Corresponding author: Sarah Usoro
Copyright © 2026 Author(s) retain the copyright of this article. This article is published under the terms of the Creative Commons Attribution License 4.0.
Reducing food waste through predictive analytics and cold chain monitoring: A U.S.
Retail Review
Sarah Usoro
1, *
Precious Orekha
2
and David Azikutenyi Galadima
3
1
College of Business, East Texas A&M University, Commerce, Texas, United States.
2
College of Computing and Informatics, Drexel University, Philadelphia, United States.
3
Hankamer School of Business, Baylor University, Texas United States.
International Journal of Science and Research Archive, 2026, 18(03), 382-388
Publication history: Received on 19 January 2026; revised on 03 March 2026; accepted on 05 March 2026
Article DOI:
https://doi.org/10.30574/ijsra.2026.18.3.0403
Abstract
Food waste represents a systemic inefficiency in the United States food supply chain, with approximately 30–40% of
produced food failing to reach consumers (Buzby et al., 2014). This study examines how machine learning-based
demand forecasting and Internet of Things (IoT) cold-chain monitoring have been deployed by major U.S. food retailers
and distributors to reduce waste and improve supply chain performance. Drawing on publicly available corporate
sustainability reports, industry case studies, and peer-reviewed literature, the analysis evaluates implementations at
Walmart, Kroger, Sysco, and FreshDirect. Company-reported outcomes suggest that algorithmic forecasting systems
improve prediction accuracy by 20–42% relative to conventional statistical baselines, while continuous IoT-based
temperature monitoring reduces cold-chain spoilage rates by 15–30%. These figures, it should be noted, are drawn
almost entirely from self-reported disclosures; the absence of independent verification limits the strength of causal
claims. Nevertheless, the consistency and scale of reported improvements Kroger's documented 26% reduction in food
waste (245,289 tons) since 2017, Walmart's 78% waste diversion rate, and FreshDirect's sub-2% waste generation
suggest that technology-enabled distribution represents a viable pathway toward USDA 2030 waste reduction targets.
The study also identifies implementation barriers including capital constraints, data interoperability gaps, and
workforce skill deficits that may limit uptake among smaller operators. Future research should prioritize independent
performance audits and cost-benefit analyses for small- and medium-scale distributors.
Keywords:
Food Waste Reduction; Demand Forecasting; Internet of Things; Cold-Chain Monitoring; Supply Chain
Optimization; Predictive Analytics
1. Introduction
The United States food supply chain loses between 30 and 40 percent of produced food before it reaches consumers
(Buzby et al., 2014). This waste is not confined to any single stage; it accumulates through post-harvest handling,
transportation, distribution, and retail operations. The economic cost exceeds $218 billion annually, while the
environmental toll approximately 170 million metric tons of CO₂ equivalent, plus disproportionate demands on
freshwater and agricultural land adds urgency to what might otherwise appear a logistics problem (ReFED, 2021;
Gunders and Bloom, 2017).
In 2015, the U.S. Department of Agriculture and Environmental Protection Agency established a national target to
reduce food waste by 50 percent by 2030 relative to 2010 levels (USDA and EPA, 2015). Progress toward that target
has been uneven. Distribution operations where perishable goods face time-sensitive quality constraints and complex
routing requirements remain among the most significant sources of avoidable waste.
International Journal of Science and Research Archive, 2026, 18(03), 382-388
383
Traditional distribution relies heavily on historical sales averages and manual inventory management. Forecasting
errors for perishable products average 30–50% under conventional approaches (Waller and Fawcett, 2013), producing
chronic mismatches between supply and actual demand. When retailers overstock, spoilage follows; when they
understock, sales are lost and consumer trust erodes. Cold-chain management has historically faced similar limitations:
temperature excursions during transport often go undetected until product arrives at its destination, at which point
remediation is no longer possible.
The proliferation of machine learning algorithms capable of processing heterogeneous real-time data streams point-of-
sale transactions, weather forecasts, regional demographics, promotional schedules offers a potential corrective to
these forecasting deficiencies (Kamble et al., 2020). Simultaneously, IoT sensor networks have made continuous
environmental monitoring economically feasible for refrigerated logistics at scale (Zhu et al., 2018). When integrated
within broader distribution management platforms, these technologies enable adaptive, data-driven decision-making
that more precisely aligns inventory positioning with actual demand.
This paper examines the evidence for such claims. The primary objective is to assess the technical implementations and
reported outcomes of automated forecasting and IoT monitoring systems at four major U.S. food distributors, evaluating
whether the reported performance gains are sufficiently robust to support broader policy recommendations. A
secondary objective is to characterize implementation barriers that may constrain adoption among smaller operators
a population whose participation would be necessary for industry-wide progress toward USDA targets.
An important caveat governs the analysis throughout: the performance figures reviewed here derive almost exclusively
from corporate sustainability reports and company press releases. These sources have obvious incentives toward
favorable presentation. Where methodological details are disclosed, they are noted; where they are absent—which is
frequent this limitation is acknowledged directly rather than glossed over.
2.
Materials and Methods
2.1.
Research Design
This study employs a mixed-methods approach, combining quantitative analysis of publicly reported performance
metrics with qualitative evaluation of system architectures and implementation contexts. The design incorporates three
components: (1) a structured review of algorithmic forecasting and IoT monitoring technologies as documented in peer-
reviewed supply chain literature; (2) case study analysis of deployments at Walmart Inc., Kroger Co., Sysco Corporation,
and FreshDirect; and (3) evaluation of sustainability and operational outcomes from published corporate disclosures.
2.2.
Data Collection
2.2.1.
Industry Case Studies
Case study organizations were selected to reflect diverse distribution models traditional retail (Walmart, Kroger),
foodservice distribution (Sysco), and online grocery (FreshDirect) and to provide documented evidence of algorithmic
or IoT-enabled interventions with associated waste or efficiency metrics. Primary source materials include annual
sustainability reports, ESG disclosures, and corporate press releases published between 2020 and 2024. Secondary
sources include trade publications (Retail Dive, Food Logistics, Food Dive) and consulting firm reports (McKinsey and
Company). The trade and consulting sources introduce their own potential biases toward technology adoption and
toward the vendors whose products they profile and results drawn from these sources are treated with commensurate
skepticism.
2.2.2.
Technical System Components
Three system categories are examined. Demand forecasting systems encompass machine learning models including
random forest, gradient boosting, and neural network architectures that generate store- and SKU-level inventory
predictions from multi-source data inputs. IoT monitoring infrastructure encompasses sensor networks deployed in
refrigerated transport and storage, transmitting continuous temperature, humidity, and location data to centralized
analytics platforms. Data integration platforms include cloud-based systems aggregating inputs from enterprise
resource planning (ERP), warehouse management (WMS), and transportation management (TMS) systems.
International Journal of Science and Research Archive, 2026, 18(03), 382-388
384
2.3.
Performance Metrics
Waste reduction was assessed through reported spoilage rate changes, absolute tonnage of food waste avoided, and
waste diversion percentages. Operational efficiency was evaluated via forecast accuracy improvement (expressed as
percentage reduction in mean absolute percentage error, or as improvement over a stated baseline), inventory accuracy,
stockout frequency, and logistics cost changes. Environmental impact metrics include energy consumption reductions,
greenhouse gas emission changes, and food recovery volumes.
2.4.
Data Analysis
Literature synthesis followed a structured review of peer-reviewed publications indexed in Web of Science and Scopus
between 2010 and 2024, using search terms combining 'food waste,' 'demand forecasting,' 'machine learning,' 'supply
chain,' and 'IoT cold chain.' Corporate document analysis extracted reported performance figures and matched them to
specific technology interventions where disclosures permitted. Comparative analysis set reported outcomes against
industry-average baselines documented in the academic literature, though the absence of common measurement
standards across organizations limits strict numerical comparison.
Limitations
This study's central methodological constraint is its reliance on company-reported data. Corporate sustainability
reports are not subject to the same verification standards as financial disclosures, and there is no independent auditing
body for food waste metrics. Organizations may selectively report favorable outcomes, define baseline periods
advantageously, or attribute gains to technology investments that coincided with other operational changes. These
concerns are not hypothetical; the reporting bias in corporate environmental disclosures is well-documented in the
literature (see, e.g., GRI Standards compliance studies). The findings presented here should therefore be read as
preliminary evidence of plausible effects rather than as rigorously established causal relationships. Controlled
experimental comparison between conventional and technology-enabled distribution operations was not feasible
within this study's scope. Long-term performance data beyond three to five years remain unavailable for most
implementations reviewed.
3.
Results and Discussion
3.1.
Demand Forecasting: Reported Performance
The most consistently reported outcome across implementing organizations is improvement in demand forecast
accuracy. Industry-facing publications estimate accuracy gains of 20–42% compared to traditional statistical methods
when machine learning models are substituted for conventional approaches (Walmart, 2023; McKinsey, 2022). These
figures require careful interpretation. Neither source specifies a common accuracy metric, and the baselines against
which improvements are measured differ across organizations. A 20% improvement in MAPE starting from a 40% error
rate is materially different from the same relative improvement starting from 15% a distinction that most corporate
disclosures do not make.
Walmart's published materials describe an AI-assisted forecasting system processing point-of-sale data from over 4,700
U.S. stores alongside weather forecasts, promotional calendars, and regional demographic data. The company reports a
21% improvement in seasonal merchandise forecast accuracy and a 16.4% reduction in stockouts during promotional
periods (Walmart, 2023). Additional reported outcomes include a 90% gain in inventory accuracy and a 10% reduction
in logistics costs figures that, if accurate, represent substantial operational improvements. However, Walmart's
disclosures do not specify the forecast horizon, the product categories where gains were most pronounced, or whether
the reported improvements were sustained over multiple seasons. These omissions matter for generalizing the findings.
Ensemble modeling approaches combining random forest and gradient boosting algorithms with meta-learners that
select optimal predictions by product category and forecast horizon appear to outperform single-algorithm
implementations in the technical literature, with reported MAPE reductions of 40–60% versus conventional statistical
forecasting (Waller and Fawcett, 2013). Smaller-scale evidence is consistent with this: House of Spices reported a 30%
improvement in forecast accuracy and a 20% reduction in inventory waste following implementation of an algorithmic
forecasting platform (Relewise, 2024). The smaller operator context is noteworthy here, as it suggests the approach
may generalize beyond large retailers, though a single case study is not sufficient basis for confident generalization.
The translation from forecast accuracy to waste reduction is not automatic. A more accurate forecast reduces
overstocking only if it is acted upon that is, if replenishment orders are actually adjusted downward and supply chain
International Journal of Science and Research Archive, 2026, 18(03), 382-388
385
partners respond accordingly. This behavioral and organizational dimension of implementation receives little attention
in the technical literature but is likely an important determinant of whether forecast improvements translate to
measurable waste reductions in practice.
3.2.
IoT Cold-Chain Monitoring
Continuous monitoring of temperature and environmental conditions in refrigerated logistics represents a more
technically straightforward intervention than demand forecasting: the counterfactual (periodic manual checks or post-
delivery data review) is clearly inferior, and the mechanism of effect earlier detection of excursions enabling faster
corrective response is well understood. Research literature places the spoilage reduction attributable to IoT monitoring
at 15–30% in cold-chain operations (Martin et al., 2023; Patel et al., 2022), though effect sizes in individual studies vary
with baseline spoilage rates and the severity of excursion events in the study period.
Sysco's investment is the most extensively documented among the cases reviewed: a $215 million capital commitment
to equip 2,350 refrigerated trucks with real-time temperature monitoring and GPS tracking. The company reports a
22.7% reduction in food spoilage attributable to this program (DCF modeling, 2023). The Sysco case is notable both for
the scale of investment and for the fact that the performance claim originates from a financial modeling source rather
than from Sysco's own sustainability report introducing a different, though not necessarily less biased, form of third-
party characterization.
Beyond spoilage prevention, IoT sensor data enables energy optimization in refrigeration units by providing the
empirical basis for predictive algorithms that reduce unnecessary cooling cycles. The energy efficiency gains are
plausibly substantial given that refrigeration represents one of the larger variable costs in cold-chain logistics, though
specific figures were not consistently reported across organizations reviewed.
Real-time alerting automatically notifying drivers and dispatch coordinators when temperature thresholds are
breached represents a qualitative improvement over retrospective monitoring that is difficult to quantify but
operationally significant. The practical limitation of this capability is connectivity: rural distribution corridors
frequently experience intermittent cellular coverage, creating data gaps precisely in regions where monitoring may be
most critical (FCC, 2020). Edge computing architectures that buffer data locally during connectivity lapses offer a
technical workaround but add hardware cost and system complexity.
3.3.
Integrated System Outcomes
FreshDirect provides the most striking summary statistic in this review: the company reports less than 2% food waste
generation through a platform that tightly integrates predictive inventory management with real-time warehouse and
delivery monitoring (FreshDirect, 2024). The online grocery model is structurally different from brick-and-mortar retail
in ways that make direct comparison difficult FreshDirect fulfills orders rather than holding open-shelf inventory, which
reduces the overstocking dynamic characteristic of traditional retail. This context does not diminish the figure, but it
does caution against treating it as a benchmark for conventional distributors.
The proposed mechanism of integration benefits is intuitive: demand forecasts reduce over-ordering, IoT monitoring
preserves the quality of what is ordered, and automated replenishment systems continuously reconcile the two against
current inventory status. Industry implementations report that safety stock can be reduced by 20–35% while
maintaining service levels above 98% when automated systems respond to real-time demand signals (McKinsey, 2022).
Dynamic route optimization adjusting delivery schedules in response to updated demand forecasts addresses both
stockouts and spoilage by directing product to where it is actually needed rather than where it was expected to be
needed.
Predictive shelf-life management represents a particularly high-value integration point. When systems can identify
products within two to three days of expiration across entire warehouse inventories in real time, automated routing to
same-day delivery orders, promotional bundles, or food bank donation networks becomes operationally feasible. This
proactive redistribution substantially reduces write-offs and has the collateral benefit of increasing food recovery
volumes, which has social value independent of its contribution to corporate waste metrics.
3.4.
Sustainability Metrics Across Implementing Organizations
Four organizations provide the primary quantitative evidence for this review's central claims.
International Journal of Science and Research Archive, 2026, 18(03), 382-388
386
Kroger reports a 26% reduction in total food waste since 2017, equivalent to 245,289 tons, with 82% waste diversion
from landfills company-wide and 106 million pounds of surplus food donated through its Zero Hunger | Zero Waste
program (Kroger, 2023). The company also reports a 15.2% reduction in greenhouse gas emissions from its 2018
baseline. Using a standard conversion factor of approximately 1.2 tons CO₂ equivalent per ton of food waste, the waste
reduction alone represents roughly 294,000 tons of avoided emissions. These figures are internally consistent and
represent multi-year trends, which makes them somewhat more credible than single-year announcements, though
independent verification remains unavailable.
Walmart reports 78% diversion of unsold products from landfills, 760 million pounds of food donated globally in
FY2023, and a 12% reduction in food waste since 2016 (Walmart, 2022; Food Dive, 2024). The donation figure is large
in absolute terms; the 12% reduction in waste generation is more modest relative to the company's stated target of 50%
by 2030 and suggests that the remaining progress will require more than incremental improvement in existing systems.
Sysco's environmental disclosures are less specific regarding waste reduction directly attributable to technology
interventions. Total GHG emissions of approximately 102,900 kg CO₂e in 2023 are reported, with ongoing investment
in fleet modernization and cold-chain optimization (Sysco, 2023). The absence of a clear waste reduction trend line in
Sysco's disclosures is notable, given the scale of its IoT investment.
FreshDirect converted 667 tons of food waste to eco-fuel and recycled an additional 829 tons in 2023, preventing 25
tons of methane emissions and maintaining a food donation partnership with City Harvest yielding approximately 2.6
million meals annually (FreshDirect, 2024). The company's sub-2% waste generation rate is the single most compelling
outcome statistic in this review, though as noted above, the online grocery context limits direct extrapolation.
3.5.
Implementation Barriers
The performance gains documented above are real, but they have been achieved by organizations with substantial
technical and financial resources. The barriers to broader adoption deserve more systematic attention than they
typically receive in technology-optimistic reviews of this literature.
Capital requirements are the most frequently cited obstacle. Comprehensive implementation covering IoT hardware,
network infrastructure, cloud computing, software licensing, and systems integration requires investments in the
hundreds of thousands to low millions of dollars for mid-scale operations serving 50–100 retail locations (McKinsey,
2021). For independent grocers operating on thin margins, this is not a marginal investment; it may represent an
existential risk if performance falls short of projections. The extended payback periods typical of these implementations
estimated at 24–36 months before positive return is achieved compound the financial challenge.
Data interoperability remains a structural problem across the sector. IoT sensors, warehouse management systems,
transportation platforms, and point-of-sale systems commonly operate on incompatible protocols and data formats.
GS1 product identification standards and EDI protocols provide partial coverage, but the absence of comprehensive
data exchange standards for real-time inventory and environmental monitoring data necessitates costly custom
integration development. This is not simply a technical inconvenience; it means that the value of analytics systems scales
with the breadth and consistency of data inputs, and smaller operators—who are least able to afford integration work—
are also least able to realize the potential value of the platforms they invest in.
Workforce capability gaps compound both of the above challenges. Predictive analytics platforms require users to
interpret probabilistic recommendations and exercise judgment about when to override automated suggestions. Many
distribution and warehouse managers lack this background, and the training investments required to develop it are not
trivial. Large retailers have developed structured training programs to address this gap; smaller operators often have
neither the budget nor the instructional infrastructure to replicate them.
Finally, connectivity limitations in rural distribution corridors create genuine technical constraints on real-time
monitoring. Edge computing architectures mitigate but do not eliminate this problem and add cost and complexity that
partly offset the value of continuous monitoring.
3.6.
Scalability and Policy Implications
Despite the barriers outlined above, there are grounds for cautious optimism about broader adoption. Modular
implementation strategies prioritizing high-impact, lower-cost components (temperature monitoring before full AI
forecasting integration, for example) and expanding incrementally as returns materialize can reduce initial capital
requirements by 50–65% while delivering 60–75% of total projected benefits within 18–24 months, according to
International Journal of Science and Research Archive, 2026, 18(03), 382-388
387
industry estimates (McKinsey, 2021). These figures are themselves drawn from consulting publications with
commercial interests in the technology they analyze, but the underlying logic of phased implementation is sound.
Software-as-a-Service delivery models for demand forecasting platforms represent a structural shift that could
substantially lower adoption barriers for smaller operators. Subscription-based access to analytical capabilities
previously requiring significant capital expenditure democratizes the technology though questions of data sovereignty,
vendor dependency, and SaaS pricing escalation deserve attention before smaller operators commit to these
arrangements.
Industry consortia focused on data standardization offer a high-leverage policy intervention. If common protocols for
IoT sensor data and real-time inventory exchange were established through bodies such as GS1 or government-
convened working groups, the integration costs that currently fall disproportionately on each adopting organization
could be substantially reduced. USDA grant programs and technical assistance targeted at small and medium
distributors could further lower financial barriers, though the effectiveness of such programs would depend heavily on
implementation design.
4.
Conclusion
The evidence reviewed here supports a qualified conclusion: algorithmic demand forecasting and IoT-based cold-chain
monitoring demonstrably reduce food waste in the organizations that have implemented them at scale, with reported
outcomes that despite the limitations of self-reported corporate data are consistent in direction and plausible in
magnitude. Kroger's 245,289-ton waste reduction, Walmart's 78% diversion rate, FreshDirect's sub-2% waste
generation, and Sysco's 22.7% spoilage reduction collectively represent meaningful progress toward USDA 2030
objectives.
The mechanisms are well-understood and technically sound. More accurate forecasts reduce structural over-ordering;
continuous monitoring preserves quality and enables rapid response to excursions; integration of the two enables
proactive redistribution of at-risk inventory to donation networks before it becomes waste. These are not marginal
improvements they represent a fundamentally different operating model for perishable distribution.
That said, the findings must be tempered by the absence of independent verification, the structural advantages of large
retailers that make their results difficult to generalize to smaller operators, and the evidence that even leading
implementers remain well short of their own stated targets. Walmart's 12% waste reduction against a 50% goal is a
reminder that early deployments, however impressive, represent the easier part of a longer trajectory.
The more pressing question for the field is not whether this technology works at Walmart it appears to but whether it
can be made accessible and economically viable for the independent grocers, regional distributors, and mid-scale food
service operators whose collective waste contribution is substantial and whose participation is necessary for national
targets to be achieved. Answering that question requires research designs capable of generating evidence that corporate
sustainability reporting cannot: independent performance audits, controlled comparisons, and rigorous cost-benefit
analyses in small and medium operator contexts. The present study identifies what the technology can do; establishing
what it reliably will do across a broader range of organizational conditions remains the work ahead.
Future research priorities should include independent audits of corporate food waste claims, longitudinal studies
tracking performance beyond the initial implementation period, and empirical investigation of adoption dynamics
among smaller distributors particularly the conditions under which modular implementation strategies succeed or fail
to deliver projected returns.
References
[1]
Buzby, J. C., Wells, H. F., and Hyman, J. (2014). The estimated amount, value, and calories of postharvest food
losses at the retail and consumer levels in the United States. Economic Information Bulletin No. EIB-121. U.S.
Department of Agriculture, Economic Research Service.
[2]
DCFmodeling. (2023). Sysco Corporation (SYY) DCF valuation model. Retrieved from https://dcfmodeling.com
[3]
Federal Communications Commission. (2020). 2020 Broadband deployment report. FCC 20-50. Washington, DC.
[4]
Food Dive. (2024). Walmart cuts food waste 12% since 2016, eyes 50% reduction by 2030. Retrieved from
https://www.fooddive.com
International Journal of Science and Research Archive, 2026, 18(03), 382-388
388
[5]
Food Logistics. (2024). Retailers are using AI to improve demand forecasting. Retrieved from
https://www.foodlogistics.com
[6]
FreshDirect. (2024). Sustainability and food waste initiatives. Retrieved from https://www.freshdirect.com
[7]
Gunders, D., and Bloom, J. (2017). Wasted: How America is losing up to 40 percent of its food from farm to fork
to landfill (2nd ed.). Natural Resources Defense Council Issue Paper IP:17-05-A.
[8]
Gustavsson, J., Cederberg, C., Sonesson, U., van Otterdijk, R., and Meybeck, A. (2011). Global food losses and food
waste: Extent, causes and prevention. Food and Agriculture Organization of the United Nations, Rome.
[9]
Kamble, S. S., Gunasekaran, A., and Gawankar, S. A. (2020). Achieving sustainable performance in a data-driven
agriculture supply chain: A review for research and applications. International Journal of Production Economics,
219, 179–194.
[10]
Kroger Co. (2023). 2023 Environmental, social and governance report: Creating a more sustainable future. The
Kroger Co., Cincinnati, OH.
[11]
Martin, S. L., Bhakta, R., and Singh, P. (2023). IoT-enabled cold chain management for reducing food spoilage in
supply chains. Journal of Food Engineering, 338, 111–125.
[12]
McKinsey and Company. (2021). How technology is reshaping supply chain management. McKinsey Digital
Supply Chain Practice.
[13]
McKinsey and Company. (2022). AI-powered decision making for the consumer goods supply chain. McKinsey
Analytics.
[14]
Patel, K. J., Patel, H. R., and Shah, M. B. (2022). Temperature monitoring systems in cold chain logistics: A
comprehensive review. International Journal of Refrigeration, 135, 45–58.
[15]
ReFED. (2021). A roadmap to reduce U.S. food waste by 20 percent. ReFED Solutions Database. Retrieved from
https://refed.org
[16]
Relewise. (2024). House of Spices case study: AI-driven demand forecasting. Retrieved from
https://www.relewise.com
[17]
Retail Dive. (2024). How Walmart is using AI to transform inventory management. Retrieved from
https://www.retaildive.com
[18]
Sysco Corporation. (2023). Sysco's recipe for good: 2023 corporate social responsibility report. Sysco
Corporation, Houston, TX.
[19]
U.S. Department of Agriculture and U.S. Environmental Protection Agency. (2015). U.S. food waste challenge.
USDA Office of the Chief Economist and EPA Sustainable Management of Food Program.
[20]
Walmart Inc. (2022). Regeneration: Walmart environmental, social and governance report 2022. Walmart Inc.,
Bentonville, AR.
[21]
Walmart Inc. (2023). Walmart advances AI-powered supply chain innovation. Walmart Corporate Press Release.
Retrieved from https://corporate.walmart.com
[22]
Waller, M. A., and Fawcett, S. E. (2013). Data science, predictive analytics, and big data: A revolution that will
transform supply chain design and management. Journal of Business Logistics, 34(2), 77–84.
[23]
Zhu, Z., Chu, F., Dolgui, A., Chu, C., Zhou, W., and Piramuthu, S. (2018). Recent advances and opportunities in
sustainable food supply chain: A model-oriented review. International Journal of Production Research, 56(17),
5700–5722.
