Route optimisation in transport: what it is, what AI adds and what the Dutch truck toll changes

What is route optimisation?

Route optimisation is the search for the best allocation of shipments to vehicles and the best order for the stops. 'Best' usually means the fewest possible kilometres or hours, or the lowest possible cost. Hard constraints apply: customers' time windows, load capacity, vehicle requirements and drivers' driving and rest times. In the field, this is known as the vehicle routing problem.

The problem is old. In 1959, G.B. Dantzig and J.H. Ramser described in the journal Management Science how a fleet of fuel tankers can supply a large number of service stations from a single depot in as few kilometres as possible (The Truck Dispatching Problem).

The difficulty lies in the numbers. Even a single vehicle with twenty stops can visit those stops in more than 2.4 quintillion sequences. With several vehicles, time windows and capacities, that number quickly grows further still. Planning software therefore does not work through every possibility, but uses clever approximations that find a good, feasible plan in a short time. Dantzig and Ramser did the same back in 1959: they looked for a near-optimal solution, not a proven best one.

Route planner, route optimisation and AI: what is the difference?

A route planner finds the way from A to B, like a navigation system. Route optimisation goes a step further: it allocates a whole day's shipments across your fleet and determines the sequence for each vehicle, within all the constraints. That is mathematics from operations research, and in itself it is not AI.

Machine learning can play a part in that process, though. It can, for example, predict travel and arrival times better from historical trip data, or reach a good approximation faster than classical calculation rules do. That is a real, gradual improvement to an existing field. It is not a new technology that reinvents planning. See also the term route optimisation (VRP) in our glossary (in Dutch).

Where AI really adds value

The practical gain does not usually lie in a spectacularly different route. It lies in handling changes faster and more consistently. A planner who is faced with ten extra trips and three absent drivers in the morning benefits from software that quickly works through several feasible scenarios. The human can then take a well-founded decision sooner.

• Faster replanning in the event of disruptions such as congestion, sickness absence or cancelled trips.
• Better estimates of travel times from historical trip data, rather than fixed average speeds.
• Several scenarios side by side, so that the planner can compare rather than blindly following one outcome.
• Early flagging of planning conflicts, for example with the available driving time, well before the trip begins.

Where the hype begins

Some vendors suggest that AI produces the optimal plan on its own and that the planner becomes redundant. For most transport companies, that is not a realistic picture. Route optimisation works with assumptions and averages. Reality is full of exceptions: a customer that can only unload in the morning, a driver who knows a region better, a vehicle that has just come back from maintenance.

Be critical, too, of savings claims in sales pitches. The effect of route optimisation varies greatly by company, by region and by type of trip. A generic percentage says little about your own situation. Ask vendors for a trial with your own trip data rather than for a marketing figure.

How widely is AI already used for route planning?

The available figures vary widely. According to CBS (Statistics Netherlands), 7 percent of companies in transport and storage used one or more AI technologies in 2025. Those are provisional figures, covering companies with two or more persons employed. Of all companies that use AI, 4 percent used it for logistics purposes.

A survey by planning software vendor Descartes, conducted in 2025 among 300 senior decision-makers at shippers and logistics service providers in Germany, the United Kingdom, France and Belgium/the Netherlands, arrives at a much higher figure: 37 percent use AI for route and load optimisation. The two figures measure different things. CBS counts all companies in a sector, whereas Descartes asked decision-makers who are responsible for transport. Moreover, the survey comes from a company that sells such software. Together, they show above all that 'AI in planning' means very different things to different people.

What does a kilometre driven cost now that the truck toll applies?

Since 1 July 2026, owners of trucks in categories N2 and N3, with a technical maximum mass of more than 3,500 kilograms, have been paying an amount per kilometre driven. This applies on almost all motorways and on a number of provincial and municipal roads. The cleaner and lighter the vehicle, the lower the amount (Rijksoverheid, the Dutch central government).

From 1 September up to and including 31 December 2026, a temporary discount of 22.3 percent applies. As a result, the average rate falls from €0.191 to €0.148 per kilometre driven (Rijksoverheid, 22 May 2026). For a truck in CO2 emission class 1 with EURO 6 and a technical maximum mass of more than 32,000 kilograms, the official tables on vrachtwagenheffing.nl give these amounts per kilometre:

• From 1 July to 1 September 2026: €0.201.
• From 1 September to 1 January 2027, with the discount: €0.156.
• From 1 January 2027: €0.206 according to the table for 2027 on the same page. On 11 October 2026, that page also stated that the amounts for 2027 would be announced in October 2026. So check the table before the turn of the year.

What does the toll mean for route optimisation?

The toll changes the calculation behind every plan. A kilometre on a tolled road now incurs the toll on top of fuel, wages and wear and tear, and the amount varies by vehicle. A system that works with one fixed rate per kilometre for the whole fleet does not see that difference. So ask whether the software can take each vehicle's rate into account, and whether it distinguishes between tolled and untolled roads.

A worked example, not a measured value. Suppose better planning saves 15,000 tolled kilometres a year for a truck in CO2 emission class 1, EURO 6, of more than 32,000 kilograms. At the discounted rate, that is €2,340 in toll. At the rate for 2027 from the table, it is €3,090. The wages, fuel and wear and tear for those kilometres come on top of that.

AI does not lower the rate. What better planning can lower is the number of tolled kilometres per shipment. The full rate tables, the changes to the motorrijtuigenbelasting (mrb, the Dutch motor vehicle tax) and what changes on 1 January are in our article on the truck toll and trip planning. You need the same trip data when clients ask for CO2 per shipment.

Joint planning with other companies: the SiLK subsidy

The biggest saving in kilometres sometimes lies not in your own planning but in that of your neighbours. Two hauliers that each drive back half empty, or three shippers that deliver to the same region, can drive fewer kilometres by working together. For such cooperation there is the Tijdelijke subsidieregeling samenwerking in de logistieke keten (SiLK, the temporary subsidy scheme for cooperation in the logistics chain), administered by RVO (Netherlands Enterprise Agency). SiLK is one of the schemes funded from the revenue of the truck toll. The examples RVO gives include consolidating loads, avoiding empty and duplicated trips, route and network optimisation and merging the planning of different parties.

The 2026 round runs from Tuesday 15 September 2026 at 09:00 up to and including Tuesday 13 October 2026 at 12:00, with a budget of €5 million. Those dates are also in the amending regulation in the Staatscourant (Dutch Government Gazette). At the time of this update, on 11 October, the round therefore still had two days to run. If you are reading this after 13 October, the 2026 round has closed. A complete application requires, among other things, a cooperation agreement signed by all participants, a project plan and eHerkenning (the Dutch electronic login for businesses) at level 3. Anyone who has yet to start on those would do better to aim for a possible next round.

Whether there will be a new round in 2027 has not yet been published. The explanatory notes to the amendment of June 2026 do say that the intention is to open SiLK for applications several years in a row. According to the original regulation, SiLK runs until 1 October 2030. The first round ran from 4 November to 30 December 2025 inclusive, with a ceiling of €3.6 million. The main points of the 2026 round, according to RVO:

• A partnership of at least three companies. At least half are active as a transport company, shipper, own-account carrier or logistics service provider. At least one participant is the holder of a truck under the Wet vrachtwagenheffing (Dutch Truck Toll Act).
• A subsidy of up to 75 percent of the project costs, with a minimum of €25,000 and a maximum of €100,000 per project.
• No more than 30 percent of the eligible costs may go to externally hired staff, at no more than €125 per hour excluding VAT. Your own hours count at a fixed rate of €60 per hour.
• Each participant bears at least 10 and at most 70 percent of the total costs.
• The project must save at least 15,000 truck kilometres per year on Dutch roads, with a lasting result. According to RVO, that is a best-efforts obligation. If you carry out the activities but do not achieve the reduction, that has no consequences for the subsidy.
• The costs of preparing the project are ineligible. The project must be completed within twelve months of the award. An application is rejected if two or more participants were previously together in a partnership that received a SiLK subsidy.

Driving and rest times still take precedence

Whichever software proposes a route, the European rules on driving times, breaks and rest periods in Regulation (EC) No 561/2006 apply in full. The daily driving time is at most nine hours, which may be extended to ten hours twice a week. After four and a half hours of driving, a break of at least 45 minutes follows, which may also be taken in two parts of 15 and 30 minutes. A system that calls a route 'optimal' without taking those limits into account does not produce a usable plan. At most, it produces a plan that has to be corrected afterwards, or that leads to infringements.

Good planning software therefore includes these rules in the calculation as a hard constraint. When choosing a system, ask how the driving time rules are built into the calculation model, and have that demonstrated with a specific example from your own planning.

With joint planning, that applies to all parties. Article 10 of the Regulation makes the transport undertaking liable for infringements by its drivers. It also obliges undertakings, freight forwarders, consignors, principal contractors and subcontractors to ensure that contractually agreed time schedules comply with the Regulation. A joint plan must therefore also fit within the driving times of your partners' drivers.

The human decides and signs

At AIBA, our starting point is that the human decides and signs. For transport planning, that means the planner remains responsible for the final route, even if an algorithm proposed it. That is a matter not only of trust but also of liability.

Moreover, Article 4 of the AI Act has applied since 2 February 2025. Since the amendment by Regulation (EU) 2026/1744, in force from 27 July 2026, it has asked providers and deployers of AI systems for measures that support the development of the AI literacy of their staff. They do not have to guarantee a specific level. In so far as your planning system is an AI system, what matters for planners in practice is that they understand what it can and cannot do, where the assumptions lie and when a proposed route must be adjusted manually. AI training for logistics (in Dutch) based on your own planning is aimed at that: a planner who can assess a proposal from the system and reject it with sound reasons.

Where to start

Before you invest in new software, it pays to establish where in your current planning most of the time and most of the unnecessary kilometres go: drawing up the base plan, disruptions during the day, empty return trips or monitoring driving times. For one type of trip, record how many tolled kilometres you drive, how full the vehicles are and what share of the kilometres is empty. That is your baseline measurement.

With that baseline measurement you can answer two questions. Does better planning within your own company deliver enough? And do the empty kilometres in fact lie between you and other companies, so that cooperation would deliver more than new software? The free Practice Scan helps you weigh that up for your situation.