How machine learning can change the shipping industry

Whether it has to do with a spam filter or planning proactive maintenance work, we have come to use machine learning in many parts of our everyday life. Machine learning involves an algorithm that evaluates the data made available to it and develops a logic on this basis. For example, doing so allows it to detect irregularities in data records, which in turn leads to the identification of spam emails. Or it makes forecasts that make it possible, for example, to plan maintenance work accordingly. The special thing about machine learning is that the algorithm for it is not programmed separately, but simply “fed” data.

Generally speaking, a distinction is made between supervised and unsupervised machine learning. Supervised algorithms require a data scientist to “instruct” the algorithm by determining which variables or characteristics the model should analyze and use to come up with its predictions. In this way, one receives a result that is precisely tailored to one’s particular needs. In unsupervised machine learning, the algorithm uses an intuitive approach – called “deep learning” – to review data and draw inferences.

Hapag-Lloyd uses machine learning technology in the areas of maritime network planning, container demand forecasting, and the un-/pairing of container flows. But the technology could also be useful in many other areas – and thereby significantly change the shipping industry. Below are five examples:

1. Trouble-free liner services
Machine learning could help better predict when maintenance work needs to be done and thereby significantly improve the planning of such work. This would make it possible to design even smoother liner services because it would be much more predictable in the long term when which ship will need to be temporarily withdrawn from service.

2. More reliable sailing schedules
Machine learning could be used to calculate the probability of a delay in certain scenarios and thereby supply even more precise arrival data. These probability calculations would make it easier for customers to plan their supply chain. It should be noted, however, that the causes of schedule delays – such as draft problems in the Port of Hamburg due to low water levels in the spring or storms in South America in the fall – aren’t always the only factors that impact schedule reliability. In addition, there are often waiting times at ports that cannot be foreseen.

3. More consistent rates
Machine learning would make it possible to optimize the handling of deficits. Better and more reliable capacity utilization would enable us to offer more consistent rates, which would directly benefit our customers.

4. Better transhipment rates
Machine learning could help improve street turns and container transhipment times with forward-looking recommendations that compare container imports with export bookings. This would make the imbalance more stable, which would have a positive impact on rates. This is because repositioning and evacuation costs are currently a major cost factor in our calculation of contribution margin. For example, it can happen that a customer always returns his empty containers to Location A but exports from Location B. In such cases, we are forced to sail around with the empty container, which results in additional costs. However, if the customer is willing to pick the container up from Location B, we can offer him a discount.

5. Plannable maintenance work
If you keep an eye on the type of uses that typically cause damage to containers, it can result in a pattern. Then, if the pattern arises, there is a high probability that a container will need to be repaired. Machine learning would therefore help optimize maintenance work on containers by supplying predictions – as we would know that defects will (most likely) occur before we can actually see them.