Data Mining Interview: Marcel Baumgartner
Data Mining Research: Could you introduce yourself? What is your journey in Analytics?
Marcel Baumgartner: My name is Marcel Baumgartner. I am one of these Swiss Germans who moved across the “Röstigraben” to live in the French speaking part of Switzerland. I graduated from EPFL, the “little” sister of the ETHZ, in mathematical engineering, and then finished a Masters in Statistics at Purdue University in the US. My first job was at Nestlé, and as for many others, Nestlé has remained my employer. I started my career (1995) first in Nestlé’s R&D organization, in its heart at the Nestlé Research Center near Lausanne. In my role as a consultant in all kinds of statistical matters, I had the privilege to more or less apply what I was taught in school. We basically explained our fellow researchers (biologist, nutritionists, chemist, food engineers, …) what to do with all the data they gather and generate day in, day out. One of the strengths of this still existing group is the use of experimental designs, and making sense of high-dimensional data using multivariate analysis methods like Principal Components or Decision Trees.
In 2001 I needed a change, and two colleagues from EPFL told me that Supply Chain Management needs a bit more statistics. A job opening was available in Vevey, in the headquarter of Nestlé, and I joined the Demand & Supply Planning team. In Vevey, we are responsible for corporate guidelines, best practices, appropriate tools and vision. I don’t do myself any operational planning.
Around 2003, Nestlé got access to systems from SAP to support the work of Demand Planners. Their role is to foresee the future orders of our customers, as we very rarely can produce based on orders: we need to build stock, therefore we need accurate forecasts. Through SAP and their module Advanced Planning and Optimization (SAP APO), we suddenly had exponential smoothing routines within our toolbox. So I jumped on this occasion, learned all about it, provided training material, gave courses, and step by step we have learned how we can use these statistical methods to our advantage.
DMR: What is demand planning about? What is the difference between forecasting and planning?
MB: Demand Planning is about ensuring that the right product is at the right place, with the right volume, at the right time. The goal is to have enough so that we deliver exactly what has been ordered (internally we say that “any unfulfilled order is a crime”), but naturally not too much, as inventory is costly: it ties up capital, uses space, and products get old and eventually become “bad goods”. Freshness of our products is key: a chocolate consumed a few weeks after production is so much better than if it is already 6 months old.
Forecasting is about a honest outlook of what will happen under the current assumptions. For example, assuming that the trend continues, that the seasonality is stable, and that the investment in the forthcoming marketing campaign will really materialize in more demand. Basically, we want to be able to identify gaps to our targets. Planning is then more about closing this gap: if our forecasts say that we are missing 2% growth, then we plan the actions that we need to close the gap. This becomes our plan.
DMR: What are your current challenges at Nestlé? Anything you can share?
MB: The challenge is that we need to become more efficient. Our typical Demand Planner is still doing way too much manual work, and therefore cannot focus on his or her value added tasks. A value added task is to have a very good relationship with the Sales team, and always stay informed about what’s going on with promotions and other customer related activities. A non-value added task is to come up with a forecast for a product that can be forecasted well statistically. This is the transformation we are going through.
We have now a very good understanding about the part of our portfolio can be forecasted statistically, or as we also call it, analytically. We now need to translate this into action ! What we will do is to create analytical competence centers that will provide these statistical forecasts as a service, so that Demand Planners don’t need to worry about the details. The members of these analytical competence centers will be called Demand Analysts, and they will use state-of-the-art forecasting software. They will then provide reliable forecasts, and ensure that the Demand Planners adopt them widely.
Another challenge is to find the right blend of people, brains that are good in statistics (not experts, simply people who know what can, and more importantly, what cannot be done with statistical methods), that are good in data management and finally individuals that can talk and live business. The new name for such talents is Data Scientist. All this in a context where hiring new people is very difficult.
DMR: Do you have an advice for people involved in business forecasting?
MB: There’s a huge potential of forecasting performance improvements, both on accuracy but often more on efficiency. Today’s forecasting products (e.g. SAS Forecast Server, ForecastPro, …) are amazing in performance, scalability and automation. The challenge is to identify what part of your portfolio can take advantage of such tools. Your historical demand volatility is a key driver: low volatile products (typically measured through a coefficient of variation, adjusted for trend and seasonality) will have high performance, high volatile products need more business insight.
Then don’t underestimate the difficulties you will face to ensure adoption of this approach. If your company does not have a strong analytical mindset, you will need to spend a lot of time to gain the trust. Fortunately, you have the right tool: facts ! In forecasting, it is straightforward to show how good you are: simply eliminate a few periods of your history, simulate the performance of your methods, and you will have a very honest and unbiased estimate of your future performance. With these facts, you will convince all your managers not only of the power of these statistical, sorry analytical, methods, but also of their beauty !
Marcel Baumgartner can be joined at email@example.com