Franz Edelman Award 2024: Molslinjen and Halfspace AI
It is always a good idea to learn about best practices in other domains. Though, one has to search for them. There are only a few awards that recognize accomplishments in analytics. Franz Edelman Award by INFORMS is one of them.
This year’s winner 🏆 is the team of Danish ferry operator, Molslinjen, and Halfspace AI. They have improved the overall operations of Molslinjen with forecasting models, revenue management practices and tailor-made systems.
“First awarded in 1972, the Franz Edelman Award recognizes and rewards outstanding contributions of analytics and O.R. in the for- and nonprofit sectors around the globe. Each year, INFORMS honors finalist teams that have improved organizational efficiency, increased profits, brought better products to consumers, helped foster peace negotiation and saved lives. Since its inception, the cumulative dollar benefits from Edelman finalist projects have surpassed $419 billion.”
So, this is a prestigious award that is given based onsignificant contributions to efficiency of organizations. Before diving into Molslinjen-Halfspace collaboration, let’s review the other finalists of 2024. Taken directly from the announcement page:
- ALDI SÜD Germany for “Data-Driven Approaches and Collaborative Intelligence for Empowering Leadership of a European Market-Leading Retailer”
- American Airlines for “HEAT — The Hub Efficiency Analytics Tool”
- McDonald’s China for “Supply Chain Network Optimization Has Reduced Millions in Transportation Costs and Fostered Economic Growth for McDonald’s China”
- Tata Steel Limited for “Strip Temperature Control in Continuous Annealing Furnace of Tata Steel India by Model Predictive Control Approach”
- Transvision for “Optimizing Mobility for Elderly and Disabled Dutch Citizens using Taxis”
The diversity of problems and solutions show that this is a really rich field. Now, let’s go back to the main topic.
Data-driven at Sea: Forecasting and Revenue Management at Molslinjen
To our benefit, problem owners at Molslinjen and Halfspace penned a beautiful article about their whole improvement process starting from 2019. You can get the preprint from Pierre Pinson’s webpage.
In summary, Halfspace craftily devised analytical tools to improve Molslinjen operations where analytical solutions are needed the most. It is not one big, state-of-the-art “AI” solution but a number of battle tested models adapted to the problems at hand. The result is a happy company, happy customers, decreased costs and increased profit margins.
Here are my brief and incomplete notes on the case study.
- Molslinjen is a ferry operator in Denmark. They carry people and vehicles (cars, motorcycles, buses etc.).
- It is emphasized that top management (CEO & CCO) is previously from airline industry (SAS). It is indicated that, top management is eager to implement practices known to the airline industry (esp. revenue management) in ferry business as well.
- Molslinjen’s main improvement points are improvement of “vehicle formation in the hull”, “demand forecasting” and “revenue management with focus on dynamic pricing”.
- Vehicle formation problem is mind-stimulating. In the example there is a ferry with a hull with three main lanes which vehicles shall park during the journey. There are two formations: 2-by-2 formation is a relaxed formation which cars form two lanes at each main lane. Zipper formation requires 3 lanes of cars at each main lane. It is very hard to change formation mid-loading. They also have to account for balance (heavy vehicles shall be in the middle) and other criteria. Halfspace provides an algorithm which recommends formation and packing tightness. Their solution also provides visualization.
- Demand forecasting tool became the backbone of all their operations feeding both vehicle formation and dynamic pricing modules. Their solution is to use a version of gradient boosted trees model (XGBoost) combined with prominent feature engineering. It is not a fancy model like Transformers or other deep learning stuff, though it gets the job done. They use a custom metric to train, and they expectedly update the forecasts periodically until recently before the departure time.
- Revenue management with dynamic pricing is based on several pillars. Molslinjen has three price categories; Low Fare, Standard and Business. They have to optimize ticket availability for capacity, demand and prices for these three classes and vehicle capacity.
- They used a combination of parametric modelling with Bayesian neural network model and a Bayesian inference framework. They make a distinction of price related and non-price related features and provide estimates for each class fare.
- These estimates are inputs for the dynamic pricing module which uses a well known algorithm from the literature, EMSRb, as base model. They use an extension called EMSRb-MR (Marginal Revenue).
- Rest of the paper, they discuss the improvements and how they are calculated through quanitative and qualitative analysis, what were the significant impacts of the implemented solutions and testimonials from Molslinjen management.
Conclusion
First of all, congratulations to Molslinjen and Halfspace, and their people who are involved in the making!
The article is well-written. It is definitely not a hardcore technical paper as it should not be, though it provides technical details where it is necessary. Problem and sub-problems are well defined. It becomes easier to solve the sub-problems instead of attacking the problem as a whole.
Top management at Molslinjen is confident that their ferry business operations can be improved, thanks to their experience in the aviation industry. This is a valuable point because such innovations are usually hard to implement without management support.
Methodology used to define and solve the operational challenges is described in detailed along with discussions on decisions to implement. In most cases, what Halfspace did was to use commoditized models and modify them to their needs. The whole process is a beautiful example of an end-to-end solution.
I’m pretty sure other finalists have also great stories. Though, Molslinjen and Halfspace make a very good case about our profession and line of work. I highly recommend those who are interested to read the 33-page long article.
