Improvement in predicting actualized numbers
7 out of 10 times the engine is closer to predicting the actualized numbers than the old manual estimation.
More business costumers per departure
Molslinjen’s most valuable segment, business customers, has seen an uplift of 14% more business customers per departure because of improved predictions.
Expected reduction in time spent
Molslinjen expects up to 60% reduction in time spent on planning staffing requirements.
Extra tickets sold
Overall uplift in annual sales by more than 4.600 extra tickets on otherwise sold out departures because of the ability to predict how many ticket holders will not show up for departure.
“Now that we have taken the first real step towards fully integrating artificial intelligence into our ferry operations, we can see from the results that there is huge business potential to extend the Forecast Engine made by Halfspace to the rest of our organization.”
Jesper Skovgaard, Commercial Director at Molslinjen
One of the most essential key success factors in optimizing ferry operations is the ability to forecast how many vehicles and passengers that arrive for departure. Not everyone that has made a reservation will be there for arrival. If operators like Molslinjen can more precisely predict how many vehicles and passengers show up, then they’ll also be able utilize the capacity more effectively and optimize pricing and ticket sales.
However, until recently forecasts were quite static, manually processed and made from historical statistics and experience. Molslinjen wanted to become more data-driven and forward-looking.
The objectives were clear:
- Improve load times
- Minimize delays
- Meet staffing and catering requirements with greater accuracy
- Guarantee reservation for business customers
- Sell unused reservations
- Prepare marketing to be more targeted
Molslinjen hired Halfspace to assist the company move from manual estimation to a fully automated AI-based solution.
Halfspace has developed a Forecast Engine which essentially works in these steps:
- Data is sourced into a data lake. The data is typically in categories such as time tables, check-ins, reservations, ticket sales, seasonality, holidays, no-shows and more.
- These data are being automatically processed in the Forecast Engine by three separate engines. One that makes the first set of predictions based on previous patterns, another that adjusts these by also looking at up-to-date online and offline bookings and a third that combines the first two to make real time predictions by considering actual check-in times and any irregularities in traffic conditions to the ferry site.
- The final step is operational decision-making. Making truly data-driven decisions and maximizing capacity utilization based on more accurate forecasts – all in real time. Forecasts that leads to improved ticket sales, more accurate staffing requirements, faster loading times etc.
The Halfspace Forecast Enginee is built from Gradient Boosted Decision Trees in a cutting edge, Databricks Apache Spark environment.
The Key Takeaway
Organizations that manage to assess, identify, analyze and prioritze their data - followed by actual development of predictive models and algorithms based on their newly structured data foundation can achieve better costumer satisfaction, optimize sales and align better with their CSR strategy. Making vast amounts of data and complex modeling simple by capturing and visualizing data and outcome from advanced modeling can break down complexity and optimize operational decision making.