Ramlee Ibrahim & Associates

How To Forecast Intermittent Demand

Do your products exhibit intermittent demand patterns? I bet that anyone who is a capital goods manufacturer or service parts inventory manager has wrestled with this common and costly inventory management problem.

Unlike most product sales and demand data, intermittent demand contains a large percentage of zero values, often 30 percent or more, with non-zero values mixed in at random. If there is great variability among the non-zero values, this demand pattern is also called “lumpy”. Whatever it is called, the costs of inaccurately estimating lead-time demand and target service level inventories in this environment are potentially huge.

What makes forecasting intermittent demand data so difficult? Largely, it’s the predominance of zero data values. Familiar techniques useful in forecasting conventional or “smooth” demand, such as exponential smoothing and moving averages, ignore the special role of zero values and other key features of intermittent demand.

In the case of service parts, there is an additional twist to the forecasting problem. Here, the forecasts are usually used as inputs to inventory control models. Inventory control theory requires forecasts of the entire distribution of possible demand values – not just a single number thought to be the most likely demand – and requires forecasts over a total lead time, not just a single time period. If these forecasts are accurate, then the inventory models can recommend correct procedures for inventory management, such as the size and timing of replenishment orders.

Traditional statistical forecasting methods fail because they assume that the probability distribution of demand over a lead time (lead-time demand) will resemble a “normal” bell-shaped curve. This certainly is not the case for most service parts. Instead, lead-time demand can have odd shapes, and classical forecasting methods can provide grossly misleading inputs to inventory control models. Most computerized forecasting tools identify recognizable patters in the data, such as trend and seasonality. But there are no easily recognizable patterns in intermittent demand data.

Researches have confirmed that exponential smoothing are effective in forecasting mean (average) demand period when demand is intermittent. But this method does not accurately forecast the entire distribution of demand values. This is true with customer service level inventory requirements – for example, a 90 percent, or 99 percent likelihood of not running out of a product item – for satisfying total demand over a lead time.

The core idea is what we call “bootstrapping”. It is a statistical method that accurately forecasts both average demand per period and customer service level inventory requirements. It does this by using samples of historical demand data to create a large number of realistic scenarios that show the evolution of cumulative demand over a lead time.

Consider the 24 monthly demand values shown in Figure 1. Suppose forecasts are needed for the next three months because the parts supplier takes three months to fulfill an order to replenish inventory. A simple bootstrapping approach to this problem is to sample from the original 24 values, with replacement, three times, creating a bootstrap scenario of demand over lead time.

For example, we might randomly select months 7, 12, and 5, which would give us demand values of 0, 9, and 4, respectively, for a total lead time demand in units of 0+9+14=13. Repeating the process, we might randomly select months 20, 8, and 20 (again), giving a lead time demand of 0+35+0=35 units. By continuing to generate bootstrap scenarios in this way, we can build a statistically robust picture of the lead-time demand distribution.

The histogram in Figure 2 shows the results of 10,000 bootstrap scenarios. (These bootstrap scenarios reflect all elements of the methodology, including real world possibility that non-zero demand values that appear in the future may differ from those that appeared in the past.)

In this example, the most likely lead time demand value is 0, but demand can extend up to 80 or more units. Obviously, the lead-time demand distribution in Figure 2 looks nothing like a bell-shaped curve – and any inventory models assuming it does will provide unreliable advice on setting reorder points and order quantities.

This bootstrapping approach provides fast and realistic forecasts of intermittent product demand over a lead time. In turn, these forecasts can be entered into inventory control models to strike the proper balance between keeping enough inventory on hand to satisfy customer demand and keeping as little inventory as possible to hold down costs.

The Theory of Constraints

The Theory of Contraints (TOC) essentially is a managerial philosophy that focuses on helping managers identify impediments to their goal(s) and effect the changes necessary to remove them. In his best-selling business book The Goal, Dr. Eli Goldratt, the founder of TOC, says the essence of management is determining the answers to three questions:

1. What to change? Pinpoint the core problems.
2. What to change to? Construct simple, practical solutions
3. How to cause the change? Induce the appropriate staff to buy into such solutions

To help managers address these questions and guide them in their quest to manage their own constraints, Goldratt offers a five-step process to improve the performance of any system.

Step 1. Identify the system’s constraints. If you could choose to add more of a single resource, which one would allow your system to increase its throughput (amount of product or service delivered to the customer)? If the constraint is physical in nature, it could be:

• Materials – the input to the process
• Capacity – insufficient amount of a specific resource relative to market demand; or
• Market – isufficient sales to consume product being produced or servioce availability with the existing capacity.

However, experience have shown that many constraints reflect managerial policies or paradigms.

Step2. Decide how to exploit the system’s constraints. Determine how to work with the system’s constraint to maximize throughput. For instance, if the constraint is a specific raw material, ensure no waste of the material. If the constraint is in sales, determine hot to capture more sales. If the constraint is a specific internal resource, ensure the resource is productive all of the time. This is a difficult process, and extracting the most throughput from the system often entails strategic decisions.

Step3. Subordinate everything else to the above decision(s). This synchronizes the rest of the organization with the capabilities of the constraint and the decisions made regarding how best to utilize it. For instance, if the constraint is a machine on the line, establish inventory buffers to protect its abilities to produce and base the release of materials into the plant on the schedule for that constraint and the amount of buffer time established.

It is here that most of the organization’s traditional peformance measures must be changed. For example, every single resource that is not the constraint severely damages the organization if it strives for 100 percent utilization. However, in traditional operational settings, that’s exactly how the resources are measured.

Step 4. Elevate the system’s constraint. In previous steps, managers ensure the organization is optimized via nothing more than procedure/policy changes. In this step, managers actually alter the constraint. For instance, if the constraint has been a machine in the plant, the manager would add physical capacity. How?

1. Reducing set-up and process times
2. Investing in other process improvements
3. Increasing overtime, hiring more staff, buying another machine, or taking any other action that eliminates the machine as the constraint.

Step 5. Don’t let inertia become the system’s constraint. Some or all of the policies established in steps two and three may not be appropriate now that the original constraint has been removed. The manager must reexamine the policies’ applicability and effectiveness under the new operating situation and, in particular, determine where the new constraint is located by revisiting step one.

By using TOC five-step focusing process, a company validates its commitment to continuous improvement.

Strategies For Trimming Inventory Carrying Costs

Each year, Asian companies spend an estimated $400 billion on logistiocs services. While most of that money is well spent, some 40 percent of it isn’t adding value becaue it falls under the line item of inventory carrying costs. In logistics terms, inventory carrying cost is the money a manufacturer has tied up in raw materials, work in progress, and finished goods, which includes materials and products in storage, transit, and production.

Although totally avoiding paying some form of inventory carrying cost is unlikely, you can substantially reduce the costs with the right logistics practices.

1. Strategic purchasing. The better price you can negotiate for your products’ raw materials, the lower your raw material inventory carrying cost will be. Larger corporations usually enjoy a significant advantage because they typically qualify for hig-volume discounts. But with purchasing cooperatives, small-to-medium-sized manufacturers have some recourse. These cooperatives compile the orders of several companies to give all a better price. Another purchasing-related strategy is blanket ordering. A company places an order for supplies well in advance – a practice that suppliers appreciate and reward with better pricing – but arrangtes to have the order delivered in convenient increments over a period of time. The company then pays for the supplies as they are delivered. Better purchasing practices are easier for large companies to implement, but they address only the inbound side of inventory carrying cost, which is not necessarily the most expensive.
2. Better inventory velocity. This method also focuses on the inbound logistics side. Just-in-time (JIT) methods aim to eliminate raw materials storage by transporting materials directly from supplier to production line in pre-arranged, carefully timed increments. Many companies have shaved millions of dollars off their inventory carrying costs and storage expenses this way, but JIT isn’t for everyone. More than one production line has been temporarily crippled awaiting delivery of a 25-cent part. The efficient deployment of JIT requires a highly disciplined production process and a reliable set of suppliers. JIT tends to be more viable for some industries (such as automotive, electronics, and other durable goods) than others. Other potential limitations of JIT are uncontrollable circumstances such as weather and political unrest. Some companies have found a way around these shortcomings by using a flow-through facility instead, where raw materials still move quickly but some actually stop for a time in a warehouse. Even with this method, you must balance the transportation costs of making frequent pick-ups from suppliers – one of the standard operating procedures of JIT and flow-through – against the benefits of high inventory velocity.
3. Smart site selection. This tactic addresses the outbound side of logistics. In an ideal world, goods would move off the production line, onto a truck, and directly into sellers’ or users’ hands. But becaue no one can predict future demand with 100 percent accuracy, companies need some products on hand. Warehousing for finished goods is often essential. By selecting a few strategic locations for distribution centers and making use of the available transportation alternatives, you can reduce the number of centers and products in reserve, which will reduce your finished goods inventory cost. One safe bet is to ensure all your locations offer multiple inbound and outbound transportation alternatives. You can’t implement this tactic overnight. Time and money are required to a distribution center network. And making the changes takes even longer, often because companies are tied into distribution centers by ownership or long-term lease agreements.
4. Inventory financing. With this method, a third party agrees to fund a manufacturer’s raw materials and finished goods in exchange for both the opportunity to manage that inventory and collect a fair market cost of capital. Manufacturers get the cost of raw materials off their balance sheets, increasing borrowing power and owner’s equity, and enhancing cash flow because they can pay upon shipment. This method can potentially reduce the cost of supplies because the interest rate for these materials often is based on the customer’s, rather than the supplier’s credit rating (which is often more favorable), so suppliers can charge less for materials.

Of these four cost-reduction tactics, only inventory financing spans the entoire supply and demand chain and can be imploemented relatively quickly. However, only a handful of third parties currently offer inventory financing, and not all of them have the best service solutions for specific client situations. You’ll have to search to find a partner that will meet your needs.