Sharpening the operational supply chain
Demand forecasting in defence has been for the most part, based on past consumption - a bit like driving forward by looking in the rear-view mirror.
The result has been costly, often with expensive overstocking, and in a fast moving theatre, 'just too late' or 'just in case' rather than 'just in time'. Commercial enterprises on the other hand have been far more transformative in their approach to demand forecasting, using advanced cognitive analytics or 'machine learning' to exploit the data it holds. The next step for supply chain strategists in defence is to use probabilistic analysis to inform its day-to-day demand forecasting.
Most defence demand forecasting has traditionally looked backwards at empirical data to identify future consumption. Yet actual defence demand patterns often look more like chaos than linear, predictable requirements. They are increasingly influenced by multiple internal and external factors that drive them up and down in ways that can't be understood by simply looking at past campaigns and wars.
Defence Forecasting - Why so difficult?
In Aerospace and Defence (A&D), the gold standard has been not deterministic, but Stochastic/Probabilistic optimisation of logistics and its end-to-end supply chains. This has primarily been focused on Initial Procurement (IP) and on rotables/repairables as this is where the big dollars sit.
As more Original Equipment Manufacturers (OEMs) look to move from manufacturing to through-life support operations, the need for better through-life demand forecasting – driving profitability from the minimum inventory whilst maintaining or even improving acceptable service levels – has increased. Optimised inventory holdings have gained a greater commercial impetus and the old 'run IP and let the users buy sustainment stocks' process just won't do if profit margins are to be maintained.
The traditional deterministic approach simply does not survive the analysis of the maths.
Enter Probabilistic Analysis
Probabilistic analysis has the potential to provide model outputs in various ranges of figures with calculated degrees of confidence, and takes into account imprecision and uncertainty when appropriate. Factoring in the probabilistic impact of variability in demand or supply plans in recommending inventory levels, accounts for the likely variation across a range of factors that make up the demand forecast - rather than simply multiplying past performance by a set number.
The ultimate aim is to balance investment in inventory against the resultant improvement in service levels. The bottom line is ‘the graph’ is not a straight line but more continuous probability curves. This model has already been used successfully in the commercial sector and there is a lot to learn from this.
How can defence learn from probability in the commercial sector?
The military’s perception of commercial manufacturers and suppliers is that they have it easy. They can choose, and know their markets and customers, a luxury the military don’t often have with their enemies. Unlike those in uniform, commercial enterprises are not subject to short-notice unexpected exercises and operations brought about by external political decisions. They are not constantly introducing new equipment with different support packages. They don’t have to grow new, complex supply chains to cope with new theatres of operations.
Yet there are similarities between what defence consider to be their special challenges, and those experienced by a series of real commercial enterprises.
• Short Notice Exercises and Operations - Some commercial enterprises typically run thousands of consumer promotions a year - these can be compared to the changing consumption patterns caused by military activity. One such company’s marketing effort creates unusual forecasting scenarios for 30-40,000 individual stock numbers, each in effect with their own ‘promotion’ or changed consumption pattern. It is vital to commercial survival that firms forecast the uplift in demand well.
• New Equipment Lines and Upgrades - One global electronics distributor routinely has 500,000-plus in-line items - this aligns with figures for the UK MoD, which must manage an inventory of 900,00 different lines. The company introduces 5,000 new products every month and fulfills more than 44,000 same-day orders every day from its operations in 32 countries. Predicting demand for such a vast array of new products is more than any demand planner can reasonably be expected to handle, especially if systems can only look at their own past consumption patterns for similar products.
• Increasing Operational Complexity - A US manufacturing firm had to manage a rapid expansion of its distribution network while introducing regional distribution centers to its supply chain structure. Like the military engaging in a new operational campaign, it had to implement this change while maintaining high service levels.
These commercial organisations acted rapidly to implement a probabilistic approach - something defence logisticians can learn from.
Not Too Late, rather, Just In Time
The companies in the scenarios above all had complexity and scale that made it impossible for their demand forecasters to plan accurately, much like we see in the defence sector. Current deterministic forecasting systems are just not accurate enough, so planning staffs only engage at the end of the process to put right clearly inaccurate stock levels.
This leads to inefficient ‘fire-fighting’ to meet priority requirements rather than inputting the required factors and drivers at the start of the process. While market or ‘operational’ feedback is valuable, constantly manually adjusting forecasts leads to bias introduced by human opinion and lack of trust in the system.
What should be a scientific process rapidly descends into a ‘blame the supply chain’. The result is inappropriate over-provisioning and continued poor demand satisfaction as the next problem is uncovered.
The solution adopted by all of the commercial companies above was to turn to a flexible probablistic analytics model in their day-to-day operations, rather than simply IP and initial stocking. This improved their forecasts, reducing inventory costs and increased service levels. It also reduced the burden on their demand forecasters and inventory planners who were able to concentrate on providing world-class solutions, rather than producing expensive fixes for stock shortages.
The 'Just in Case' scenario – probability in action
During the 2003 invasion of Iraq, the commander of a coalition Armored Division demanded that the equivalent of five ‘turret loads’ of ammunition be positioned forward in the theatre of operations for each battle tank and armored vehicle. Simple operational planning and the number of available targets indicated that this was too much ammunition by a significant factor. All of this ammo would need to be moved, maintained and accounted for.
With no trusted science to back up the logistic experts’ assumptions, 25 old staff tables and inherently conservative planning tools designed nearly 40 years before were used. When the smoke cleared and the Armored Division vacated the battlefield, the logisticians had to recover and return or destroy the equivalent of more than one turret load per armored vehicle in theatre.
This is a perfect example of how poor deterministic forecasting erodes confidence in the supply system, leading to inflated safety stocks and over-demanding at the front-line. It is clearly time to consider the adoption of non-deterministic technology in support of their demand forecasting staffs.
Where next - enter machine learning
Probabilistic systems have traditionally had challenges around the theoretical and practical problems of data acquisition and representation. The next step is to understand what is happening at each stage of the supply chain and at all inventory locations rather than taking a simple 'whole inventory' view. With more granularity, the factors that are relevant and actually make a difference can be separated out from the 'noise', and issues such as differing velocity at various points in the supply chain can be taken into account in the analysis.
Enter machine learning. This is a computer-based discipline in which algorithms can actually 'learn' from the data. In simple terms, machine learning systems capture and model all of the relevant factors that shape demand whilst filtering out, or ignoring, random and unpredictable fluctuations - noise. Because of this, they can manage forecasting models that use many different types of data, just like those in defence. Once the system knows which indicators can be related to success, it will update the model over time and change alongside behavior. In defence, this could be linked to higher operational and maintenance schedules so that detailed future consumption patterns can be forecast.
The overall aim must be to model the real life world and to improve operational performance, rather than re-engineer it.
Realising the benefits
However, to realise the benefits of moving from a deterministic to a probabilistic approach, or even to machine learning and drive the forecast into day-to-day processes and avoid leakage of strategy, organisations need to integrate this approach with a modern, agile application with enterprise breadth. The approach also needs to be linked to the MRO processes, which generate much of the demand, so it becomes a two way process.
The agility of modern Enterprise Resource Planning (ERP) software – with its inherent modular capabilities – removes the time and pain required to adapt new processes compared to traditional, more monolithic ERP systems.
These ERP solutions are designed to rapidly adjust to developments in technology. The breadth of such a solution enables reliable and accurate forecasting information to be fed back into planning, projects and operations within the enterprise solution. This allows optimisation of the end-to-end maintenance process and parallel actions across the wider organisation, making life easier for the equipment managers, enabling them to understand the effects and impact of demand drivers such as operational plans and new product introductions, as well as implementing a more efficient and effective support chain.
Perhaps at last, the right stock, in the right amounts, in the right place – at the right time.
