Platzhalter für Bild, Bild auf Titelfolie hinter das Logo einsetzen Ann Melissa Campbell, Jan Fabian Ehmke 2013 Service Management and Science Forum Decision Support for Attended Home Deliveries in Metropolitan Areas

 Online retailing is the fastest growing retail sector  Efficient and reliable delivery of orders is critical for lasting success  This is particularly challenging for attended deliveries Examples: grocery, large appliances, repairmen Challenges for Attended Home Deliveries

 For attended deliveries Customers and service providers agree on a time window Customers expect on-time deliveries, narrow time windows Online retailers need to maximize the number of customers visited Deliveries in metropolitan areas suffer from congestion  Not surprising many of the early businesses failed! Challenges for Attended Home Deliveries

Strategic: Decide what service time windows to offer in each customer area. Operational: 1. Selection of a service time window  Decide what windows are feasible for each customer  Decide when to close time windows in different areas 2. Order picking  Picking in warehouse vs. stores 3. Vehicle routing and scheduling  Minimizing costs of delivery Key Decisions © PTV, Karlsruhe

Strategic: Decide what service time windows to offer in each customer area.  Agatz, Campbell, Fleischmann, Savelsbergh 2011 Operational: 1. Selection of a service time window  Assumed deterministic travel time  Not look at impact of congestion/time dependency  Campbell & Savelsbergh 2005/2006, Asdemir et al Order picking 3. Vehicle routing and scheduling  Consider a time-dependent cost matrix  TDVRPTW  Ichoua et al. 2003, Eglese et al. 2006, Fleischmann et al. 2004, Maden et al Related Work © PTV, Karlsruhe

►How do we decide which requests to accept in each service time window?  How do we balance the need to provide reliable service with profit maximization?  Examine several acceptance mechanisms Differ in level of information, ease of implementation  Evaluate them using simulation framework Simulated demands Simulated travel times to reflect congestion in metropolitan areas Our Focus

Home Delivery Problem  Retailer offers a predefined set of time slots on a day  Delivery requests arrive before start of service  Customer selects time slot  Each customer has first and second choice  Provide first choice if available  Provide second choice if first choice not available  Customer leaves if neither available  Need to quickly decide if each request can be handled in a particular time slot  Objective: maximize the number of accepted requests  Assumption: vehicle capacity not binding Time Slot [12:00,12:59] [13:00,13:59] [14:00,14:59] [15:00,15:59] [16:00,16:59] [17:00,17:59] [18:00,18:59]

Solving the Home Delivery Problem  Need to quickly decide if each request can be handled in a particular time slot  This can be done in a rough/approximate way Rules of thumb Fast/easy Refer to as “static approaches”  This can be done in a more dynamic way Use more detailed information, build routes Can accept more deliveries May accept too many deliveries!  Congestion, stochastic travel times Refer to as “dynamic approaches” Time Slot [12:00,12:59] [13:00,13:59] [14:00,14:59] [15:00,15:59] [16:00,16:59] [17:00,17:59] [18:00,18:59]

Home Delivery Problem Static Approaches SLOT:  Accept a fixed number of requests per time slot  If first priority time slot cannot be accommodated, check alternative option, else: reject  Common/simplest TD-SLOT:  Allow the number of requests to vary depending on time of day  Requires general knowledge on the evolution of travel times Feasibility of scheduled routes is not guaranteed. Time SlotSLOTTD-SLOT [12:00,12:59]67 [13:00,13:59]67 [14:00,14:59]66 [15:00,15:59]65 [16:00,16:59]64 [17:00,17:59]64 [18:00,18:59]65 SLOTTD-SLOT

Home Delivery Problem Dynamic Approaches (1/2)

Home Delivery Problem Dynamic Approaches (2/2)

 Arrival time distribution at customer not based on simply combining means and variances of travel times Many assume this!  Arrival time at one customer based on previous travel times and time windows  Consider if majority of distribution of arrival times is before a customer’s time window The opening of the window determines the departure time This reduces the variance in arrival times at subsequent customers Summary: Service time windows impact how you propagate the variability Home Delivery Problem Propagation of Arrival Times (1/2)

Home Delivery Problem Propagation of Arrival Times (2/2) Summary: We compute a unique buffer for each customer based on information from earlier in the tour

Platzhalter für Bild, Bild auf Titelfolie hinter das Logo einsetzen TW 15:00-16:00 DYN: Arrival 15:56 BUF: Arrival 15:32 TW 15:00-16:00 DYN: Arrival 15:59 BUF: Arrival 15:51 TW 17:00-18:00 DYN: Arrival 17:46 BUF: Arrival 17:35 TW 18:00-19:00 DYN: Arrival 18:58 BUF: Arrival 18:14 TW 19:00-20:00 DYN: Arrival 19:42 BUF: Arrival 19:28 DYN: 34 customers Proportion served late: 94% BUF: 34 customers Proportion served late: 7% Impact of DYN-BUF

Computational Experiments Experimental Design Real-world inspired metropolitan network  Two zones of customer locations Inner city and suburbs 100 instances from each zone  Mean time-dependent travel times generated from speed multipliers To reflect daily congestion patterns  Parameters of distribution based on fit with real travel time data (Stuttgart)  Basic test: 7 time slots of width: 60 minutes Service time at customers: 20 minutes Three vehicles Solve each instance with each acceptance mechanism Simulate 1000 times

 SLOT accepts the least requests  DYN accepts the most but is late in 86% of realizations  DYN-BUF: smaller request acceptance, but occurrence of lateness is reduced a lot  DYN-SBF and static approaches accept far fewer but are also rarely late – tend to be conservative Computational Experiments Suburban Delivery

Computational Experiments Inner City Delivery  Extent of lateness is less due to shorter distances  DYN-SBF is very strict in inner city areas – fewer deliveries than TD-SLOT  DYN-BUF reduces lateness occurrence, but accepts almost as many requests as DYN

Conclusions and Outlook  Static approaches do not maximize utilization of logistics resources Parameter estimation difficult and inflexible  Dynamic approaches are able to increase efficiency Basic dynamic approaches good for close locations in downtown Most important to consider travel time variation with longer distances + congestion (suburbs) shorter time windows  Design routing algorithm to minimize lateness propagation

Platzhalter für Bild, Bild auf Titelfolie hinter das Logo einsetzen Thank you for your attention!