« April 2015 | Main | June 2015 »

2 posts from May 2015

11 May 2015

Supply chain management professionals, let's play Moneyball

Many of you have probably heard of the term Moneyball. It was introduced by author Michael Lewis in his book named, well, ‘Moneyball’. The book was quite successful and it was even turned into a movie starring Brad Pitt

The story was about a small time baseball team in the US trying to survive in the league of superpowers. A guy came along and convinced the owner to apply data analytics to drive up team performance. They relied on statistical analysis extensively to support their decisions on player drafting, using limited budget to buy undervalued players on a cheap and ship out overpriced ones on a high. Turned out the team started doing really well and even finished top of league a couple of times.

So you might say, we are not playing baseball or rather, we are already the superpowers in our market.

The majority of organisations (regardless of their size or market position) are facing similar challenges: increasing competition, customer expectation, scale and complexity of global operations. There is a rising pressure for organisations to be even more efficient in many departments, in particular, their supply chain management. Relying on traditional supply chain execution systems is becoming harder and there is very little room for waste.

So how can the ‘Moneyball’ concept help with supply chain management?

Imagine that you have all the supplier data, manufacturing data, order data and customer data sitting in silos across different departments. Wouldn’t it be nice to have an end-to-end supply chain visibility solution, which allows you to see everything that’s going on with only a few clicks on your laptop, tablet or mobile device? Or even better, by implementing an attribute based product segmentation tool, you can zoom in to track supply chain behaviour on a product segment level and tailor supply chain strategies for your products.

Imagine again that you are regularly monitoring your supply chain operations, you might be thinking: ‘gosh, it’s such a big beast, I want it lean! But where do I begin?’ How about considering a real time data driven diagnostics tool to help you? A tool that builds learning algorithms around your data that:

1)       alerts you to saving opportunities and operational risks, in real-time

2)      informs what you can do to improve your supply chain operations

3)      tells you how much improvement you can benefit from each action

Data and analytics driven supply chain management is a key capability for an organisation to be successful. In a world where everybody is talking about ‘Big Data’, the story of Moneyball tells us that success doesn’t come with just sitting on the data. Put faith in data analytics and develop the capability to execute the optimal analysis on the right datasets can reap infinite benefits. 

If you would like to discuss these issues, or the impact of emerging technology or data and analytics on your industry, then contact our Data & Analytics team.

07 May 2015

The importance of getting Demand Forecasting right

The financial reporting season for retailers, Fast Moving Consumer Goods (FMCG) companies and car manufacturers has brought some interesting features to light. The ability to gauge consumer demand accurately is a key theme.

For retailers, overstocking (especially leading up to the critical holiday trading period) leads to under-utilisation of floor space and clearance sales at a later point in time, eroding margins. Understocking leads to unfulfilled demand in the short term, which in turn leads to loyal customers either having to try a different store or on-line fulfilment, or (worse) jumping across to a competitor with a substitute product.

For FMCG companies, over-estimating demand leads to over-production and then the need to run down stocks, while factories are operated at lower capacities. Conversely, under-production leads to short term supply shortages, causing distributors and retailers to seek out alternative suppliers or substitute goods, which may lead to more damaging long term consequences.

Car manufacturers face broadly similar issues, albeit customers are more used to the idea of waiting for delivery of personalised vehicles in terms of specification and options, particularly in the premium market segment.

Against this backdrop, what can retailers, FMCG companies and automotive manufacturers do to try to hit the sweet spot and maximise shareholder returns? A top down and bottom up approach could be adopted as follows:

  • Macro level demand forecasting – using relatively aggregated data – can seek to address the impacts of economic and other external factors to provide forecasts by product at a headline global or major geographic region level, which then feed into business planning.
  • At the same time, predictive analytics, using much more granular data, has a critical role to play in determining likely future demand at the country-specific and regional in-country geographical level. This is achieved through explicitly modelling the underlying drivers of demand for any given product.

Industry experience, local knowledge and expert judgement also provide critical inputs to both of these approaches, naturally.

Macro level demand forecasting can leverage a number of well-established disciplines, including econometrics, statistics, Monte Carlo simulation and actuarial techniques. We can assess changes over time across a range of indicators including:

  • the macroeconomic environment
  • cost of raw materials
  • regulatory or government actions (for example significant changes in levels of tax or duty on goods)
  • attendant geo-political risk,
  • level of consumer confidence,
  • level of average real wages,
  • foreign-exchange movements (which can materially impact levels of imports),
  • demographic and social trends

By using stochastic modelling techniques, ranges of possible outcomes and probabilities of outcomes can be determined, and provide management with valuable additional information.

For FMCG companies and car manufacturers, such forecasts can then feed into high level decisions around raw materials and key components purchasing, and in relation to required production and distribution capacity by geography. For retailers, these projections can feed into decisions around requirements around overall in-country floor space and distribution centre capacity.

Correctly forecasting demand for individual products at a sub-geography level, and consequently optimising stock levels, is highly data-intensive. One way of doing this is to model future customer behaviours, which I discussed in a previous blog. But this approach tends to fare less well for businesses where the proportion of customers for which this is possible is relatively low.

Instead, an alternative approach is to build generalised linear models or use either Bayesian Inference techniques or machine learning, where the objective is to fit historical sales data to algorithms with multiple variables in order to predict future demand. Many of these variables can be calibrated from internal data (for example  sales volumes by brand, types, colours and sizes of garments; car sales by make, model and engine size).

The ability to incorporate variables which reflect regional external drivers (such as socio-economic and demographic trends, the competitive landscape, average non-mortgage household debt, average disposable income for the geographical region in question) by leveraging external, third party data sets, suitable proxies and even social media sentiment, will improve the goodness of fit of the models and hence their predictive capabilities.

Naturally, these algorithms will never achieve 100% predictive accuracy, and understanding the level of volatility of historical demand levels is also important. This will in turn allow statistically derived buffers in required stock levels to be established.

I should also stress that such models should be seen as important tools in making key decisions, and will never replace expert judgement. To illustrate this, as these models are highly reliant on timely data, forecasting demand for new products will be better served in the short term by other approaches, until sufficient data is captured. In relation to existing products, it will also take a number of months before a new trend becomes statistically significant, whereas astute individuals with their finger on the pulse can get ahead of model predictions, and steal a competitive edge. Harnessing both of these capabilities to improve demand forecasting can deliver a big step forward in financial performance.

If you would like to discuss these issues, or the impact of emerging technology or data and analytics on your industry, then contact our Data & Analytics team.