theme is self-described as a “cautionary tale” of job rotation.


Sure, some companies have abused job rotation, rotating employees instead of spending the time and money to actually reduce MSD risks. Despite these abuses, job rotation is not the villain it's made out to be. Over our years applying ergonomics in industry, we have seen situations where job rotation, combined with ergonomic risk abatements, was highly effective at reducing MSDs and associated costs. For job rotation to be effective, it is necessary to have jobs to rotate into that provide adequate rest/recovery such that the employee’s cumulative exposure dose for the shift does not exceed acceptable limits.


Amazon’s announcement of a job rotation program will no doubt be viewed with skepticism by many in the ergonomics community. To support their argument, detractors may cite the lack of scientific evidence supporting the effectiveness of job rotation for reducing MSDs. Amazon does not strive to be average, typical, or swim with the academic consensus. In his Letter to Shareholders, Bezos describes how he doesn’t want Amazon to be “typical”. A case can be made that Amazon’s announced invention in the area of job rotation and related algorithms is on the cutting edge of innovation in ergonomics. Allow us to describe how Amazon might possibly achieve Bezos’s job rotation vision by utilizing software algorithms and multi-task ergonomic evaluation models.


Multi-task evaluation models predict human tissue damage and localized muscle fatigue. Model inputs are typically based upon the percentage of a worker’s maximum strength used to perform a physical action. If you can describe manual work as a percentage of maximum strength (i.e., % of maximum grip strength, % of maximum shoulder torque, etc.), then it is possible to combine exposures across multiple jobs/stations to calculate a cumulative shift exposure dose. It would first be necessary to calculate an exposure rate by body part/region for each job/station considered in the rotation plan. The exposure rate could be the average exposure rate of an employee working in a job/station. Leveraging advanced warehouse management software enables a more advanced approach, calculating the physical exposures real-time for specific employees while they work in Amazon Fulfillment Centers. For jobs in a Fulfillment Center, we suspect that the location, size, weight, and count of each item handled by employees can be known. Therefore, it doesn’t require a stretch of the imagination to foresee advanced warehouse management software calculating employee exposure metrics real-time, continually updating based upon actual physical demands. In addition to calculating the physical exposure metrics real-time, it would be possible to calculate anticipated future metrics based upon forecasted/scheduled work. Note that all of this could be accomplished without using wearable sensor technology - simply leveraging advanced warehouse management software and related algorithms.


Although job rotation may not be the holy grail solution to MSDs that Bezos is looking for, understanding it is certainly necessary for evaluating and understanding the risk exposure that employees face. The exposure to risk that an employee faces is the result of what they actually do while at work. By tracking the job rotation schedule and using multi-task ergonomic evaluation tools we can determine actual risk and mitigate the specific aspects of the job that are the riskiest. The current focus on the exposure at a given work station, even if using modern multi-task tools misses the point. Modern workers in manual materials handling jobs rarely work at a single station doing a single task all day. This requires us to track the employee’s job rotation in order to calculate their actual exposure. The path IS the job. The rotation path through the various work stations and time spent at each work station is the job, and that job will continually change as the rotation path changes. The job, as actually performed, is what results in risk exposure; NOT the individual job/station metrics or the job rotation path through those specific stations. We need to know both in order to calculate the actual exposure level. Exposure becomes an algebraic equation comprised of variables that represent specific job metrics and time spent at each station. Whether Amazon/Bezos’ approach recognizes this fact or not, job rotation information is critical for controlling ergonomic physical exposure levels and reducing MSDs.


The exciting part is the potential for software algorithms to plan or adjust job rotation schedules, or even modify an individual worker’s job assignments … realtime. If an employee approaches unacceptable exposure limits for a particular body part/region, there are a number of risk mitigating alternatives. The algorithm could rotate the employee into a different job/station having lesser demands on that body part/region, providing much needed “recovery”. Alternatively, the employee’s upcoming work assignments could be adjusted (in real-time) by the algorithm such that appropriate rest/recovery needed is provided in future tasks. For example, if an employee experienced a higher than average volume of heavy items picked from high pick locations (resulting in high cumulative loading to the shoulders), this employee’s future work assignments could be adjusted real-time by algorithms such that future order picks were less physically demanding to the shoulders. An important distinction is that job rotation can be between job/station, or within job/station by adjusting future work assignments.


Amazon has the big data capability to track all of their employees at a given site, ensuring that rotating one employee out of a given work station does not rotate another employee into that work station that would have their risk exposure become unacceptable. By aggregating data over the entire team, this model will be able to identify certain risks that should be prioritized for fixing when for example there are no employees that can be rotated into a given job with acceptable risk. This will allow Amazon to identify those job risks that cannot be “rotated safe” and instead need to be mitigated. Some problems may be be able to be solved through rotation alone, but other problems will require certain work stations or specific work tasks to be changed or eliminated in order to bring risk to an acceptable level.


Another potential technology advancement of such an algorithm-based job rotation program is to leverage ML (Machine Learning) to discover patterns/trends between the past occurrence of MSDs and actual employee physical work exposure histories. A natural extension of this same technology is to leverage PA (Predictive Analytics) to forecast the impact to ergonomic risk metrics of future/proposed changes in job rotation schedules, cycle time reductions (i.e., increasing work pace), process modifications, etc. The ergonomics process of the future is likely to be very different than what we are accustomed to today. Instead of reacting to problems and injury trends, future ergonomics initiatives may very well leverage multi-task evaluation models, advanced software algorithms, and data scientists to not only learn from the past, but look into the future.


Amazon’s plans for a job rotation program in 2021 may or may not have anything in common with the ambitious approaches described above. But technology advancements such as these are certainly on the horizon. Everyone may not agree, but we believe Amazon’s interest in job rotation is a good thing for safety, because understanding job rotation is necessary for reducing MSDs in the workplace. Read the Jeff Bezos Letter to Shareholders and decide for yourself (see link below).


https://www.aboutamazon.com/news/company-news/2020-letter-to-shareholders


Copyright 2021, Saturn Ergonomics Consulting, LLC.



AUTHORS:


Murray Gibson, PE, CPE

Founder

Saturn Ergonomics Consulting, LLC

murray@saturnergonomics.com


Bob Sesek

PhD Student

Auburn University

bob@auburn.edu

Founder

Innovation Monkey LLC

bob@innovationmonkey.com



ACKNOWLEDGEMENT:

Amazon company name and logo used under rights of the Fair Use Doctrine, https://en.wikipedia.org/wiki/Fair_use

REFERENCES:


Amazon, Jeff Bezos 2020 Letter to Shareholders, https://www.aboutamazon.com/news/company-news/2020-letter-to-shareholders


Gallagher S, Schall MC (2017) Musculoskeletal disorders as a fatigue failure process: evidence, implications and research needs. Ergonomics 60:255–269. https://doi.org/10.1080/00140139.2016.1208848


Gallagher S, Schall MC, Sesek RF, Huangfu R (2018) An Upper Extremity Risk Assessment Tool Based on Material Fatigue Failure Theory: The Distal Upper Extremity Tool (DUET). Human Factors 60:1146–1162. https://doi.org/10.1177/0018720818789319


Gibson, M., & Potvin, J. R. (2016). An equation to calculate the recommended cumulative rest allowance across multiple subtasks. Association of Canadian Ergonomists Conference, Niagara Falls, NY.


Hani, Dania Bani, et al. “Shoulder Risk Assessment Based on Fatigue Failure Theory.” Proceedings of the Human Factors and Ergonomics Society Annual Meeting, vol. 63, no. 1, Nov. 2019, pp. 1117–21. https://doi.org/10.1177/1071181319631395


Mehdizadeh A, Vinel A, Hu Q, et al (2020) Job rotation and work-related musculoskeletal disorders: a fatigue-failure perspective. Ergonomics 63:461–476. https://doi.org/10.1080/00140139.2020.1717644


Mossa, G., et al. “Productivity and Ergonomic Risk in Human Based Production Systems: A Job-Rotation Scheduling Model.” International Journal of Production Economics, vol. 171, Jan. 2016, pp. 471–77. https://doi.org/10.1016/j.ijpe.2015.06.017


Padula, Rosimeire Simprini, et al. “Job Rotation Designed to Prevent Musculoskeletal Disorders and Control Risk in Manufacturing Industries: A Systematic Review.” Applied Ergonomics, vol. 58, Jan. 2017, pp. 386–97. https://doi.org/10.1016/j.apergo.2016.07.018