The results from the previous section provide clear evidence that labor availability has an effect of IRI.
Furthermore, it is reasonably easy to identify scenarios where IRI, or its consequence, shelf-OOS, could have an impact on the workload of store employees. For example, as discussed in §3, once IRI is present, the restocking system becomes unreliable, increasing the probability of shelf-OOS. Empty shelves translate into confused or angry customers that interrupt normal operations and require special attention through help with searches (either physical or in the computer system), trips to the back room, or placing orders for the missing SKUs. Eventually, high IRI will trigger a response from management that will require employees to perform inventory counts or walk-by shelf inspection. All these activities represent distractions from ‘regular’ operations, thus reducing the labor available to perform them. As per §4, lower labor availability translates into even higher IRI, thus creating a reinforcing loop. Figure 5 operationalizes this reinforcing loop (R1) through the construct of work pressure – the ratio of labor required to effective labor available (Oliva, 2001).
Fig. 5. IRI-Induced work pressure
Nevertheless, it is hard to imagine a situation where this reinforcing loop would run out of control and all employees will be responding to IRI-created work. Customers, for one, would quickly realize that the establishment is not the best option for their purchases and would search for alternative retailers. This would reduce the amount of base labor required to handle regular traffic and would bring work pressure back to normal operating range. However, a quick inspection of the shape and strengths of the links in loop R1 reveals that the gain of this loop saturates very quickly as only a limited fraction of total labor goes into IRI activities.4 Specifically, to assess the gain of the loop around the backlog-shrinkage error,
4 The gain of a link is defined as the ratio of the output to the input, and the gain of a loop is the product of all the link gains in the loop. A reinforcing loop has a positive gain, but to create exponential growth the gain has to be greater than one.
we first used the model described in §3 to estimate the impact of the error rate on IRI. Simple OLS regressions explained 99% of the observed variance and yielded tight estimates for the error rate coefficients on average IRI (βεB-=907.0, p-value<0.0001). We further assumed that work pressure (WP) has a multiplicative effect on the operational error rates defined in §3, that is, = ∗ , and that IRI can take up to 10% of the base labor required per SKU once it reaches the magnitude of the desired shelf level (S*). Finally, we assumed that the impact of IRI is not felt instantaneously by the workforce, but takes time to accumulate. We approximated this adjustment through a first order exponential process with a time constant of one week. With those assumptions the gain of the loop around εB- is only 0.004 at the average observed IRI and assuming a normal initial work pressure.5 Even if all the other error rates were similarly activated, the addition of the five loop gains will still be less than one, and it would take a work pressure greater than 50, i.e., a severe understaffing, for loop R1 to have a gain greater than one.
Note finally, that as formulated, the gain of the loop decreases with IRI as only a limited fraction of labor can realistically be allocated to handling IRI issues.
However, the feedback loop depicted in Figure 5 is not working in isolation and poor data integrity could induce negative consequences with different time delays. Figure 6 shows two simple, yet relevant, feedback loops that characterize the intermediate and long-term impacts of work pressure on labor availability. The loop R2 suggests that IRI-induced work pressure could lead to higher work intensity.
Extended periods of work intensity results in employees’ fatigue, thus reducing their effectiveness and reinforcing work pressure. Extended periods of fatigue, in turn, results in employee burnout, which eventually increases FTE turnover rate, not only reducing headcount (R3), but also affecting the nominal labor productivity as new FTEs will require time to assimilate (not shown in the figure) (see Oliva and Sterman, 2001 and 2010 for evidence and calibration of these feedback loops in service settings). Of course, IRI is not the only source of increasing work pressure: normal variations in customer arrivals, delays in the recruiting processes, and introduction of new product lines are just few examples of normal variations that could trigger the cascading effects on operational error rates and compound the growth of IRI.
5 The gain of the loop is given by the expression / ∗ ∗, where IRI is the average |Ir-Ia| (~20 units), S* is the desired shelf inventory level (50 units), εi is the base error rate for error type i, βi is the estimated impact of the error rate on IRI (from regression of model output), τ is the time constant for the impact of IRI to affect labor requirements (7 days), δ is the maximum effect of IRI on labor requirements (10%) and γ is a parameter that controls the shape of the impact of IRI on labor requirements in the range [0, δ], in our case γ=5.
Fig. 6. IRI-Induced fatigue and burnout
The situation described above can easily be prevented if the effective labor available is maintained above the base labor required and work pressure is maintained below one. However, common hiring practices in the retail industry do not seem to be in line with this basic tenant. First, as shown in §4, there is no evidence that PTEs have an effect on improving IRI performance – if anything; there is weak evidence that they make things worse. Second, addressing IRI issues (helping customers, placing orders or realizing audits) is most likely done by FTEs, a reasonable assumption given the low training of PTEs.
Consequently, the current practice of using PTEs to cover temporary staffing gaps has no substantial effect on the labor available for the purposes of the structure depicted in Figure 6, as PTEs are not capable of containing the IRI from regular operations and are not capable of supporting the customer to deal with the consequences of IRI. We integrate this structural element in Figure 7 and show in a dashed link the weak effect that PTEs have on nominal labor availability. While the effect of PTEs on the described dynamics is inconsequential, the fact that management perceives PTEs as being helpful and reduce the FTEs’ hiring rate does have an effect on the dynamics as it sustains the work pressure that triggers all the reinforcement loops described above.
Fig. 7. Myopic PTE adjustment to labor availability 6. Discussion
Our study takes an empirically grounded multiple-modeling approach to derive insights that help managers prevent occurrence of IRI. We began with modeling a continuous (Q,R) system with explicit considerations of recorded and actual inventories in the backroom and on the shelf. We used this model, through a full-factorial simulation experiment, to assess the impact five different operational errors that are thought to drive IRI. We set our error rates to values observed in our research context, but the impact of those rates in IRI was inferred through the model. We found strong support for the premise that inventory accuracy is vulnerable to operational errors. More importantly, while the literature (Kok and Shang, 2007; Chen and Mersereau, 2015) identifies three operational sources of IRI – shrinkage,
transaction errors, and misplacement – no formal effort had been made to assess their relative impact. Our experimental design identified that all operational errors were significant, but we also found that transaction errors (e.g., checkout and data capture) and misplacement (e.g., under-shelving) have negligible impact on IRI and that shrinkage, either shelf or backroom, is the largest contributor to IRI.
We then empirically estimated the relationship between labor availability and IRI. We
operationalized labor availability as a function of the level as well as mix of store workers. Our estimation model focused on a structural explanation of IRI, as opposed to a correlational study to test hypotheses, and we were able to identify significant labor effects on operational outcomes. While the negative association between FTEs and IRI conforms to the advocate that increasing FTEs improves operational execution (Ton, 2012; 2014), the partially significant positive impact of PTEs on IRI reveals the potential drawback of increasing temporary workers. Finally, we articulated the reinforcing relationships between
labor and IRI by formally assessing the gain of the feedback loop based on our empirical findings and presented multiple feedback loops with the intermediate and long-term impact of IRI on labor availability.
The feedback modeling exercise not only illustrates unanticipated consequences of IRI with different time delays but also reinforces the need for more effort and dedication to tackle IRI. Allocating adequate labor to prevent operational errors identified in §3 would be instrumental in enhancing process conformance and data accuracy. Our analysis of labor and IRI, however, does not rule out alternative drivers of IRI (e.g., those identified by DeHoratius and Raman, 2008). We simply aim to inform retail managers that employees matter when it comes to IRI and add to the newly emerging stream of studies on labor effects, store execution, and retail performance.
A major limitation of our paper is the generalizability of labor effects on retail data quality. Similar to DeHoratius and Raman (2008), we analyze secondary data from stores of a single retailer. Although the significance and appropriateness of random effects modeling in §4 help us generalize beyond this
organization, statistically we may not be able to claim that our finding will hold beyond this retail chain.
Another limitation of our analysis is that we only focus on quantities of labor. Labor quality could also moderate the effects of full-time and part-time labor on IRI. Although we could not obtain data on education, age, experience, engagement, and attitude of employees to assess such moderations, our conversation with the senior manager gives us no reason to suspect that significant differences in labor quality/training exist among the five stores. While we focused more on labor allocation, it would also be interesting to investigate the impact of section manager skills and background on the relationships identified in our paper.
Despite the limitations, our modeling efforts carry pragmatic implications for retail managers. Our findings suggest that management should re-direct their efforts to address shelf and backroom shrinkage.
Extensive surveys from both US (Hollinger and Davis, 2003) and European retail companies (Bamfield, 2004) suggest that employee, customer, and supplier theft cause approximately 80% of shrinkage — the other 20% being caused by internal errors (e.g., poor compliance, spoilage, accidentally damaging goods).
While a technical fix for shrinkage would be item-level RFID (Lee and Ozer, 2007), this solution is too expensive to be adopted by most retailers. Instead, our findings suggest that improving labor availability could be an effective solution, at least in reducing employee theft and internal errors, which account for
~63% of shrinkage in US and ~ 47% of shrinkage in Europe (Bamfield, 2004). Ensuring sufficient staffing levels and adequate labor mix may be the easiest thing for retailers to do in order to tackle execution failures and IRI induced by those errors. That said, retail managers should not just expand labor capacity as doing so may lead to over-staffing. Managers ought to adopt prescriptive labor planning models that avoid myopic wage minimization and consider costs of errors/IRI associated with staffing levels and labor mix. For example, Chuang et al. (2015) propose a data-driven staffing heuristic based on
observed store traffic. Following the principle of profit maximization, traffic-based labor planning significantly outperforms actual staffing decisions that are primarily budget- or cost-driven. Moreover, labor decisions should explicitly strike the balance between FTEs and PTEs rather than naively increasing PTEs to address staffing shortfalls. As mentioned earlier, lack of commitment and training may cause PTEs to commit errors that hamper data integrity more easily. For retailers who are not able/willing to increase staffing levels or change labor mix, labor quality can be improved by aligning incentives (Gino and Pisano, 2008), building awareness (Fisher, 2004), or developing the appropriate processes (Oliva and Watson, 2011).
Our study also carries theoretical implications for researchers. First, the proposed simulation model can be used to assess the impact of random errors on not only IRI but also lost sales. By expanding our model, researchers can reassess the backroom effects (Eroglu et al., 2013) in a generic setting subject to different execution errors while exploring the impact of decisions on shelf space, reorder point, order quantity/case pack size on inventory performance. Researchers can also design and test the efficacy of different shelf inspection policies (e.g., zero-balance walk, cycle-counting, daily counts) after adequately modifying the model. Moreover, since the errors investigated through simulation are common enough, the assumption that the decision maker knows the actual inventory level is not credible (Cachon, 2012). Our findings support the necessity to infer erroneous inventory records and incorporate statistical estimates into inventory control (DeHoratius et al., 2008; Mersereau, 2013).
Second, while previous studies show that sufficient staffing levels help improves service quality (Oliva and Sterman, 2001) and conformance quality (Ton, 2009), our empirical estimation suggests that full-time store workforce enhances data quality as well. The dependence between labor availability and data integrity warrants future investigation. More efforts are needed to articulate the impacts of labor mix as well. Even though PTEs could be an effective means of responding to demand spikes (Kesavan et al., 2014), inexperienced PTEs often need assistance from FTEs and this mentoring requires increasing the workload of FTEs. As a result, it becomes more difficult for FTEs to prioritize assignments and causes productivity loss (Oliva and Sterman, 2010). These dynamics arisen from labor allocation deserve further examination since it is important for managers in service settings to understand the pros and cons of building operational flexibility through mixed workforce.
Lastly, our combination of multiple methods to analyze IRI creates advances in empirical research.
We enrich research angles and insights applying simulation of differential equations, Bayesian statistics, panel data econometrics, and causal loop diagrams. On the empirical front, we are fairly cautious about imposing assumptions on the data generating process. Instead of following the predominant Gaussian thinking criticized by Singhal and Singhal (2012), we take an empirically-grounded approach to set up a sampling model in Bayesian shrinkage, which results in robust econometric estimation. From the system
dynamics approach (Forrester, 1958; Sterman, 2000), we deploy two ideas that hopefully we will find broader adoption in operations management research. First, our empirical exploration of the relationship between labor and IRI is operational. That is, we focused on identifying the functional form of the relationship and providing operational explanations and consequences of our findings (Richmond, 1993).
We believe that this type of analysis is more meaningful than hypotheses-testing correlational studies and is more conducive to generating operational recommendations. Second, our study expanded the boundary of the IRI problem to include the feedback mechanisms and managerial decisions that cause and sustain the problem. We believe that this broader perspective and the formal assessment of the feedback
mechanism are becoming increasingly relevant as operations management is tackling ever broader issues and problems e.g., supply chain management, strategic operations, etc.
References
Agrawal, P.M., Sharda, R., 2012. Impact of frequency of alignment of physical and information system inventories on out of stocks: A simulation study. International Journal of Production Economics 136(1), 45-55.
Anderson, E.G., 2001. The nonstationary staff-planning problem with business cycle and learning effects.
Management Science 47(6), 817-832.
Angulo, A., Nachtmann, H., Waller, M.A., 2004. Supply chain information sharing in a vendor-managed inventory partnership. Journal of Business Logistics 25(1), 101-120.
Baltagi, B.H., 2006. Panel Data Econometrics: Theoretical Contributions and Empirical Applications. Emerald Group Publishing, Bingley, UK.
Bamfield, J., 2004. Shrinkage, shoplifting and the cost of retail crime in Europe: A cross-sectional analysis of major retailers in 16 European countries? International Journal of Retail & Distribution Management 32(5), 235-241.
Bell, A., Jones, K., 2015. Explaining fixed effects: Random effects modeling of time-series cross-sectional and panel data. Political Science Research and Methods 3(1), 133-153.
Box, G.E.P., Cox, D.R., 1964. An analysis of transformations. Journal of the Royal Statistical Society, Series B 26(2), 211-252.
Boyer, K.K., Swink, M.L., 2008. Empirical elephants-why multiple methods are essential to quality research in operations and supply chain management. Journal of Operations Management 26(3), 337-348.
Cachon, G.P., 2012. What is interesting in operations management? Manufacturing & Service Operations Management 14(2), 166-169.
Cameron, A.C., Trivedi, P.K., 2010. Microeconometrics Using STATA, second ed. Stata Press, College Station, TX.
Chen, L., Mersereau, A.J., 2015. Analytics for operational visibility in the retail store: The cases of censored demand and inventory record inaccuracy. In: Agrawal, N., Smith, S.A. (Eds.), Retail Supply Chain Management.
International Series in Operations Research and Management Science Vol. 223. Springer, New York, pp.
79-112.
Chuang, H.H.C., Oliva, R., Perdikaki, O., 2015. Traffic-based labor planning in retail stores. Production and Operations Management, http://onlinelibrary.wiley.com/doi/10.1111/poms.12403/abstract (forthcoming).
DeHoratius, N., Raman, A., 2007. Store manager incentive design and retail performance: An exploratory investigation. Manufacturing & Service Operations Management 9(4), 518-534.
DeHoratius, N., Raman, A., 2008. Inventory record inaccuracy: An empirical analysis. Management Science 54(4), 627-641.
DeHoratius, N., Mersereau, A.J., Schrage, L., 2008. Retail inventory management when records are inaccurate.
Manufacturing & Service Operations Management 10(2), 257-277.
Eroglu, C., Williams, B.D., Waller, M.A., 2013. The backroom effect in retail operations. Production and Operations Management, 22(4), 915-923.
Fan, T., Chang, X., Gu, C., Yi, J., Deng, S., 2014. Benefits of RFID technology for reducing inventory shrinkage.
International Journal of Production Economics 147(Part C), 659-665.
Fiddaman, T., 2010. MetaSD Model Library, available at http://models.metasd.com/pink-noise/.
Fisher, M.L., 2004. To me it’s a store. To you it’s a factory. ECR Journal 4(2), 9-18.
Fisher, M.L., Krishnan, J., Netessine, S., 2006. Retail store execution: An empirical study. In: Working Paper. The Wharton School, University of Pennsylvania, Philadelphia.
Fisher, M.L., Krishnan, J., Netessine, S., 2009. Are your staffing levels correct? International Commerce Review 8(2-4), 110-115.
Fleisch, E., Tellkamp, C., 2005. Inventory inaccuracy and supply chain performance: A simulation study of a retail supply chain. International Journal of Production Economics 95(3), 373-385.
Forrester, J.W., 1958. Industrial dynamics – A major breakthrough for decision makers. Harvard Business Review 36(4), 37-66.
Forrester, J.W., Senge, P.M., 1980. Test for building confidence in system dynamics models. TIMS Studies in the Management Sciences 14, 209-228.
Gino, F., Pisano, G., 2008. Toward a theory of behavioral operations. Manufacturing & Service Operations Management 10(4), 676-691.
Hamada, M.S., Wilson, A., Reese, C.S., Martz, H., 2008. Bayesian Reliability. Springer, New York.
Heese, H.S., 2007. Inventory record inaccuracy, double marginalization, and RFID adoption. Production and Operations Management 16(5), 542-553.
Hoff, P.D., 2009. A First Course in Bayesian Statistical Methods. Springer, New York.
Hollinger, R.C., Davis, J.L., 2003. 2002 National Retail Security Survey: Final Report. University of Florida, Gainesville, FL.
Hollinger, R., 2009. National Retail Security Survey. University of Florida, Gainesville, FL.
Ingene, C.A., 1982. Labor productivity in retailing. Journal of Marketing 46(4), 75-90.
Jackman, S., 2009. Bayesian Analysis for the Social Sciences. Wiley, New York.
Johnson, E.J., Moe, W.W., Fader, P.S., Bellman, S. Lohse, G.L., 2004. On the depth and dynamics of online search behavior. Management Science 50(3), 299-308.
Kang, Y., 2004. Information Inaccuracy in Inventory Systems. Unpublished PhD Dissertation, Massachusetts Institute of Technology, Cambridge, MA.
Kang, Y., Gershwin, S.B., 2005. Information inaccuracy in inventory systems: Stock loss and stockout. IIE Transactions 37(9), 843-859.
Kapoor, G., Zhou, W., Piramuthu, S., 2009. Challenges associated with RFID tag implementations in supply chains.
European Journal of Information Systems 18(6), 185-205.
Keating, E.K., Oliva, R., Repenning, N.P., Rockart, S.F., Sterman, J.D., 1999. Overcoming the improvement paradox. European Management Journal 17 (2),120-134.
Kesavan, S., Staats, B.R., Gilland, W.G., 2014. Labor-mix and flexible response to demand spikes: Evidence from a retailer. Management Science 60(8), 1884-1906.
Khader, S., Rekik, Y., Botta-Genoulaz, V., Campagne, J., 2014. Inventory management subject to multiplicative inaccuracies. International Journal of Production Research 52(17), 5055-5069.
Kok, A.G., Shang, K.H., 2007. Inspection and replenishment policies for systems with inventory record inaccuracy.
Manufacturing & Service Operations Management 9(2), 185-205.
Kok, A.G., Shang, K.H., 2014. Evaluation of cycle-count policies for supply chains with inventory inaccuracy and implications for RFID investments. European Journal of Operational Research 237(1), 91-105.
Lee, H.L., Ozer, O., 2007. Unlocking the value of RFID. Production and Operations Management 16(1), 40-64.
Lee, H.L., Ozer, O., 2007. Unlocking the value of RFID. Production and Operations Management 16(1), 40-64.