Demand Fulfillment in Multi-Stage Customer Hierarchies
Final Report Abstract
Demand fulfillment aims at optimally matching customer orders with available resources. In manufacturing environments, this problem is often challenging since purchasing and production need to be planned with sufficient lead time, demand is uncertain, supply may be scarce, and not all customers are of equal importance or profitability. The particular structure of sales organizations adds another layer of complexity because they often have multi-level hierarchical structures that include multiple geographic sales regions, distribution channels, customer groups, and individual customers. In such a setting, demand fulfillment is typically not carried out by an omniscient central planner, but through a decentralized and iterative process in which higher-level sales quotas are disaggregated one level at a time by multiple local managers. We denote this process as hierarchical demand fulfillment (HDF). Academic literature on HDF is scant. In practice, pre-defined rules are used to determine the break-down of sales quotas in a planning hierarchy. However, the available rules are simple heuristics; no guidelines exist as to which rule to choose in which business context and how to optimize its input parameters. Our research fills this void and provides scientifically sound methods for the practically relevant problem of HDF. The project has been a collaborative effort of three research teams from the Universities of Hohenheim, Mannheim, and Würzburg. Each team made specific methodological contributions (in coordination with the other teams) to the development of new approaches for HDF. These are reflected in Work Packages (WP) 1-3 of the project. WP1 (Hohenheim) focused on the development of deterministic models for the HDF problem in a multi-period setting, aiming to maximize profits while fulfilling the demand from supply originating at the top of the hierarchy in different periods. WP2 (Mannheim) incorporated demand uncertainty into HDF and focused on the development of stochastic single and multi-period approaches with the objective of maximizing expected profits of the sales hierarchy. WP3 (Würzburg) also focused on the development of single and multi-period stochastic approaches, however, with the major objective of meeting service-level targets of different customer groups. The specific contributions of the project can be summarized as follows: First, we made a substantial methodological contribution by devising 10 novel methods for the relevant yet previously unaddressed problem of HDF. Our developed approaches cover a broad range of methodologies, including deterministic and stochastic, single- and multi-period, profitmaximizing and service-level oriented models. Second, we gained theoretical insights into the properties of the HDF problem. We identified (multi-) knapsack sub-problems as important building blocks of HDF, which can be solved by balancing marginal profits. We used this property in several of our models. We also established a sound theoretical connection between profit-oriented and service-level based HDF. Third, we built a simulation environment (in WP4) that allows for a fair comparison between deterministic and stochastic planning approaches in HDF. Fourth, we documented the performance of our methods by means of extensive numerical studies. The results show that all of our methods outperform previously available methods, reducing their ex-post optimality gap by roughly 90% on average. Moreover, our decentralized methods consistently perform close to their centralized benchmarks. Thus, we demonstrated that decentralized decision making, which reflects organizational realities, does not prohibit effective demand fulfillment, if designed properly. Fifth, we provided insights into the question of what information is crucial for effectively decentralizing demand fulfillment. One key result is that, in addition to expected demand, information on both customer heterogeneity and demand uncertainty substantially benefits the quality of fulfilment decisions. Interestingly, however, a fairly coarse representation of these factors is sufficient, in general, to reap these benefits. Another insight is that taking stochastic demand information into account appears to be more important than multi-period planning, unless shortage rates are very high and/or strongly fluctuate across periods.
Publications
- (2018): Allocation planning in sales hierarchies with stochastic demand and service-level targets. OR Spectrum
Kloos K., Pibernik R. and Schulte B.
(See online at https://doi.org/10.1007/s00291-018-0531-5) - (2019): Deterministic allocation models for multi-period demand fulfillment in multi-stage customer hierarchies. Computers & Operations Research 101, pp. 76-92
Cano-Belmán J. and Meyr H.
(See online at https://doi.org/10.1016/j.cor.2018.09.002)
