Distribution Planning for Stable Networks

INTRODUCTION
Over the last few years there has been a great deal of press and hype surrounding the significant return on investment that can be generated by wave of supply chain planning (SCP) and advanced planning and scheduling (APS) solutions. What seems to be an all-encompassing solution, however, turns out to be somewhat limited in application. The basic framework of most optimization solutions is aimed at strategic optimization of a supply chain network. The underlying assumption is that cost differentials exist between plants and freight alternatives, and that these costs will change over time. An optimal solution thus requires some form of linear programming modeling. If the cost differential assumption is removed, these applications have little applicability.

Many enterprises are organized around stable networks of plants and distribution centers. For each product there is a preferred source, and for each network lane there is a preferred transportation mode. The minimal cost objective is implied in the manner in which plants and distribution centers are organized and the preferred transportation mode. The network-planning problem is thus a tactical and operational management issue rather than a dynamic sourcing issue. For these environments a much less complex solution is more appropriate. The solution must focus on demand roll up, supply allocation and transportation reservation.

Recently we have seen the emergence and rapid growth of a range of memory resident enterprise, distribution and production planning systems which allow concurrent testing of constraints. At the enterprise level there are still only a limited number of offerings, however, within the manufacturing sector there are now a wide range of finite scheduling systems with varying capabilities. Some are aimed at the "configure to order" sector, some at the "make to order" market, while others aim for the "made to stock" market. They vary greatly in their capabilities, generally due to their industry sector focus.

Distribution planning has had a similar growth in planning systems, although not to the same extent as finite scheduling. Early entrants into this market used the MRP logic to provide time phased 'gross to net' calculations with time offset for delivery time. However, these models did not take into account other constraints in the network. More recently we have seen the emergence of distribution planning systems that can concurrently plan customer service achievement, carrier service capacity, loading and unloading capacity and warehouse storage capacity. Some systems can handle complex drop ship routings, cross docking and cross network transfers.

The dramatic improvement in usability of these systems has come primarily from the increase in capability of the PC. Massive increase in memory has allowed more complex models to be manipulated in memory with near instant simulation capability. Significant increases in processing power have made complex linear programming (LP) and other mathematical or process intensive models able to reach viable solutions within reasonable time. Concurrently, the software vendors have developed packaged solutions and 'user friendly' interfaces. For the first time complex operations research solutions have become practical in day to day operational decision making.

Advanced Planning and Scheduling (APS) systems have tended towards 'optimization' solutions built around LP. It is interesting to note that much of the initial focus around distribution planning has also been based on LP models. Early models have concentrated on 'cost optimization'. The literature pervades with stories of freight rate tables and multi-sourcing. One is almost led to believe that a distribution planner would not be doing his job without a dynamic network router and the capability to dynamically source across multiple plants or multiple suppliers.

However, not every planner needs complex modeling features to achieve an optimal plan. Cost optimization models rely heavily on cost differentials, either in freight alternatives or differing plant manufacturing costs. Distribution environments that face common manufacturing costs across plants and have longer visibility of demand can achieve optimal solutions with dramatically less complexity in their modeling. In particular, planners in made to stock manufacturing companies can achieve optimal replenishment with minimal modeling data.

The Non-complex Environment
A significant part of the manufacturing sector manufactures against a forecast of demand and holds product in inventory awaiting sales orders. This 'make to stock' environment occurs frequently in food and beverage, specialty chemicals, rubber and plastic products, building products, paper products and over the counter pharmaceuticals. A large segment of this group is involved in high volume, low value product manufacture. Most of the long life consumer packaged products belong to this segment.

The high volume, low value, made to stock manufacturer has some readily identifying characteristics:

• multiple small to medium-sized plants spread geographically
• high proportion of dedicated or preferred production equipment
• many packaging variants
• regional distribution networks
• most likely work with large named accounts
• often involved in vendor managed inventory (VMI) sites
• distribution cost is high relative to product value

Figure 1.	Stable networks have the advantage that they are relatively easy to describe and model

We see their products most often in our local supermarket or home improvement store. Products are made in multiple locations across a region using common production processes and usually the same formula. Each plant services a number of major accounts in the region with direct deliveries to customer warehouses and a network of their own regional warehouses. Generally inventory volumes are high due to the variety of products offered and the volumes distributed. Most of the time the manufacturing cost per unit is near identical from plant to plant, thus eliminating the sort of cost differentials that encourage plant specialization and dynamic sourcing.

Because of the high volume of low value products shipped, the distribution network is relatively stable and also has some readily identifiable characteristics:

• Shipping locations change infrequently, thus the network is relatively stable
• Each outlet usually has a single, preferred source for each stocked item. This is reinforced by the regional nature of manufacturing and the desire to minimize shipping costs.
• Due to the cost of freight relative to the product value, items are shipped in quantity and thus only periodically. Generally dry or packaged items are packed in pallets.
• The replenishment frequency is planned so that full truck loads are used. Often a load will be made up with the higher volume products to ensure full truck loads.
• Because of the longer term visibility of made to stock replenishment, the shipping method is planned based on a default minimum cost service. Alternatives are only considered when product shortages require either delivery times or shipping quantities to be reduced. This is normally done on an exception basis.

The nature of the stable network means that considerable planning complexity is removed from replenishment planning. Routes are fixed rather than dynamic. Long term contracts can be placed for each service due to the regular and predictable nature of the capacity required. The planning system can be set up so that the bulk of the planning task is routine with exceptions being handled by planner intervention.

The distribution planning tasks in this environment are most often split between corporate and regional responsibilities. Corporate staffs define the network, the inventory policy and undertake the allocation of demand. The plant is then responsible for the allocation of supply to the network and the management of the transportation logistics.

Demand Rollup
The objective of the inventory planning function is to have the right quantity of an item at a specific stocking location to meet the target customer service level while at the same time arriving at a replenishment plan that will minimize the total of the replenishment and carrying cost. In general, the quantity required to meet customer service levels is determined by the safety stock. As this next table will show, that is a function of the forecast error and the target customer service level.

The replenishment quantity is normally determined through a consideration of the costs of replenishment vs. the costs of carrying inventory. This is somewhat problematic where full truckloads carry a mix of products, each of which has varying seasonality characteristics. However, by reviewing resupply costs over various transportation assumptions, a preferred delivery frequency can be determined for each class of products. The cargo type (pallet, container, etc.) is then used to determine the shipment method for each.

Once these assumptions have been determined, the demand rollup undertakes a 'gross to net' calculation up through the network back to the preferred source using a planned delivery calendar. This is the statement of demand that the plant capacity planning and production scheduling will use to drive its own calculations.

It is useful in building the demand values to be able to separate demand into components for later replenishment planning. For example, actual sales orders could be classified by priority and separated from forecast demand. It may also be useful to know the quantities that would maintain safety stock or the volume that would be needed to satisfy inventory policy.

Supply Planning
Most make to stock manufacturers use common carriers for transportation of product to regional warehouse and VMI sites. The difficulties and complexities of dealing with vehicle maintenance, union workers and transportation regulations encourage most companies to outsource this activity. In doing so, it greatly reduces the replenishment planning complexity in a stable network. Given the visibility provided by the demand rollup, the transportation requirements can be determined days, if not weeks, in advance and a relatively firm shipment schedule determined.

The major problem in supply planning that the planner has to address is the issue of over or under availability of product. Where insufficient quantities are available to meet the network demand, some reduced allocation must be undertaken. Conversely, where an excess of quantity is available, a choice must be made as to where the product should be held in inventory.

Conventional DRP systems generally used a 'fair share' allocation rule for shortage allocation. With more advanced processing power, a wider range of possibilities can now be supported. For example, the allocation may consider the following:

• Sales order priorities
• Location priority
• Customer priority
• Maintenance of safety stock
• Reduced days of cover

The allocation can take into account the preferred frequency of replenishment in order to identify anomalous situations that may need some manual intervention.

Where excess supply is available, a situation where actual and forecast demand can be met, the allocation of product may use different rules. Consider the following:

• Hold excess at a designated location
• Satisfy maximum levels at specified locations
• Maintain only safety stock at the plant and disperse the balance
• Equal days of supply

The supply plan takes into account the MPS allocation as well as the preferred delivery calendar. It does not, however, assume any limitation on delivery capacity. This type of plan is very useful for companies that outsource transportation as they can simply provide shipment details to their external carrier. The external carrier then has the responsibility of determining the final transportation details.

Transportation Planning
Where the manufacturer controls his own transportation resources, or needs to work with services like barges, ships or trains, where capacity is constrained, a more constrained supply plan is needed. This may still need to be passed to a transportation execution or transportation management system for final resolution, but significant refinement can be done at a higher level.

Transportation Planning in stable networks has as its primary objective the task of building economic loads. This process takes the supply plan and adjusts it to the specific service capacity limitations that are defined for each lane in the network. By combining planned deliveries over a longer time period, better utilization of carrier capacity might be achieved. Limitations on automatic adjustments are imposed by the need to satisfy demand and safety stock requirements. Where target loads cannot be achieved, these can be reported as exceptions and the planner can make the final decision.

At a detail level, transportation management may require that each load be defined in terms of load placement, stacking, etc. At the planning level, however, the load is being determined over days and weeks. At this level, the fine-tuning need not be undertaken. Crude values for volume and weight by type of cargo can be used to refine the plan down to a daily need. This can then be used to drive a detailed plan, on a load by load basis, over the first few days through a manual load plan or through a load building system.

At the transportation planning level the planner is looking for exception situations where the load is either under the target or over the available capacity. By sourcing additional capacity or pulling forward deliveries, the planner arrives at an acceptable plan.

Planning in stable networks assumes that the preferred carrier can be used most of the time. Thus minimum cost is achieved once the network characteristics are set up. Exceptions are identified and are handled on a one-off basis. This may mean using more expensive options, but the nature of the exceptions mean that each problem has to be individually resolved. Certainly more complex models could reduce the planning problems by resolving more of the exceptions automatically, but the complexity cost is very high relative to the benefits in this type of environment. Since many of the plants are small, they simply cannot support the data collection and data maintenance issues associated with more complex models.

Stick to Basics
High volume, made to stock environments are demand planning driven. The key to optimal planning is the forecast. Driving down forecast error through better forecasting processes, use of promotions planning and collaborative forecasting, will have the greatest impact on costs. The forecast is used to calculate the inventory needed at each location, by item, to achieve targeted customer service levels. From that point on, the planning process just gets the product to the location. The real decisions have already been made.

Stable networks are relatively easy to manage in this situation. The forecast provides enough visibility to firm up delivery frequency and service preferences. This provides the basis for the optimal shipment schedule. Preferences on product shortage are usually well defined. A similar preference exists for excess product allocation.

What the planner needs at this point is a pragmatic system that will do the simple calculations. This includes the 'gross to net' calculations, taking into account a planned delivery calendar, as well as the allocation logic. This does not require complex linear programming or advanced mathematical modeling. Next the planner needs a system that will identify the exceptions and allow him/her to use their personal knowledge to arrive at an acceptable solution.

Stable distribution networks are very prevalent, especially in CPG markets. Basic computer based decision support is ideal in this environment. Much of the distribution planning involves relatively simple calculations, unfortunately there are a lot of them to do. At the same time, memory resident modeling of constraints allows the planner to test out solutions to exceptions so that a viable plan can be arrived at quickly.

Make It Happen
Classical DRP systems always had an advantage in that they could take a lot of the grunt work out of getting a distribution plan done. Unfortunately they were hampered by a lack of reality. Without considering variable delivery times, carrier availability and capacity and product availability, its use was limited. Now with the enhanced computing capabilities available, memory resident modeling can overcome those limitations. In practice, stable networks don't require complex mathematical modeling to achieve viable solutions. Pragmatic heuristics-based models can achieve optimal results quickly, thereby allowing planners to intervene and resolve exceptions.

The savings to the corporation for this type of decision support are well proven and easy to achieve. The solution for distribution planning in stable networks really concentrates on the inventory target more than the complexity of distribution dynamics. Simply put, it aims to get the stocking levels right. That requires, first and foremost, a good forecast. From there we can derive safety stock levels, replenishment amounts and timing and a shipment plan. With better visibility into demand and planned delivery patterns, the shipment choices are often obvious. A good daily planning system can look after most of the calculations.

The other area of concern to most planners is how to resolve shortage allocation. Again, with some simple heuristic logic coupled with increased computing power these issues can be quickly resolved. This leaves the exceptions to be dealt with by the planner.

Stable networks are very prevalent and yet have had very little attention from the APS vendors, perhaps because their systems are simply too expensive and too complex to be supported in simpler environments. However, new solutions are now becoming available which are aimed at the made to stock manufacturer. These provide decision support systems for knowledgeable planners in situations where default and heuristic choices achieve a viable plan quickly. Planners resolve outstanding and exception problems interactively. The ROI is high and the implementation time is relatively short.

What should you do if you are in this type of environment? Get some education on constrained planning, ask questions of the vendors, go see some references and then do the numbers yourself. You will be surprised how easy it is to justify a constrained distribution planning system for even relatively small stable networks.

About the Author
Dr. Tom McKaskill
President
Distinction Software, Inc.
Atlanta, GA, USA

Dr. Tom McKaskill is President of Atlanta based Distinction Software, Inc. a company that offers software products and consulting for supply chain planning to process manufacturing companies. Distinction Software has signed an agreement to merge with PeopleSoft in October of 1999. Prior to Distinction Software, Tom co-founded UK based Pioneer Computer Group Ltd. with his wife in 1979. Pioneer developed the COMMAND MRPII system, the GEMBASE 4GL and the PROMIX Process Manufacturing System. In 1991 Pioneer was merged with ROSS Systems, Inc. Dr. McKaskill is a frequent speaker at manufacturing conferences and workshops.