• Home
• Careers
• Intranet Access
• Contact

• Company
  Institutional
• Business
  Client Portfolio
• Cases
  Projects
• Partnerships
  AIMMS

Retail Cases

Freight and Shipping Logistics Scheduling – Casas Pernambucanas

UniSoma conducted studies evaluating the routing and cargo selection procedures done at Casas Pernambucanas. To do so, a heuristic was developed based on multi-agent iteration and evolutionary learning techniques, allowing the generation of dynamic routes, fleet sizing and shop resupplying frequency.

Problem Description

Casas Pernambucanas constitute a network of over 250 shops throughout several Brazilian city, mainly at the South, Southeast and Midwest regions of Brazil. The companies 5000+ products in Bed & Bath, Kitchen and Clothing departments are resupplied according to season changes and market conditions, while electronics are supplied on demand.

Resupplying shops is done through an only distribution center, located in Barueri-SP, where product selection and loading is done, and where all transportation vehicles depart from. The distribution center operate in variable work shifts which lead to different daily product selection and loading capacities and limit distribution capabilities. Product selection is conducted in different sectors at the distribution center (with variable productivity's) and generally is the process bottleneck.

Fixed distribution routes are used to cover each city, with predetermined resupply frequencies, so that orders are adapted to these frequencies, which can inhibit sales when the next delivery is not imminent.

The delivery scheduling consists in choosing, amongst previously picked routes, material type and vehicle load quantity, so that total transportation costs are maintained at an acceptable level and shop orders are delivered. The product choice determines the unload time at every shop, so the planner must guarantee that time constraints aren't violated, such as time windows for product delivery at every shop, which have great influence in the shop delivery order of each route.

Product transportation is done by third party common carriers, whose fleet have different types of trucks, categorized in classes. For each class, a medium capacity load is established and a freight cost is negotiated with the carrier based on the distance and the duration of trips. The carrier choice and best truck pick are essential while routing the distribution. The cargo insurance offered by carriers, available truck capacity, restrict access to some unload locations and the existence of common carriers specialized in certain regions limit and influence the kind of products that are transported.

The Challenge

Although delivery locations are fixed, the choice for fixed routes and fixed delivery frequency is not an efficient strategy for the company, because the product demand is strongly influenced by (a) market variation, (b) dynamic stock at each delivery location (CMI), (c) client orders and (d) sales strategies and marketing (VMI).

Typically, to create the daily distribution plan 2 to 3 employee are occupied for more than 8 hours generating routes that meat constraints that range from store delivery frequency up to detailed delivery sequencing

In July 2004, UniSoma was hired to conduct studies to identify methodologies that would allow the optimization of daily routing while guaranteeing that all constraints were met, simultaneously analysing all possible variables in the process and generating superior performance than that originally obtained by the company

The Solution

UniSoma carried out a study to show that other forms of delivery scheduling, based on optimization techniques, could generate better levels of services (delivery frequency, load transported, vehicle occupancy level, unitary costs, etc). To do so, a methodology was developed to generate dynamic routes and suggest optimized delivery frequencies, according to stock variations and client orders

The interconnection between several sets of decisions involved in the routing process leads to a combinatorial complexity problem that impossibilitates the use of exact solution methods. An alternative was adopted with the development of a heuristic algorithm to build optimal routes. based on a multi-agent scheme. According to this scheme, given an initial route, built using an algorithm that reproduces the current scheduling process, disturbances are applied (insertion or switching of delivery locations between routes, new shops are allocated to routes, etc.) by several agents (improvement, constructor, destructor, etc.) with the purpose to improve the overall routing.

The agents are procedures that seek solutions (local search, exact mathematical models, constructive or improvement heuristics, etc.) which act on the set of routes modifying them to improve their performance. At each algorithm iteration, the agents act on routes that where modified by other agents, coordinated by a learning system.

The cost per cubic meter is the economical criteria evaluated and used to select routes by, simultaneously, minimizing total cost and maximizing transported volume and occupancy rates (less idleness). This criteria, on the other hand, prevents the construction of routes with low demand and locations far from the distribution center. To solve this problem, the final optimization criteria is the cost per cubic meter weighted by delivery frequency, objectives typically contradictory.

Based on a delivery history maintained by the algorithm of each daily routing scenario, the algorithm seeks to maintain an equilibrium between performance metrics and delivery frequency.

The big advantage of using a multi-agent architecture is the flexibility to insert new agents, update sub-agents with efficient techniques and the removal of agents that, evaluated by a point system, don't present a significative impact on final results.

Benefits

The algorithm was applied to a set of 10 planning days and its results were compared with the performance metrics obtained by the routing plan manually generated for the same set of days. For comparison, only regular company routes where used, to avoid that non-conventional routes would privelge the algorithms performance.

The use of the algorithm during this period indicated a possibility to increase the average number of delivery locations visited daily by approximately 39% (from 116 to 161, average). At the same time, the delivery history and the weighted algorithm objective lead to a potential reduction in the average number of days that a store was unsupplied, as illustrated bellow.

The results also indicate the possibility to increase the number of orders being met and the replenishment of stocks at delivery locations from 50% to 70%, average. The graphics bellow compare the supply service levels (in volume and object numbers) of the companies plans and the ones produced by the algorithms.

Lastly, associated to a better level of delivery, the use of the algorithm indicated an opportunity to reduce the cost per cubic meter of approximately 26%.



To generate the distribution routing plans for the 10 days, the algorithm needed approximately 6 hours, which represents an enormous productivity gain in comparison to the almost 80 hours of planning done by the company.

Unfoldings

Capacity Sizing and Logistics Process Simulation for Hanging Products
Still in 2005, UniSoma conducted, for Casas Pernambucanas, a capacity sizing study for their storage and distribution system for Hanging Products, a process, at the time, in deployment. To do so, discrete event simulation and optimization models were developed which resulted, among others, in (a) future bottleneck identification, (b) equipment and labor needs and (c) the definition of optimized operation policies.

Customer Quote

"The project was developed from 07/2004 to 03/2005 with a broad participation of UniSoma consultants at the daily distribution planning activities conducted at Casas Pernambucanas. Its completion allowed the visualization of the potential gains with the adoption of optimization techniques: reduction in the cost per cubic meter, improvement in the vehicle occupancy rates, increase in the number of deliveries in the evaluated period and the reduction of average days without a delivery."
Roberto Umehara - Logistics Manager - 14th of November 2006

About Casas Pernambucanas

Founded in 1908, it is currently one of the largest chain stores in Brazil. The company participates in the commercialization of diverse products such as clothing, bed & bath, appliances and computers. It has over 270 branches, scattered in 7 Brazilian states in the South, Southeast and Midwest regions and over 150,000 employees.