Textbooks cover the basics of forecasting and inventory management but, as Malcolm Wheatley relates, the scale and speed of aftermarket parts supply requires increasingly complex IT solutions
When Piaggio Group Americas had problems with aftermarket spares availability, incoming director of spare parts Marco Ciccolini worked long hours expediting inbound replenishment shipments and dealing with backorder complaints from dealerships selling the company’s iconic scooter brands.
With service level fill rates that had dipped as low as 50%, the company was relying too much on expediting shipments via air freight. Worse still, high levels of very slow moving or obsolete stock had resulted from an inability to distinguish regular demand from specialised intermittent demand.
What to do? To begin with, explains Ciccolini, implementing the basic inventory management and forecasting functionality contained within the business’s recently acquired SAP R/3 ERP system helped to improve performance. Service levels quickly rebounded into the 70%-80% range, albeit with erratic and inconsistent peaks and troughs.
But prompted by dealer concerns to initiate a 15-day guaranteed replacement part policy, Ciccolini decided to implement a system that offered the prospect of 90% service levels on a more consistent basis–ToolsGroup’s SO99+, a forecasting planning solution that was already used to manage and optimise the inventory for approximately 180,000 aftermarket parts in Piaggio’s European divisions.
The plan was for the business’s SAP R/3 system to feed data into a parts database and inventory file so SO99+ could take this information, apply analytic forecasting models and generate replenishment proposals which would become the basis for parts orders, within SAP.
Within just two months of implementation, the impact of SO99+ was evident. Service levels across Piaggio’s brands leapt into the 90% range–sometimes exceeding 95%–and, most importantly, remained highly consistent.
The company was able to dramatically reduce its reliance on air freight, and increase the ratio of standard ocean shipments compared to flying. Even better, despite the significant rise in service levels, Piaggio’s inventory turns improved, evidence that it was now stocking the right parts to suit demand.
Piaggio’s experience is a classic case study of the relationship between forecasting and inventory management, and a lesson for other manufacturers–get it wrong and the result will be, as at Piaggio, too much inventory of what isn’t required and too little of what is in demand, leading to low inventory turns, low customer service levels and high expediting costs.
Get it right, and what’s on the shelves corresponds with what is being demanded, and in the quantities that are required. Inventory turns and customer service levels promptly recover, while expediting costs plummet.
It isn’t rocket science. Forecasting and inventory management are covered in university undergraduate courses, and IT has been harnessed to the task since the 1960s. Most ERP systems provide the requisite capability as standard.
The trouble is, the real world doesn’t work how textbooks suggest it should–especially in the automotive industry.
Automotive aftermarket demand, for instance, doesn’t necessarily follow a predictable normal ‘bell-shaped’ distribution. Nor, given the industry’s long supply chains, are the lead times as stable as might be expected. Service level expectations are also higher in the automotive industry than in industries such as consumer goods, because the point of final demand can often represent a vehicle that is off the road, awaiting the arrival of the spare part in question.
There are other peculiarities in automotive. Distribution is multi-echelon, with national, regional and local stockholdings. Local stockholdings can also exist in parallel to independent distributors and OEM dealers holding the same part in stock, within the same city–and competing on price and availability.
“It’s a complex, multi-tiered environment,” says Michael Martin, vice president of business development for the automotive and industrial sectors at DHL Supply Chain, which handles aftermarket logistics for Volvo, Mitsubishi, and Mercedes, and operates a shared-user distribution service for tyre manufacturers Pirelli, Cooper Tires, and Hankook, with tyres for all three being transported on a single fleet. “So much depends on how automotive manufacturers choose to configure their aftermarket supply chains, and where they want to place the burden of forecasting.”
The inevitable third-party dimension complicates life in the shape of equivalent parts, too. Is the OEM-specified part out of stock? No matter, an often cheaper, generic third-party equivalent will do the same job. With cost-conscious fleet operators and consumers increasingly aware of the third-party option, predictable demand patterns can quickly become less so, especially in the face of price promotions.
There are also ‘macro’ factors such as fleet characteristics, automotive engineering trends, vehicle life and so on.
“Aftermarket volumes are in long-term decline, due to a number of trends and structural factors, such as elongating service intervals on new cars, increased reliability and the slow but inevitable shift in parc mix,” says Ben Waller, senior researcher with UK-headquartered international automotive industry membership-based research body ICDP.
Modelling analyses undertaken by ICDP, for instance, estimate that the impact of these trends on future aftermarket demand in the parc of the main four EU markets will be an average decline in operations per vehicle of around 13% from 2009 to 2015–although this average sees significant differences by individual market and brand.
Industry insiders also say that parts usually follow a demand lifecycle that means there is a tendency to underestimate demand as the volume of vehicles on the road increases, but also an equivalent tendency to overestimate demand as the number of vehicles starts to decline.
That is because as the fleet grows, historic demand history is based on fewer vehicles than are actually on the road. In an ageing and declining fleet, a forecasting system looks at a demand history based on a vehicle fleet larger than that which will generate parts consumption going forward.
Seasonality and geography play a part, too. The demand for air conditioning components in Florida, for example, will be greater than in cooler northern states, and skewed towards the summer months. Other demands, though, will be weatherdependent, such as snow tyres, for instance.
It’s clear that even if techniques such as ERP-based exponential smoothing could deliver accurate projections for high-volume items, they are far from ideally suited to predicting the demand for low-volume ‘failure mode’ items, especially in circumstances where there is a limited demand history from which to project.
These challenges make automotive aftermarket forecasting one of supply chain management’s more complex operations. So how is the automotive industry dealing with these issues? In 2012, what constitutes best practice? What role is played by 3PLs? And what opportunities are there for improvement?
Almost inevitably, specialised niche forecasting solutions come typically, but not always, from niche vendors.
Smart Software, for instance, co-founded by former university academic Charles Smart, is one software firm tackling the problem. Its flagship product, SmartForecasts, uses a patented empirical approach called ‘bootstrapping’ to rapidly generate tens of thousands of future demand sequences scenarios, building up cumulative demand values over an item’s lead time.
The major challenge, says Charles Smart, is forecasting intermittent demand: while conventional statistical forecasting techniques such as exponential forecasting may deliver solutions–even if sub-optimal–for items in moderate demand, they fail almost completely when demand is intermittent.
“And that’s the problem,” he says. “In many automotive businesses, there’s a huge ‘tail’ of items that are intermittently demanded–perhaps 60% or 70% of the total SKUs. And the more expensive the item, the less appropriate conventional techniques become. Take axles, for instance: you don’t want to be out of them, but nor do you want to be holding more in inventory more than you need. They are just too expensive.”
Martin Woodward, MD of ToolsGroup UK–whose SO99+ forecasting transformed Piaggio’s aftermarket–agrees.
“In automotive, the overriding characteristic is the very long tail of very slow-moving and obsolete items,” he says. “It needs managing, and it needs managing at scale, as volumes are just too high for manual intervention.”
American bus manufacturer, Motor Coach Industries, is a SmartForecasts user. The company points to a parts inventory of more than 200,000 individual items, of which just 40,000 parts are active, meaning they have experienced demand at some time during the prior calendar year.
“The sheer magnitude of our inventory is staggering,” says Stan Dzierzega, executive director operations at Motor Coach Industries. “Planning is especially challenging when you consider that we have 13 distribution centres, each with high service level expectations.”
Even at the tier one level, points out Woodward, the scale of the challenge requires automation. At ZF Industries, for instance, where SO99+ is managing aftermarket forecasting, the package sits on top of an in-house ERP system managing 12,500 items in 17 warehouses and 160,000 SKU locations.
It’s similar at SKF Vehicle Service Market, the aftermarket parts division of Sweden’s SKF, where SmartForecasts reduced inventory by 16% while maintaining or improving service levels. Demand planners couldn’t cope with the 60,000 bearings, seals, and U joints the business stocks, says Matthew Schiele, SKF’s vehicle service market supply chain manager.
If it looks like a fish, it might not be a fish
But if niche software packages are the answer, they needn’t necessarily come from niche software vendors, which is some comfort, perhaps, to automotive industry giants wary of overreliance on small companies.
Mighty enterprise applications and database vendor Oracle, for instance, is keen to showcase its automotive aftermarket expertise, pointing to successful implementations at 3,000-branch retail parts distributor Carquest; the aftermarket operations of truck and defence manufacturer Navistar; and a slated deployment at Mazda. The software is Demantra Advanced Forecasting and Demand Modelling, a capability acquired in 2006.
“The mistake that companies often make is trying to predict demand at a distribution centre level,” says Andrew Spence, Oracle’s pre-sales manager for manufacturing, retail and distribution. “We try to model the supply chain, rather than attempting to forecast and manage a single echelon at a time.”
But whatever the vendor, it seems that best practice is moving away from forecasting using a prescriptive algorithm– such as exponential smoothing, or even regression–to ‘best fit’ approaches based on iterative techniques that aim to make the data itself delineate the algorithm that describes it best.
“There isn’t a silver bullet,” says ToolsGroup’s Woodward. “It’s about having a toolkit of techniques, and working with the ones that are most appropriate. And it’s about using a tool that’s appropriate for the problem you’re trying to solve: OEMs have one set of issues, distributors have another.”
Charles Smart, for his part, cautions against the use of specific statistical distributions. “Intermittent demand is typified by an asymmetric distribution with a long tail over to the right,” he says.
“It might look like a Poisson distribution [a statistical distribution used for predicting low-frequency events], but we don’t specify a Poisson distribution. We carry out tens of thousands of simulations and construct a distribution, letting the data speak for itself.”
And while Oracle’s Demantra is more structured, offering nine forecast models in its Demand Modelling application and a further six in the Advanced Forecasting module, the basic logic is the same: self-learning and self-adapting algorithms that combine the different models to maximise predictive accuracy–rather than historic fit–enabling a faster and more accurate response to changing market conditions.
That said, ToolsGroup’s Woodward cautions against an over-reliance on the mathematics underpinning the various algorithms that the firm uses.
“We often get asked how our forecasting application works,” he says. “But that really isn’t the point. It’s like Google–no one asks how the algorithm works before they use it. They use it, and if it works, they carry on using it.”
All of which could be said of carmaker Jaguar, which for more than 20 years has relied on Unipart Logistics to provide it with a full automotive parts service, including the sourcing, storing, processing and despatching of parts from 16 warehouses– located in the UK, USA, Canada, Germany, Spain, Russia, China, Japan and South Africa–to more than 700 Jaguar dealers in over 60 countries.
“There’s an expectation of same day delivery in major conurbations, and next day delivery in others,” says Chris Roberts, Unipart Logistics’ global account director for Jaguar Land Rover. “Typically, we see requirements such as ‘Order by 6pm, deliver to a lock-up facility by 8am the next day.’ In terms of the logistics, it’s an inventory holding strategy just as much as a distribution strategy.”
What’s more, he adds, “You need strategies for handling fast moving items, and different strategies for handling slow moving items. Specific storage facilities for the slow movers, located perhaps some way from the main picking face, and concentrated and efficient storage facilities for the fast moving items, in order to minimise movement times and waste.”
And this complexity is exacerbated by the need for some facilities to not just store incoming parts, but also carry out work on them, in a process that he characterises as “semimanufacturing”.
“Around 90% of the incoming parts that we receive have to be processed,” he explains. “We receive the parts, inspect them, pack them and label them–as well as carry out kitting operations, where we’ll pack several components together as a specific service kit.”
He adds that it’s usual for bulk items to be shipped to distribution centres without retail packaging–nested together to use shipping cubic volume more efficiently.
To meet Jaguar’s demanding customer service targets, slick forecasting is vital and Roberts says that Jaguar tasks Unipart with undertaking it. What’s more, the software in question remains Unipart-built, despite several attempts over the years to identify equivalent third-party alternatives.
“Our core enterprise system is SAP,” he relates, “and we looked at SAP’s Advanced Planning and Optimisation forecasting product in 2000, and its then-new Service Parts Management module in 2007–even working with the latter as a beta-customer. But we couldn’t get it to work as we wanted and abandoned the attempt after several months.”
Instead, the company relies on a solution it built itself back in the early 1990s, although it has been developed and enhanced since then. Sophisticated algorithms–including the use of the Poisson distribution for slow-moving items–are part of the rationale, he explains, but only part.
“At Unipart, we’re very into visual management, and the forecasting system reflects that,” says Roberts. “Given the volumes we’re dealing with, it’s exception-driven, and issues alerts when excessive demand or low stocks breach set limits. It displays all the information that anyone needs in a very graphical format, allowing them to re-model the forecast using the very latest demand data. SAP is very good, but to get to an equivalent decision you had to look at several screens–and we don’t have time for that.”
And the results, he reckons, speak for themselves. Jaguar drivers across the world experience the same excellent levels of aftermarket service and support, regardless of their location, a continuity of service and support that helps drive brand value and brand loyalty across global markets,” sums up Roberts. “Forecasting is sometimes seen as boring, but we believe that the benefits of good forecasting go right through to the bottom line.”
