Modelling & integrating probabilistic demand and supply inputs to optimize inventory – EyeOn

 Overview

Company Description

We are EyeOn, innovative supply chain experts. We integrate consultancy, solutions and data science to help improve forecasting and planning processes for our customers. As a graduate intern, you will be part of our data science team. Within EyeOn data science, we help customers improve their planning decisions anywhere on the spectrum from quick-wins to industry-leading advanced data science projects.

Project Description

Project description:

The demand forecast is an important input of any supply chain. The traditional demand forecast (mostly used in industry) is a deterministic forecast (or point forecast); it gives a single demand prediction for each point (e.g., each week per SKU per customer). In other words, you are predicting the most likely outcome. A different approach is
probabilistic where one assigns a probability to every possible outcome. In other words, all future events remain possible, they are just not equally probable. A similar reasoning holds for determining (forecasting) supply lead times.
One of the primary use cases for forecasting demand and/or lead time within the supply chain domain is inventory optimization: When should I order more inventory? How much do I need to order? What should my safety stock level be? These questions can be answered with taking a deterministic forecast as input. One of the limitations of this approach, however, is that one should make assumptions regarding the demand and lead time distribution. With probabilistic forecasts we don’t have to make these assumptions because the forecast engine would output the entire probability distribution (considering demand and lead time).
One example could be the application to estimating a reorder point. Instead of adding the safety stock to the expected demand, we could predict the amount of inventory that we need to meet a certain service level. E.g., for a 95% (cycle) service level target we look at the 95th percentile of our forecasted distribution.
This project aims to investigate how probabilistic forecasts for demand and lead times can be used for inventory optimization, and how these should be integrated. It provides you with the challenge of making advanced forecasting theories work in real-life supply chains.

Goals of the project:

The goal of this project is to look at both the forecasting as well as the inventory side of the equation. Below are some example questions you can think of.
Forecasting (demand and supply inputs)
– How can we generate a robust and high-quality probabilistic forecast?
– Which forecasting models/techniques can we use for this and what are       the trade-offs?
– How should we evaluate forecast accuracy of a probabilistic forecast        (e.g., Pinball loss, Bayesian likelihood)?
Inventory optimization
– How can we effectively use a probabilistic demand and supply forecast      for inventory optimization?
– How are both forecasts for demand and supply best integrated?
– What benefits (if any) does this bring over traditional methods?
– How to implement the approach in practice considering that current            APS  systems mainly rely on a deterministic MRP?

Deliverables:

Exact deliverables to be determined, but include:
• Model(s) implemented in our data science platform (Python code)
• Master thesis
• In-house knowledge sharing session
• Support EyeOn-colleague hosting a webinar on the topic or writing a commercial blog post

Essential student knowledge:

You already have (solid) programming skills and (basic) knowledge of time series forecasting and inventory optimization

More information: escf@tue.nl