The manufacturing value chain changed drastically in the last 40 years. As globalization extended, companies started looking for suppliers in other parts of the world to reduce costs. As a consequence, manufacturers now work with contract logistics and shipping companies to forward the parts from their suppliers’ factories to theirs.
Hence, contract logistics and shipping companies have become major players in the manufacturing world. They ship parts worldwide and store them for their customers. They also provide them with additional services to anticipate potential delivery delays and adapt their production plan. This is where Small Data and Predictive Analytics come into stage.
Gathering data on delivery delays
We worked with a contract logistics company to help them identify the key causes of delay and predict their duration. They provide one of their customers with shipping and storage services for parts coming from different parts of the world. As they collect data on these shipments, they thought of building prediction services to help anticipate potential shortages.
They only had data for the past year for each order. This data was correlated with information on the expected shipping date, quantities ordered, time of duration and type, routes taken as well as additional information on the previous orders. The Data Set was about 25000 rows and 30 columns. It was more than the usual Small Data set, but still too little for classic Machine Learning algorithms used in Big Data projects.
They wanted to predict two outcomes:
- if there would be some delivery delays or not;
- Then, what type of delay can they expect: short, medium or long term. Both were classification problems.
The challenge here was first to predict potential delays and their duration. Additionally, identifying the key causes of delay interests the shipping company so that they can implement actions to reduce them. Indeed, reasons for delays are rarely as obvious as the Ever Given Suez crisis of 2021!
When Explainable AI impacts the whole organization
The results were quick to obtain thanks to Explainable Artificial Intelligence. We were able to build quick, explainable predictive models to answer their business questions.
The first model was right 75% when it came to predicting if there would be a delay or not in normal shipping conditions. In that specific case, the key reasons were related to the suppliers’ history and the quantity of goods they ship for each order. These reasons make sense, but our Explainable AI was able to quantify and correlate them to provide the contract logistics company with a simulation tool to predict if a delay would happen.
For the second use case, we predicted the delay duration with a quite similar precision of 74%. This time period was classified between short (0 to 7 days), medium (8 to 15 days) or long (more than 15 days) terms. Moreover, it isolated several interesting features that were mostly related to the type of routes used to transport the goods. Indeed, ground routes were far likelier to cause delays than other types of routes. The type of supplier also played a part to explain delivery delays.
Here, Explainable AI mostly confirmed existing insights. Furthermore, it did put them into perspective and shape to help shippers get better information about delivery delays. Thanks to this information, their manufacturing customers are now able to better predict their needs and improve planification forecasts. This has a huge impact not only on the production side of things but also on the working capital requirement – allowing the manufacturers to optimize their financial organization.
From Delivery Delays prediction to Operational Efficiency
This use case is a good illustration of how Small Data Analytics can impact a whole organization by improving forecasting. And it didn’t need to build a Big Data infrastructure and spend millions on training. So if you are looking to improve the operational efficiency of your whole operations, feel free to reach – we at MyDataModels love to help companies improve their processes and performances!
