Measuring the Hidden Cost of AI Errors in Supply Chain Decision-Making
International Journal of Business and Economics Insights · 01.08.2025 · 15 (08) · 01-26
DOI: 10.63125/y33vkx18 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Business and Economics Insights, online erschienen am 01.08.2025. Verfasst von Tasnim Binte Afsar.
Abstract (Originalsprache)
Artificial intelligence (AI) forecasts are judged by their accuracy at the point of decision, but the cost of their errors is paid across the supply chain. This paper asks how large the costs of AI forecast errors are that never appear where the forecast is used, and what determines them. Building on the theory of information distortion, we propose a research model in which AI forecast error raises hidden upstream cost through order distortion and bullwhip amplification, error structure matters at equal accuracy, and human oversight attenuates the effect. We test it in a computational experiment: a three-echelon supply chain in which the retailer plans with an AI forecast whose error structure (white noise, rare spikes, autocorrelation, under- and over-forecast bias), magnitude and oversight regime are manipulated in a full factorial design with 9,000 runs and common random numbers. Errors of identical root mean squared error produced very different costs. Unbiased noise and rare spikes placed 79% and 72% of their cost increase upstream, so that the total cost increase was up to 7.0 times the increase visible at the retailer, whereas over-forecast bias raised only the retailer's cost and under-forecast bias cost little anywhere. Contrary to our second hypothesis, persistent errors generated far less hidden cost than transient ones. Order distortion and upstream bullwhip serially mediated the effect of error magnitude on hidden cost (indirect effect 6.93, 95% CI [6.55, 7.32]). Accuracy measures explained 31% of the variation in the cost increase, against 53% once error structure was also known. A simple override rule cut the hidden cost of rare spikes by 60% but made large biases more costly. A post hoc text analysis of 755 research-article titles shows that research on AI in supply chains and research on the cost of forecast errors rarely meet. The findings imply that AI forecasts should be evaluated by the cost they cause across the chain, not by accuracy at the decision point.
Zitieren
APA
Afsar, T. B. (2025). Measuring the Hidden Cost of AI Errors in Supply Chain Decision-Making. International Journal of Business and Economics Insights, 15(08), 01-26. https://doi.org/10.63125/y33vkx18
Vancouver
Afsar TB. Measuring the Hidden Cost of AI Errors in Supply Chain Decision-Making. International Journal of Business and Economics Insights. 2025;15(08):01-26. doi:10.63125/y33vkx18
BibTeX
@article{Afsar2025,
author = {Afsar, Tasnim Binte},
title = {Measuring the Hidden Cost of AI Errors in Supply Chain Decision-Making},
journal = {International Journal of Business and Economics Insights},
year = {2025},
volume = {15},
number = {08},
pages = {01--26},
doi = {10.63125/y33vkx18}
}
EUVIMED