SUPPLY CHAINS HAVE TRADITIONALLY OPERATED IN THE BACKGROUND, but in the past few years they have become front page news. From a global shortage of electronic components to supermarket shelves without toilet rolls, flour or salad, a fast-food chain without food and a container ship stuck in the Suez Canal, the importance of supply chains to our global businesses and people are now clear.
These issues have highlighted the risks that are embedded in our supply chains. Major disruption happens, and more often than most people consider. But how do we go about designing supply chains, ideally without hitting the front page of the popular press?
Supply chains are a series of nodes that connect raw material suppliers via factories, warehouses and transport links to end consumers. That is a relatively simple statement, but typically there is a lot of detail behind this. In the typical literature, it is often represented in a linear fashion, but the reality is often a spider’s web of interconnected supply chains.
The science of these supply chains, at a static level, is pretty well understood, such as optimisation models for network design and transportation routing and simulation models for stochastic problems in factories and warehouses. But models are based on data and assumptions, and the question very quickly becomes how well these reflect reality.
PRACTICAL CONSIDERATIONS
Let’s take network locations as an example. These are typically used to assess where to construct new warehouses and will take inputs such as customer and supplier locations, demand and transport distances. Since demand is often correlated against population, running these models for different businesses typically gives similar outputs, leading to a concentration of warehouses around the M1 between Northampton and Leicester. shows an example for Magna Park Lutterworth, with around 800,000 m2 of warehouses, but relatively limited access to a local labour pool.
FIGURE 1 AERIAL VIEW OF MAGNA PARK LUTTERWORTH (SOURCE: GOOGLE EARTH PRO)
The models typically ignore soft requirements (such as labour and skills), but also often don’t clarify the relatively sensitivity of location to the overall solution. At the same time, moving a location 40 miles north or south may not have a material effect on the solution, but a two-mile diversion from a major transport network can have a huge impact. For a major retailer, this ‘minor’ change would add nearly 750,000 lorry miles per annum.
Equally, it is easy to assume that data is widely available but getting timely and accurate data across a multi-tier supply chain is very hard. The reason is not that the data doesn’t exist, but that the data is proprietary and different entities in the supply chain have different levels of visibility. And since most supply chains consist of businesses with interlinking commercial relationships, this data is sensitive.
As a consequence, whilst it seems easy to draw a supply chain map, it is much harder to extend this across multiple tiers. We saw this when the Rana Plaza building in Dhaka collapsed in 2013. Several retailers did not know that their products were produced in these factories, partially due to a long list of commercial contracting and subcontracting relationships. Whilst this may have initially resulted in a low-cost sourcing of products, the ultimate cost in human lives (and resulting brand damage to the retailers) was huge.
Long supply chains are typically not static. Businesses change suppliers, or relocate to a new facility, and transport providers change which nodes they use (especially relevant for rail, sea and air transport, where different terminals might be used). The models would need to keep track with the levels of change, and in particular sourcing options can change both frequently and potentially temporarily.
the length of supply chains, with large number of nodes, typically results in models that are mathematically and computationally challenging
Finally, the length of supply chains, with large number of nodes, typically results in models that are mathematically and computationally challenging. At BearingPoint we use Gurobi as our mathematical optimisation engine, which combines a best-in-class engine with the power of cloud computing. But we still hit limits to what is an acceptable model complexity and resulting run-time.
BALANCING DIFFERENT ELEMENTS
In order to model supply chains effectively, we therefore need to blend the rigour of the data and modelling with a practical application of those factors that are either unknown, changeable or difficult to model. That doesn’t mean that we can’t use data to assess some of these factors, it is just treated outside the core model.
As an example, we have recently completed the global network review for a business with a large product range and a relatively short order lead time. A key factor in the decision making was how we could balance cost, investment in inventory and infrastructure and service levels. This is a fairly typical application, where advanced models can support the decision making. “The power of combining the quantitative modelling with the qualitative assessments resulted in a series of workshops where the key stakeholders could discuss, using facts and data, what the best options are to balance cost, investment and risk for our future supply chains”: CEO, global distributor.
To start, we constructed a set of conceptual supply chain models. In simple terms, these were used to assess at a high level what a realistic set of options might look at. It very quickly became obvious that the inventory investment, based on the number of products, makes it very difficult to split the range over a large number of stock holding locations. This allowed us to constrain the model to a smaller number of options.
We then used segmentation analysis to split the range based on a series of characteristics, such as value, cube and weight, rate of sale and life cycle risk. This allowed us to further restrict the number of options into a more manageable set of options, again based on the criteria of cost, investment and service.
Finally, we ran the models to understand the optimum solution, balancing the costs and investment whilst meeting a minimum service requirement, then again with increasing service requirements to understand the sensitivity. This resulted in a series of scenario outputs (each individually optimised), which could then be assessed against the overall scoring.
In parallel, we developed a series of ‘qualitative’ scoring areas, and then assessed each scenario on these. Some of these categories can take data driven inputs, such as taxes and tariffs and political and disruption risks. Others are more subjective, such as freight availability and some risk factors.
shows one of the elements of this evaluation, highlighting both quantitative and qualitative measures against a series of options.
FIGURE 2 EXAMPLE EVALUATION MATRIX
Ultimately, it is important to recognise that there typically is not one optimum solution; different stakeholders will put a different focus on the different elements. It is therefore required to review both the analytical elements (where there may well be an optimum, but it depends on the inputs) and the softer or more qualitative elements.
it is important to recognise that there typically is not one optimum solution
BLENDING ART AND SCIENCE
The requirement to use non-numerical data does not mean that we should not use the optimisation models. Rather, it should remind us that these models can be very useful but need to be treated with a degree of caution: they are limited by their complexity, by the difficulty to get perfect data and by the fact that there are factors that are not in the model.
The key element is to take the models, ensure they represent the actual underlying problem and then to test the outputs, understand their sensitivity and to be very clear about other factors that are important decision criteria but may not be in the model.
Finally, it is also worth remembering that your competitors are running the same models. Sometimes, you can get a competitive advantage by doing something that may not be ‘optimum’ but allows you to differentiate yourself.
Designing supply chains is hard, but by combining the power of analytics with a real-world understanding of the limitations of the model, we can deliver results that are both cost effective, more environmentally friendly and with the right balance of service and risks.
Additional information
Notes on contributors
Emile Naus
Emile Naus is a Partner in BearingPoint. He has worked in supply chain across both operational roles and consultancy projects across the globe, with a background in analytics (https://www.linkedin.com/in/emile-naus/). BearingPoint is a 5,000 strong consultancy with a focus on business and technology (www.bearingpoint.com).