5/5 (3)

Last summer, I found myself stepping into a role that was both exciting and intimidating: a data analyst that needed to built a data pipeline. Armed with only basic Python skills and no prior coding experience, I had to figure out how to build an end-to-end data transformation pipeline, also known as ETL (Extract, Transform, Load). This involved interacting with APIs, manipulating large datasets, and uploading data to a Snowflake database. The learning curve felt steep, but due to ChatGPT’ I was able to finish the project with succes. The tool I used in this case was the “code copilot” GPT from ChatGPT.

I initially struggled with where to begin. Although I had a foundational understanding of data analysis, I wasn’t familiar with the technical intricacies of building a complete ETL pipeline. However, with the help of ChatGPT, I was able to gradually piece together the process. I would input small tasks and concepts I wanted to tackle, and the AI would provide explanations and snippets of code. This iterative back-and-forth helped me demystify many of the tasks.

By the end of the project, I had created a fully functional ETL pipeline that automated the collection and transformation of data. Without the assistance of generative AI, what seemed like a daunting and nearly impossible task turned into a fulfilling learning experience. It empowered me to stretch beyond my initial capabilities. Generative AI truly served as a valuable tool, transforming what could have been a steep learning curve into a collaborative and enjoyable project.

Pitfalls

While generative AI was incredibly helpful in building my data transformation pipeline, there were some notable limitations. Debugging, for instance, still required significant manual effort. ChatGPT struggled with complex bugs, and I often had to turn to StackOverflow for deeper insights and solutions to fix it myself.

Additionally, technical knowledge was crucial. While the AI helped structure code, I needed to understand APIs, Snowflake, and XML parsing to implement specific details. ChatGPT couldn’t generate the entire solution on its own. ChatGPT is great for creating the global parts of code but I needed to adjust the code most of the time to make it fit in my project.

Moreover, I realized the importance of asking precise questions. You can’t ask the AI to write code without providing clear technical requirements. It’s similar to how a product owner communicates with developers: even if you don’t know how to code, you need to speak their language to convey what you want. In the end, generative AI was a valuable tool, but success depended on my ability to guide it with the right queries.

5/5 (2)

Demand forecasting

¹Demand forecasting is nothing new within supply chain management. It involves forecasting future demand using statistical methods. Examples of traditional methods include: (weighted) moving averages, exponential smoothing, regression analysis, bottom-up and top-down forecasting (3SC, 2022). However, these methods are very time-consuming to maintain and rely on ‘manual’ processes. In addition, they are also limited with handling external variables such as weather, social media trends etc. An alternative will be discussed in this blog post.

Machine learning as a solution

Machine learning can provide a solution to replace traditional forecasting methods. Machine learning models can adapt quickly to changes, providing real-time insights that drive smarter decision-making, minimize stockouts, and improve overall supply chain performance and efficiency. The advantage of a machine learning model is that it learns by doing: The longer the model is used, the more accurate the predictions will be. But this also means that the model needs massive amounts of data to train itself. The image below shows some data source examples for a machine learning model. This amount of variables is complex to include in manual forecasting methods.

Is machine learning able to outperform traditional demand forecasting methods
Machine learning (ML) techniques have been explored in supply chain management since the early 2000s. At that time, however, ML methods showed no significant improvement over traditional forecasting methods (Carbonneau et al., 2007). Today, the situation is different. Recent studies demonstrate that ML methods now outperform traditional techniques (Feizabadi, 2020). Additionally, a comprehensive literature review has confirmed that ML frequently delivers better results than conventional methods (Aamer et al., 2021). This is mainly due to larger amounts of (usable) data and improved algorithms.

My view for the future

I personally think machine learning forecasting is the future within supply chains. Even though traditional methods are reasonably accurate. Machine learning methods have the ability to include immense amounts of data and variables that traditional methods could never include. In my opinion, the biggest obstacle to large-scale machine learning application lies in the transparency of the entire supply chain. Today’s supply chains are so complex that insight into an entire supply chain is very difficult. But this transparency is needed to feed machine learning models. After all, a model is only as good as the data it consumes.  When this supply chain transparency is enhanced, e.g. by large-scale block chain application, the world lies open for large-scale machine learning application, allowing global supply chains to operate more efficiently (and thus often more sustainably).

Literature

3SC. (2022, September 12). Overview Of Demand Forecasting Techniques and Impact Of AI/ML On It. https://3scsolution.com/insight/demand-forecasting-techniques-and-planning

Aamer, A., Yani, L. P. E., & Priyatna, I. M. A. (2021). Data Analytics in the Supply Chain Management: Review of Machine learning applications in demand Forecasting. Operations and Supply Chain Management an International Journal, 1–13. https://doi.org/10.31387/oscm0440281

Carbonneau, R., Vahidov, R., & Laframboise, K. (2007). Machine Learning-Based demand Forecasting in supply chains. International Journal of Intelligent Information Technologies, 3(4), 40–57. https://doi.org/10.4018/jiit.2007100103

Feizabadi, J. (2020). Machine learning demand forecasting and supply chain performance. International Journal of Logistics Research and Applications, 25(2), 119–142. https://doi.org/10.1080/13675567.2020.1803246

Wilson, E. W. (2019, August 26). Forecaster’s & Planner’s Guide To Data. IBF. https://demand-planning.com/2019/08/26/forecasting-data-types/