Choice-Based Demand Management and Vehicle Routing in E-Fulfillment

Abstract

Attended home delivery services face the challenge of providing narrow delivery time slots to ensure customer satisfaction, whilst keeping the significant delivery cost under control. To that end, the firm can try to influence customers when they are booking their delivery time slot so as to steer them towards choosing slots that are expected to result in cost-effective schedules. We estimate a multinomial logit customer choice model from historic booking data and demonstrate that this can be calibrated well on a genuine e-grocer data set. We propose dynamic pricing policies based on this choice model to determine which and how much incentive (discount or charge) to offer for which time slot at the time a customer intends to make a booking. A crucial role in these dynamic pricing problems is played by the delivery cost, which is also estimated dynamically. We show in a simulation study based on real data that anticipating the likely future delivery cost of an additional order in a given location can lead to significantly increased profit as compared to current industry practice.

An approximate dynamic programming approach to attended home delivery management

“…Similarly, Cleophas and Ehmke (2014) discuss decision making of acceptance or rejection of delivery requests, but also propose to reserve transport capacities for specific delivery areas and time windows with a high expected order value. Yang et al (2016) estimate an MNL choice model from real e-grocer data and demonstrate numerically that using this model in time slot pricing to influence demand can improve overall profitability. They employ insertion heuristics to update a pool of feasible routes as orders are coming in over the booking horizon, and derive marginal delivery cost estimates from them that are being used as estimates of the opportunity cost of accepting an order into a particular time slot.…”

“…Furthermore, our opportunity cost estimates are not only based on delivery cost but also take potential future order displacement cost into account. The work of Cleophas and Ehmke (2014) is related to Yang et al (2016) in that both papers combine demand fulfillment and revenue management, but the latter work is concerned with time slot pricing and incorporates customer choice modeling (in contrast to the static demand model of Cleophas and Ehmke (2014)). …”

“…Bühler et al (2016) propose several linear mixed integer programs to approximate delivery costs based on a fixed pool of potential routes. integrate a linear mixed integer program (MIP) formulation into the dynamic pricing approach of Yang et al (2016) in an attempt to anticipate future customer requests. This approach aims at obtaining opportunity cost estimates that feature both delivery costs and revenue implications like the ones we produce in this work.…”

“…We model this booking horizon as being discrete starting at time t = 1 and ending in time period t = T . Customers arrive over this time horizon to book their delivery for this specific delivery day; we follow the demand model of Yang et al (2016) in that we do not consider delivery time slot choice beyond a single day. The customer arrivals follow a homogeneous Poisson process λ.…”

“…Whilst customers expect convenient delivery slots, they are not prepared to pay much for peak-time delivery; around £6 to £7 is the maximum that UK retailers currently charge. However, Yang et al (2016) show in an empirical study that even small fee differences can influence customers' delivery time slots decisions, and may lead to overall improved profitability. Mintel (2016) further find that about 80% of online grocery customers in the UK pay per delivery, whilst the remaining 20% have delivery passes.…”