AI & mobility · Transport and logistics

AI in transport.

From route optimisation and dynamic ETAs to predictive maintenance, document AI and computer vision at the loading dock: AI is changing how transport companies, fleet owners and logistics planners run their operations. We help organisations that want to move from isolated pilots to AI that genuinely lives in day-to-day planning.

DomainTransport & logistics
Real-timeYes, edge + cloud
GoalPlanning, costs, CO2
ApproachSprint-based, start small
IntegrationsTMS, telematics, ERP
HostingCloud or hybrid

What does AI in transport actually mean?

The term AI in transport covers a broad field: from classic optimisation algorithms (routing and planning models based on OR-Tools and MILP solvers) to modern machine learning for ETA prediction, computer vision for damage inspection and large language models that read freight documents. In practice, transport companies use a combination, as no single technique covers every problem.

We see the same patterns at couriers and last-mile players, at 3PL and 4PL planners, at fleet owners with trucks, at multimodal shippers combining rail, water and road, and at MaaS and car-sharing platforms. The problem is rarely 'we have no data'. It is usually 'we have ten systems with separate data and nobody joining it up'. That is what we build: integrations between TMS, telematics, ERP and planning systems, with an AI layer on top that supports operational decisions rather than taking them over.

One thing to make clear up front: AI in transport is not a self-driving truck. Autonomous vehicles are a separate field with their own regulation and R&D horizon. What we are talking about is the AI layer on your existing operation. Drivers, planners and dispatchers stay in control, and AI does the work that people cannot do quickly or consistently enough.

Multi-stop
Routes with time windows, working-hours rules and fuel as constraints
Real-time
ETAs based on traffic, weather and history
Edge + cloud
In-cab tablet plus central models
CSRD-ready
CO2 traceable per shipment

When is AI in transport relevant for your organisation?

01
Planning

Routes are still cut and pasted by hand

The planner works with Excel, a map and years of experience. That works, but it is fragile: one sick planner and operations stall. AI-assisted route optimisation gives you a reliable baseline that people can steer from.

02
ETA

Customers complain about unreliable arrival times

Static ETAs from the TMS do not account for current traffic, weather conditions or the history of a specific route. Dynamic ETAs significantly reduce the number of 'where is my shipment' phone calls.

03
Maintenance

Trucks break down unexpectedly

The data from the FMS bus and telematics is already in Webfleet, Trimble or Fleetboard. With predictive maintenance, patterns become visible before a breakdown, and you schedule maintenance at a suitable moment instead of at the roadside.

04
Document flow

CMRs, customs papers and invoices take up too much time

Document AI reads consignment notes, CMRs, customs forms and tax documents with high accuracy. A planner or administrator still checks the output, but no longer types anything in.

05
Last-mile

Urban distribution is becoming unsustainable

Time windows, locker boxes, customer-at-home, low-emission zones, load mix: the last-mile puzzle has too many variables for a human planner. AI routing with live time-window optimisation reduces the number of failed deliveries.

06
Sustainability

CSRD reporting requires scope 3 emissions per shipment

Before long, you will need to be able to state the CO2 footprint of every shipment. Carbon-tracking AI calculates this from route, vehicle and load, and at the same time shows where reductions are achievable.

Three ways we build AI in transport.

Approach 01

On top of your TMS

You already work with Transwide, Descartes, Manhattan, Trimble Transportation or Centric. We build an AI layer on top that combines data from your TMS with telematics and external sources, without forcing your planning team onto a new system.

Approach 02

Custom alongside off-the-shelf packages

Sometimes the optimisation logic you want doesn't fit into an off-the-shelf package such as Loginext, FarEye, Onfleet or Routific. In that case we build a targeted custom component that integrates seamlessly with what you already run.

Approach 03

Fully custom for specific flows

Multimodal flows, complex yard management at a large DC, or a matching platform for empty return trucks: for specialist flows we build a platform from the ground up, using OR-Tools, our own ML models and integrations at CAN bus level.

Practical AI applications in transport and logistics.

An overview of where AI delivers value in practice, based on conversations with planners, fleet owners and logistics directors. Not every application makes sense for every organisation; often one well-chosen quick win is worth more than five simultaneous pilots.

Route optimisation

Multi-stop, multi-constraint: time windows, collective labour agreement rules, fuel consumption and driver preferences in a single model.

Dynamic ETA

Arrival times that adapt to traffic, weather and historical patterns on that specific corridor.

Demand forecasting

Which route, which type of vehicle and which volume you can expect in the coming hour, day or week.

Yard management

Which truck goes to which dock, with insight into timely availability and turnaround time on site.

Predictive maintenance

Engine and sensor data from the FMS bus and telematics predict failures well before they occur.

Computer vision

Damage inspection at pickup, licence plate recognition at the gate, dock cameras for loading and unloading times.

Document AI

Automatically read, validate and forward CMRs, customs papers and invoices to your TMS or ERP.

Last-mile micro-routing

Urban routes with locker box allocation and customer time-window optimisation.

Customer chatbot

Handle track and trace, delivery updates and simple claims without burdening the service desk.

Crew rostering

Plan in line with collective labour agreements, monitor rest times and respect driver preferences.

Carbon tracking

Calculate CO2 emissions per shipment for CSRD reporting and internal reduction targets.

Return matching

Match empty return trucks with demand, in the spirit of Quicargo and sennder.

The stack: from CAN bus to LLM.

Good AI in transport starts with the right data at the right level. On the vehicle itself we work with OBD-II, FMS bus and CAN bus; in the cab with telematics platforms such as Webfleet, Trimble and Fleetboard. At infrastructure level sits a streaming layer (Kafka or Pulsar) that merges engine data, GPS, load events and order flows into a single source for models.

For routing and planning problems, we often reach for classic optimisation solvers: OR-Tools, MILP solvers and, depending on scale, purpose-trained reinforcement learning models. For ETA prediction and demand forecasting, we use gradient boosting and time-series models. For computer vision at the dock, damage photos or licence plates, we work with YOLO variants and modern vision APIs. For document AI, we use LLM extraction, integrated as described on our custom LLM integrations page.

We handle routing and geo questions with the right mix: OpenStreetMap, HERE, Google Maps and TomTom each have their strengths. For dense urban last-mile, HERE or TomTom is often more accurate than Google Maps; for open-source routing with your own restrictions, OSRM based on OpenStreetMap is the right choice. We choose per use case and isolate that choice in a service layer, so switching later doesn't require a redesign.

On the off-the-shelf side: on the TMS side we come across Transwide, Descartes, Manhattan, Trimble Transportation and Centric; on the WMS side Manhattan, Blue Yonder and Centric. AI vendors such as Loginext, FarEye, Onfleet, Routific and Optidrive cover specific sub-problems. We are agnostic: if something works well off the shelf, we use it. Only where off-the-shelf packages hit their limits do we build custom.

Compliance: rules specific to transport AI.

A
GDPR

Driver tracking is particularly sensitive

People's location data is personal data. Under the GDPR you need a legal basis (collective labour agreements, works council consent) and data minimisation. We build privacy by design: raw data is kept briefly, aggregated patterns for longer.

B
AI Act

Employer monitoring falls under high-risk

The EU AI Act classifies AI systems that monitor employees as high-risk. That means extensive documentation, human oversight and transparency. We build these requirements in from sprint one rather than adding a compliance stamp afterwards.

C
Driving and rest times

AETR / EU 561/2006 and the road haulage collective labour agreement (CAO)

A scheduling model that ignores driving and rest times or the CAO is immediately unusable. We model these rules as hard constraints, not 'preferences'.

D
CMR & e-CMR

The digital consignment note is a reality

The digital consignment note (e-CMR, in line with DCD) is gaining ground internationally. The document AI flows you build now will need to connect to it seamlessly later, and we design with that in mind.

E
CSRD

Scope 3 emissions per shipment

For most clients in transport, CSRD reporting is coming faster than expected. A carbon-tracking layer in your AI stack will soon be an obligation rather than a luxury.

F
NIS2

Critical infrastructure

Transport is one of the sectors covered by NIS2. Securing the AI layer (access management, logging, audit trails) becomes part of your broader security policy.

How do you get started?

In our experience, a good start in transport AI is not a big-bang roadmap but one carefully chosen use case where the impact is measurable. That is usually either ETA prediction on a single corridor, a document AI flow for CMRs, or predictive maintenance on one vehicle type. Something that is in production within a few sprints and that your planning department can use straight away.

We build on that first success. Only once the first flow runs stably and has earned the trust of planners and drivers do we expand to the next domain. That order isn't organisational caution; it's practical: AI systems that aren't useful to the people on the shop floor from day one are rarely adopted, however good the model is.

We often work alongside your existing TMS vendor, telematics partner and, where relevant, the fleet management department. The pages on building transport software and developing a fleet management app say more about how our wider transport practice works. If you would rather explore the visual side of AI (damage inspection, licence plates, dock cameras), see our computer vision applications.

For organisations that want to think beyond a single use case — multiple domains, a genuine AI roadmap, governance, staff training — we work from our wider practice in enterprise AI implementation. The bigger the ambition, the more important a good sequence becomes. We help with that order, not just with the code.

Our AI transport practice serves a broad audience: transport companies with their own fleet, couriers and last-mile players, 3PL and 4PL planners, fleet owners, multimodal shippers (rail, water, road), distribution centres that connect logistics and warehousing, and MaaS or car-sharing platforms that want to work data-driven. Public transport organisations (NS, GVB, RET) and autonomous vehicle start-ups are more the market context we operate in than our direct client base, although we regularly share insights with them from other projects.

Frequently asked questions.

What exactly does AI do in transport?
AI in transport is an umbrella term for route optimisation, ETA prediction, demand forecasting, predictive maintenance, computer vision (damage inspection, licence plates), document AI on freight documents, carbon tracking and customer communication. It is rarely about autonomous vehicles; it is about decision support for planners, dispatchers and fleet owners, based on data from your TMS, telematics and external sources.
Will AI replace our planners?
No. A good AI layer in transport extends the planner's work: routes are suggested rather than assembled by hand, exceptions surface faster and the planner stays in control. In practice, we see planners getting more time for the truly complex cases (customer contact, emergencies, multimodal puzzles) instead of repeatedly keying in routes. Human knowledge also remains crucial: a planner knows customers, drivers and roads in a way that no model will match any time soon. AI is a colleague, not a replacement.
Which quick wins are there in AI for transport?
We usually see the fastest results with document AI (automatically reading CMRs and invoices), dynamic ETA on a busy corridor, and predictive maintenance on a single vehicle type where telematics data is already available. One well-chosen use case that is in production within a few sprints and has a visible effect often delivers more than an ambitious AI strategy programme.
How do you get started with AI in a transport company?
Start small, with one measurable use case and one clear owner within your organisation. First take stock of what data is already available in your TMS, telematics and ERP, as that determines where the first quick win is realistic. Avoid large transformation programmes with several parallel pilots; they fragment attention and rarely lead to adoption.
What about privacy and the AI Act when tracking drivers?
Driver location data is personal data under the GDPR and, when monitored using AI, falls under the high-risk category of the EU AI Act. In practice, that means a clear legal basis (collective labour agreement and/or works council consent), data minimisation, transparency towards drivers and human oversight. We build these requirements into the design from the first sprint, not as a compliance layer added afterwards.
What determines the cost of an AI project in transport?
Mainly: how well your data is already integrated (TMS + telematics + ERP), how many use cases you want to tackle in parallel, whether you use off-the-shelf packages such as Loginext or FarEye or want custom, and how strict the compliance requirements are around driver tracking and CSRD. We work with a fixed sprint budget and give you a concrete project estimate after the planning phase.
Can you integrate with our existing TMS and telematics?
Yes. We have experience with TMS packages such as Transwide, Descartes, Manhattan, Trimble Transportation and Centric, and with telematics platforms such as Webfleet, Trimble and Fleetboard. At CAN bus and FMS bus level, we integrate via standard protocols and open-source libraries. We build the AI layer on top so that your planners and drivers don't have to work in a new system.
Does this also apply to multimodal transport (rail, water, road)?
Yes. Multimodal flows do add complexity: transshipment points, different emission factors per mode, and different document flows and regulations. We model the modes as separate links in a single optimisation graph and respect each mode's own constraints. For distribution centres that act as a hub, we combine yard management AI with routing on the inbound and outbound flows.
How do you stay vendor-agnostic between AI vendors such as Loginext, FarEye, Onfleet and Routific?
By assessing, for each use case, what works well off the shelf and what doesn't. For narrow, well-defined sub-problems (simple last-mile dispatch, for example), a specialised vendor is often faster and cheaper than a custom build. For flows with specific constraints, integrations or compliance requirements around driver monitoring, a dedicated solution is unavoidable. We give you honest advice, even if that means: 'buy this package and you're done'.

Talk to us about AI in your transport operation.

A half-hour introduction where we go through where your data currently sits, which use case shows the most promise, and what a first sprint would look like. No obligation, in your language, without a demo of an AI platform you don't actually want.

Reply within 1 working day
No-obligation conversation
Westerdoksdijk 599, Amsterdam

Edit content