The Quantitative Supply Chain

Any nontrivial demand forecasting model becomes a black box for supply chain practitioners, that is, an opaque subsystem that produces numbers that are difficult to understand and to challenge. Whiteboxing, as part of the Supply Chain Management practice, is the answer to this problem. Practitioners don't need to understand the 'how' but need to understand the 'why'.

Pricing optimization is typically not considered as part of the Supply Chain Management (SCM) practice. Yet, pricing is a factor that strongly influences customer demand. Thus both production capacities and stock levels are highly dependent on prices, and must be jointly optimized.

A tight control over lead times is important in order to optimize a supply chain. Yet, while electronic inventory management started to provide transparency on stock levels back in the 70's, tracking mobile assets only came much later. The term 'supply chain transparency' refers to the real-time positioning of all of a supply chain's assets, including the mobile ones.

Data lakes are data storage technologies intended for bulk reads and bulk writes. They are particularly well suited to address supply chain challenges, because many situations require an inspection of the company's entire history of orders and stock movements.

Supply chains are complex systems made of many moving parts: goods, people, machines. POCs (Proofs of Concept) routinely fail when attempting Quantitative Supply Chain initiatives because problems get displaced instead of getting solved.

The relevant amount of historical data when considering large supply chains frequently exceeds one terabyte. As a result, inventory control requires two distinct flavors of software: transactional software (e.g. an ERP) to manage the resources, and predictive software (e.g. Lokad) to optimize the resources.

Safety stocks are an inventory optimization method that enforces an extra quantity of stock beyond the expected demand in order to maintain a target service level. This method relies on key statistical assumptions about the demand forecast, most notably that the error is normally found in the distribution.

The inventory optimization of Bridgestone's multi-echelon supply chains entails many challenges. Probabilistic forecasting is used to assess the consequences of every single stock movement and every single production capacity adjustment within the supply network.

Machine learning is an umbrella term that includes diverse algorithmic approaches. In supply chain, the historical way of doing machine learning was time-series forecasting. However, this approach has been superseded by a series of superior forecasting approaches.

Fashion brands are driven by novelty. This puts a lot of pressure on fashion supply chains to reduce overstocks, both to reduce the discount rates but also to make room for the new collections. Overstocks are typically caused by an incorrect prior forecast of the customer demand.