00:27 Warren, perhaps you could start by telling us a little more about yourself?
02:00 What are the uncertainties you can observe within supply chain management?
04:37 Why are variations in time frames interesting? What do those variations mean from a technical perspective?
08:48 How do you measure how good the decisions you are making are?
12:04 Does simulation actually work?
18:12 How about building machine learning algorithms to take policy based decisions?
22:33 How much confidence can we have in these policy based approaches?
29:38 When you are building policy based forecasts, how do you get people to visualize them?
30:48 Joannes, how would you compare your journey to the one of Warren?
33:43 Warren, what are your hopes for the future? Can you see everyone using the policy based method one day?
In supply chains, there’s often somewhat of a catch-22 between the optimum decisions you can take today and how these can affect the decisions you can take tomorrow. For this episode of LokadTV, we’re delighted to be joined by Warren Powell to discuss the difference between policy and point forecasts and how these can be used to optimize those catch-22 decisions.
Warren was a Professor at Princeton University, where he taught for over 39 years, he also founded CASTLE Labs in 1990, working directly with industry to help solve their problems with computers. Together with the students who were part of the lab, they published over 250 publications. From this, three consultancy firms were born, one of which is Optimal Dynamics, which Warren co-founded and works at as Chief Analytics Officer.
There are a vast litany of uncertainties when it comes to supply chain management. Additionally, wide variations in time frames when it comes to decisions is a very important element. But to really get to the core of the problem, you have to think about the sequential decision making process. Here, it’s the future that is shaping the past, which feels very wrong and can be hard to get your head around.
What we mean by this is that the decision that you wish to optimize in the present depends on the decision you will take later. For example, you order from an overseas supplier who has an MOQ (minimum order quantity). So you order many different products from the supplier to reach a full container - is it a good order? This entirely depends. When will you pass the order for the next containers? If you run out of a product a few days after placing the order, you’re stuck. The reality in supply chain when you start thinking “is it a good decision?” is much like a game of chess A move is only a good move in relation to all the other moves that are about to be played.
Warren and Joannès debate about the role of simulations and forecasts to navigate this uncertain landscape of decisions. They also talk about the key element of trust within all of this. To conclude, they expand upon how looking at averages can hide potential problems and what point forecasts mean for the future.