In today’s world, many operational decisions (e.g., regarding capacity sizing and utilization, order quantities or job scheduling) have to be taken under conditions in which relevant planning data such as demand, order sizes, machine outages or maintenance needs are uncertain. Moreover, predicting the realizations of the relevant planning parameters has become extremely difficult due to complex interrelationships of underlying drivers. In the meantime, with the rise of affordable sensors, smart production equipment, cheap storage for very detailed transactional data and tracking systems for data regarding customer behavior, e.g., from E-Commerce applications, organizations have started to collect and consolidate data sources that are related to their processes and their customers’ behavior. In data-driven operations management, the overarching goal is to use this (newly) available data to make better decisions regarding capacity, inventory levels and pricing, e.g., by identifying patterns, understanding relationships between different uncertain parameters and their drivers, and, more importantly: by incorporating this knowledge into existing and newly developed models.
Our research deals with the development of new approaches to integrate data into decision support models for operations and supply chain management. Data-driven operations management, as we understand it, does not only aim for a better prediction of uncertain planning parameters (which in itself is valuable), but also forces us to re-think and re-design decision models. We use advanced machine learning techniques not only to predict demand (which is the classical approach) but to directly prescribe decisions. The following figure illustrates the main differences between the classical sequential estimation and optimization approach and the joint approach we propose. In addition, we analyze the structural differences between both approaches and how each approach deals with uncertainty in the data.
Recently, our team has worked on various projects in data-driven operations management. We have attracted a number of companies as collaborators and – most importantly for this particular domain – obtained a number of attractive and very interesting datasets (ranging from service companies, such as restaurants or logistics service providers to manufacturing companies and retailers). Some of our collaborators are Lufthansa Technik Logistics Services, DAW (Caparol/Alpina), YAZ (a casual fast food chain), Gesellschaft für Internationale Zusammenarbeit (GIZ) and Maisha Meds and Main-Post Logistikgruppe, a local postal and logistics service provider. All of them have in common that they want to exploit “Big Data” to improve their operations. The following table provides an overview on the industrial settings and the according datasets we are currently working with.
|Lufthansa Technik Logistik Services: Logistics for aircraft maintenance||Forecasting & Capacity Management for inbound deliveries||Announced and actual inbound delivery dates over 1 year; 120000 individual shipments|
|Yaz: fast-casual restaurant chain||Inventory Management||Sales data and stocking quantities over 2 years for 7 different products|
|DAW: (Manufacturer of paint and rendering)||Prioritization of sales activities||40000 observations in raw data; each describes one construction site with 10 relevant features (won/lost; volume; companies involved; #visits;…)|
|Inventory Management||1,626,118 individual shipments; Weather data from nearest weather station; ~150 features derived from time series and weather data|
|Main-Post Logistikgruppe: Local postal and logistics services provider||Forecasting & Capacity Management||Individual shipments over 3 years; |
~150000 per day; for each shipment information about destination and type of shipment
|Maisha Meds: Start-up that improves medical supply chains in Africa||Forecasting & Inventory Management||Sales data from 5 pharmacies over 2 years (200000 observations); |
Information about according inventory levels
Our future research projects aim at extending our current work into two directions. On the one hand, we want to integrate additional, interesting challenges within our setting: As an example, we currently work on incorporating the substitution behavior of customers in stock-out situations and also derive models to estimate these effects in a data-driven manner. On the other hand, we intend to transfer our data-driven approaches to different industries and operations management problems such as data-driven capacity management and pricing.
2018 [ to top ]
Data-Driven Capacity Management with Machine Learning: A Novel Approach and a Case-Study for a Public Service Office. (2018). 105--115.
Big data on the shop-floor: sensor-based decision-support for manual processes. in Journal of Business Economics (2018). 88(5) 593-616.
Privacy-preserving condition-based forecasting using machine learning. in Journal of Business Economics (2018). 88(5) 563-592.
2017 [ to top ]
Prescriptive Analytics for fast-casual Restaurants: Integrating Machine Learning with Inventory Management at Yaz. (2017).
2016 [ to top ]
Capacity planning for a maintenance service provider with advanced information. in European Journal of Operational Research (2016). 251(2)