chuttersnap-255214-unsplash

Should you trust AI to support your planning decisions?

  • By
  • Anders Olivarius
  • Artificial Intelligence
  • Containers
  • Maritime
  • Optimization
  • Shipping

As new sophisticated planning tools are finding their way to market, many in the maritime industry are asking how they can trust a complex algorithm to help make their hardest, and most important, decisions. In our work to improve core planning processes in liner shipping with artificial intelligence, we have developed a method to help you limit the uncertainty about new technologies and take control of the process. This paper shows you how.

Photo by kinsey on Unsplash

Every day experienced carrier and terminal planners make tough decisions to move around ships, containers, and cranes when plans are adjusted. Such decisions often have thousands of possible outcomes, and picking the optimal solution is virtually impossible. Planners therefore spend a great deal of time manually calculating and comparing different scenarios. As one vessel operator explained, “I feel like a calculator, and there is never enough time to evaluate all options.”

Fortunately, recent advances in processing power and algorithms have made it possible to identify the best solutions among all the possible options. With its track record of tackling some of the hardest analytical challenges known to man, artificial intelligence — computers mimicking and vastly outperforming human problem-solving capabilities — is particularly well suited for this task. For instance, our AI-powered Berth Optimization Engine can assign key terminal resources to increase the utilization and handling capacity of a typical sea port by 5–10% without any new investment.

The industry has been remarkably accommodating to digital innovation in recent years. Just think about the rise of terminal automation, digital incubators like Ze Box (CMA CGM) and unboXed (PSA), or the Maersk/IBM blockchain joint venture. Yet, it is not unusual for new AI planning software to be met with some initial suspicion.

What many find worrying about technologies like AI is the apparent lack of user control. Planning processes may be manual today, but they are guided by the experience and human ingenuity of the planner, who can step in at any time to resolve unexpected problems. On the face of it, this provides a lot of control and transparency for the user, while the sophistication of artificial intelligence is such that the human mind cannot aspire to trace all steps in the computation process.

Photo by chuttersnap on Unsplash

This would seem to imply a trade-off between control and intelligence in industrial planning tools, and in fact the limited usability of first-generation AI models has held back their application in other industries such as health care and finance. However, when it’s done right, artificial intelligence can significantly increase the power of an organization to control and deploy its resources effectively. The key question is whether the algorithms are used only to speed up an existing process — think of a hyper-powerful calculator — or the tool provides such options to configure the input parameters and choose between different scenarios that the user remains in charge of the process and gains entirely new decision-making abilities.

Coming back to the main question, there are two criteria you can use in deciding whether to trust a particular AI planning tool:

  • Performance: Does it allow you to execute your planning tasks with better results and faster than before?
  • User-friendliness: Is it easy to use with configurable inputs, clear outputs, and ability to tweak the plan in any way you see fit?

If both criteria are satisfied, the new instrument will not only improve the key performance metrics of the organization, but also empower the planning team to help shape business priorities they did not have much influence on before. To use our Berth Optimization Engine as an example, the performance of the algorithm depends on its ability to allocate terminal assets, while its user-friendliness is determined by the planner’s ability to manipulate, monitor, and interpret key elements in the process. For example, planners can select specific optimization priorities that have not been available before and as such improve the competitive position and customer service of the port.

It may not be obvious whether a given AI-powered planning tool meets these criteria, but we have developed a simple 4-step approach to help you evaluate the promises of new, intelligent software products. Here’s how you can do it:

1. Define your metrics:

First you should decide exactly what you want to improve in the organization. It could be key operational metrics such as labor utilization or vessel idle time, or it could something more immediate like the time it takes a planner to manually revise the full plan. Specifying your objectives at a granular level initially will help you make an informed decision later.

2. Set a baseline:

To understand if the new software results in an improvement in your metrics, it is helpful to first define their pre-trial level. In today’s data-driven vessel and terminal operations, carriers and terminal operators can quickly tease out most KPIs of interest, such as fleet utilization in a specific trade lane or service. You may also ask the software vendor to help you establish the baseline. For example, we offer our terminal customers a diagnostic free of charge to determine the current level of their operational KPIs and estimate the probable impact of our collaboration.

3. Take control: 

Once you start testing the software application, you can immediately begin to explore its user-friendliness. Like an airplane needs a pilot, a terminal or service lane requires an operator to make the last call when important operational goals need to be ranked. Verify that the tool lets you select and rank multiple objectives for each optimization cycle and that you can assess the feasibility of each optimized solution before you implement it in the plan. It is not enough that a plan optimizes your KPIs, it also needs to work under real-life operational constraints. Relevant quality metrics would include the number of changes committed and any violations detected.

4. Evaluate: 

Finally, sit back and review the results of the trial. To evaluate whether you should trust your new AI decision support tool, go back to your baseline and see if the trial resulted in an operationally meaningful improvement in your metrics. And confirm that the tool gives you the ability to select its objectives, assess its recommendations, and even control elements of the operational environment that were not accessible before.

Industrial planning tools that use AI to augment live decision-making may be hard to trust at first, but if you break down their promises and evaluate their performance and user-friendliness, they can result in big operational improvements. To learn more about how Portchain uses artificial intelligence to help container carriers and marine terminals take planning operations to the next level, please write me a note at anders@portchain.com


Previously published on Medium