Like many parts distributors, Aero Fastener, a distributor of spare fasteners to airlines and MRO organizations, is in the midst of remaking its business model. About two years ago, it began offering more breadth to its customers. “Instead of carrying a multi-year inventory of a set of fasteners, we want to carry smaller quantities of more part numbers,” says Matt Avery, director of sales for the Westfield, Mass.-based distributor.

Its biggest challenge as part of the transition is determining the right level of stock it needs to meet demand. “Our lead times have to be less than a month,” says Avery. “Our manufacturers need up to 22 weeks to make new fasteners. If we don’t have the product on the shelf at the time of an order, that’s a lost sale.” 

There are a number of tried-and- true ways to plan for the demand of parts that are ordered on a regular basis. However, planning for parts with unusual order patterns is a particular challenge. “Some of our parts only get ordered once a year or even once every two years,” says Avery. “And we never know how many of them will be needed when they are ordered.” 

Among inventory managers, parts with highly variable order patterns are referred to as intermittent demand. Managing them can have a significant impact on how much money is invested in inventory and how those funds are allocated. “When parts are expensive and have long lead times, the last thing you want to do is carry a bunch of safety stock to compensate for poor forecasting,” says Ed Wodarski, aviation domain expert for PTC, formerly known as Servigistics. At the same time, he adds, being out of those parts when they are required can ground a plane or extend the time required for a repair. “This is a significant problem for organizations that are trying to forecast intermittent demand using the demand planning tools they use for other parts.” 

Nearly two years ago, Aero Fastener implemented a new class of software developed by SmartSoftware for this type of forecasting. In addition to SmartSoftware, several other software and consulting companies, including PTC and LMI have developed tools for intermittent demand. 

Each takes a new approach to determining the right stock level to meet the investment and service level goals of a parts distributor, airline, MRO or other organization warehousing spare and service parts. Done right, the solutions can reduce inventory levels among parts that fit the intermittent demand profile by 15-30% while improving service levels, according to the solution providers. 

Meanwhile, researchers at the University of North Carolina at Chapel Hill are developing new approaches to forecasting based on market behavior, while others see 3-D printing and advanced manufacturing technologies as a potential solution for intermittent demand in the future (see sidebars).

 

Irregular Intervals

When it comes to managing inventory, there are at least three types of parts: high-volume parts, low-volume or slow-moving parts, and intermittent demand. 

While the first two also bedevil inventory managers, they are predictable. High-volume parts are used on a regular basis and in large quantities. Low-volume parts are used often or in large quantities, but the volume and intervals between orders is predictable. “Low-volume says you don’t use much of a part, but over time, you do get some predictability that you can forecast with conventional tools,” Wodarski says. For example, there may be parts that are replaced only during an extensive check; but an airline or MRO that knows how many of those checks it will perform in a year can plan how many parts will be required for the work and when they’ll be needed. 

With intermittent demand, the size and interval between orders are independent of one another. There may be long periods when no parts are required at all. There is also a high degree of variability in the number of parts needed when an order is placed. For that reason, some refer to them as “now you see it, now you don’t” parts. Other planners refer to intermittent demand as lumpy—plotted out on a graph, the demand for the parts looks like the ups and downs of a roller coaster with long, flat stretches between the hills and curves. 

“These are parts that you think you won’t need until you do,” says Tovey Bachman, a senior consultant with LMI. “You may only need them 50% of the time you do a maintenance event, but the demand is at irregularly spaced intervals and for varying amounts.” For example, during a forecasting project for the Defense Logistics Agency (DLA), LMI identified parts that might be used in a range of one to five units a month, but on occasion as many as 500 units were required; other parts were ordered just a few times in a five-year period, but in quantities that varied from as few as five to as many as 100. 

Most inventory managers forecast their stock levels for intermittent parts using historical averages, just as they would for high- and low-volume parts. With intermittent demand, that can be a hit-and-miss process, because conventional forecasting tools assume that the long periods of time with no demand must be an error. 

Over a two-year period, for instance, the average requirement for a part might be three units a month. To be safe, parts planners put five units on the shelf just in case. While the average use may be three parts, in reality there may be months where no parts are ordered and months where maintenance needs 15 units. “They’re thinking about averages but not about availability,” says Tom Willemain, a senior vice president and co-founder of SmartSoftware. That can lead to extra costs to beef up just-in-case inventory levels, lost revenue for aircraft that are out of service, or increased logistics costs for priority shipping. 

 

New Tools

Since conventional demand-planning software doesn’t work well for intermittent parts, software providers like PTC, SmartSoftware and LMI have developed new tools. While each has its own better mousetrap, what they share in common is an approach that aims to take the lumps out of lumpy demand. The tools are designed to create a smooth forecast for the size of orders and separately create a smooth value for the order interval as well. They do this by looking at demand over long periods of time—as much as five years—to discover more accurate usage patterns and create a baseline forecast. 

That, however, is just a first step. “The smoothing approach is better than sticking a wet finger in the wind, but it still has a wide range of accuracy,” says Wodarski. “It still requires some other techniques to create a more accurate forecast.” 

PTC, for instance, may forecast the total, or aggregated, demand for a part across a network of line maintenance locations, parts depots and MRO facilities. Once the total number of parts is calculated, the inventory is then allocated across the network based on operating conditions. Avionics, for instance, are affected by cold weather. “If I’m looking at cockpit electronics, in the winter I would stock more in Chicago and New York than in Los Angeles and Phoenix,” Wodarski says. The PTC tool also applies root-cause analysis to highlight odd demand patterns that may be the result of improper maintenance or a manufacturing defect rather than just intermittent failure. 

SmartSoftware is developing a replenishment engine that will use the forecast to create a replenishment plan based on constraints. “We will bring in information that is normally inside an ERP or MRO program, such as what quantities of items are purchased, lead times for the part, whether that part is already on order, whether a similar part is out for repair and whether a repaired part is on its way back,” says Willemain.  

In a project for the DLA, LMI used the power of Big Data to forecast the demand for some 330,000 items. “One of the things we found is that the demand for any one item was sometimes so sparse that it was difficult to predict with confidence,” says Bachman. “However, if we did a large number of simulations for 500 or 1,000 items, we could come up with a strategy across a group of parts that was right more often than it was wrong.” That forecast was then mapped against key business metrics, such as customer service agreements, the cost of inventory and the cost of an out-of-stock. “The idea is to give the high-level decision maker the tradeoffs around key business metrics,” says Bachman. “Instead of just forecasting demand, we’re making risk-management decisions.”

 

Implementation

These new applications take time to implement, according to Aero Fastener’s Avery. First, they require a significant amount of historical data—a five-year look back is common. Also, it takes a different mindset for inventory managers used to conventional forecasting and planning. “Intermittent demand is a different way of looking at data,” Avery says. “It requires a cultural change, and it’s a challenge to merge all of the data.” 

For example, while Aero Fastener has been using the application for nearly two years, it is not yet fully integrated with its ERP system. Right now, the software is used to differentiate between slow-moving parts—those that are ordered in small quantities but on a fairly regular basis—and intermittent parts. “The software allows us to identify parts that are outliers and then set the service level we want to offer our customers,” Avery says. “It may cost me $10 million to maintain enough inventory to have a 90% service level, meaning the part is in stock 90% of the time it’s ordered. I might be able to maintain an 80% service level for $8.5 million. With that information, we can decide how much of that part we want to have on the shelf.”

Ultimately, the goal is to use the tool to recommend stocking levels based on investment and service level agreements, set minimum and maximum inventory levels for the warehouse and then integrate that information into the ERP system to automate purchasing decisions. 

“We’re not there yet,” Avery says, “but we hope we’re moving towards making smarter inventory decisions.”