What can AI do in transport planning, and what can't it?

"With AI, the system will soon plan everything by itself." I hear it a lot, and it is only half true. AI can do a lot in transport planning, but not everything, and not everything that is called AI is AI. If you know where the line is, you get the most out of it.

Calculating and AI are two different things

Making a good plan is essentially a calculation puzzle: which truck does which trips, within time windows, driving and rest times, loading capacity, and now also the battery and the grid connection. Smart calculation methods (optimisation) have existed for this for decades. They are fast, reliable and give the same answer for the same input.

Language AI, like the chatbots everyone knows, is good at something else: understanding what someone means, explaining, summarising and thinking along. But language AI does not calculate reliably. Ask a language model to work out a plan or a cost price by itself and you sometimes get a convincing answer that is wrong.

The strongest combination is therefore: the calculation method calculates, the AI thinks along, the planner decides.

Where AI is strong

• Answering questions in plain language. "Why doesn't this truck do the afternoon round?" "Why is this driver in the red on driving time?" The AI looks into the plan and the calculation and explains it, without the planner clicking through five screens.
• Thinking along with disruptions. With a driver off sick or a broken charger, the AI puts a few solutions side by side and has the calculation method work them out: what does it cost, how much slack is left, does everything stay within the rules?
• Translating for management. "Why has the cost per stop gone up?" The AI turns figures into a story a managing director understands.
• Finding patterns in data. Which customers structurally run late, which trips are often tight, where consumption deviates.

Where AI falls short

• Calculating by itself. Costs, battery levels and driving times should come from a verifiable calculation, not from a language model.
• Knowledge that isn't in the data. That a customer always opens late on Tuesdays, that a driver has just become a father, that a street is being dug up: the planner knows that.
• Responsibility. If a proposal turns out to be unworkable, "the AI said so" is not an answer. Someone has to decide, and that is the planner.
• Poor data. If the time windows, unloading times or charging capacities in the system are wrong, even the best AI makes a wrong proposal.

What this means for the planner

The work shifts from making trips to orchestrating: assessing proposals, handling exceptions and explaining choices. That takes new skills. A planner must be able to see whether a proposal is right, ask the right questions and know when to step in. See also Ground rules for AI in planning.

Experience it yourself

On the AI day "AI in transport planning" you spend a day working with an AI route planner: you assess proposals, ask questions and plan a real day in teams while things go wrong. The day is day 2 of the Certified E-Fleet Planner programme and is also available separately.