Ground rules for AI in planning: what does the system decide, what does the planner?
By Maarten Alta · 18 October 2026
AI calculates faster than any planner. The system consolidates trips, chooses routes and suggests which truck charges where. But who is responsible when a proposal turns out not to be feasible? Without clear ground rules, the system decides more than you think, or the planner ignores it altogether.
Why ground rules are needed
In The transport planner of 2026 I wrote that AI calculates and advises, but the planner remains responsible. That only works if it is agreed in advance where the system may decide on its own and where not. Otherwise one of two things happens:
- Blind trust: the planner accepts everything, including proposals that clash with what he knows.
- Distrust: the planner overrides everything, and the investment delivers nothing.
Three levels
A practical way to make agreements is to put each decision in one of three levels.
| Level | What it means | Examples |
|---|---|---|
| The system decides | The proposal is carried out automatically; the planner reviews afterwards | Order of stops within a trip, route choice at equal cost, charging time at the depot within agreed limits |
| The system proposes, the planner decides | Nothing happens without the planner's approval | Which truck drives which trip, consolidating orders, a charging stop en route, a diversion around a closed zone |
| The planner decides, the system calculates | The system provides scenarios; people make the choice | Deviating from driving and rest times in emergencies, refusing an order, using a subcontractor, anything that affects a driver personally |
Which decision belongs in which level differs per company. Start cautiously and move up as the system proves itself.
Five ground rules
- Legal limits are hard. Driving and rest times, time windows and zones are constraints, not trade-offs. A proposal that breaks them does not go ahead. With electric trucks, this includes whether charging counts as rest; see Charging time as rest time.
- The planner can always step in. And he does not need permission to reject a proposal.
- Overrides are recorded. Not as control, but to learn. If planners keep overriding a type of proposal, something is wrong in the system or the data.
- Explanations come with it. The planner must be able to explain to a driver or customer why a trip is the way it is. A proposal nobody can explain is a risk.
- Data ownership is assigned. Someone is responsible for addresses, time windows, vehicle data and charging points. An AI proposal is only as good as the data underneath.
What the EU AI Act requires
The European AI Act also affects planning:
- AI literacy: since 2 February 2025, organisations that use AI must make sure their people understand it well enough. In 2026 this obligation was worded more flexibly, but taking measures remains necessary.
- Systems that assess people or allocate tasks based on behaviour: AI that assigns work based on personal traits or behaviour, or that evaluates the performance and behaviour of workers, falls into the strictest category. Think of a system that scores drivers on driving style and links trips to that score. After a delay, the extra requirements apply from 2 December 2027.
A system that optimises trips on kilometres, time and charging power generally does not fall under this. But as soon as the driver as a person enters the equation, it is worth checking.
What you can do now
- List the decisions your planning system already makes or proposes, and put them in the three levels.
- Agree on a review moment, for example every quarter: which proposals were overridden, and why?
- Involve the drivers. Explain which choices the system makes and where the planner decides. That prevents distrust.
- Train your planners to assess AI proposals. There is a separate course on AI and data in planning. See the courses or get in touch.
This article is not legal advice.