AI in logistics and transport, where the investment comes back
Routing, capacity planning, documentation and visibility: the cases where artificial intelligence reduces operating costs in logistics.
In logistics margins are thin and costs are visible, which means any percentage improvement shows up quickly. It's an industry where artificial intelligence has proven cases and easily verifiable results.
Routing and delivery planning
The case with the most direct return. A system accounting for time windows, vehicle capacity, service times per stop, access restrictions and traffic produces routes that manual planning rarely matches once there are many stops.
The benefit appears in three places: fewer kilometres, more deliveries per vehicle, and better compliance with committed windows.
The condition is input data quality. Badly recorded addresses and unloading times estimated from memory will ruin the best algorithm.
Capacity planning
Anticipating volume by zone and day allows fleet and shift sizing with less error. It avoids both the cost of idle capacity and the cost of hiring emergency transport, which is always the most expensive.
It requires volume history at daily and zone granularity.
Documentation and trade
Delivery notes, bills of lading, packing lists, customs declarations, carrier invoices. High volume, variable formats and manual transcription work.
It's the case that gets the least attention and returns some of the most hours, especially in international operations.
Visibility and exceptions
The value isn't showing where each shipment is — it's anticipating which ones will have a problem and flagging them before the customer calls.
A system that detects deviations from plan and prioritizes exceptions by impact turns a reactive team into one that manages ahead of events. The indicator that improves is the number of incidents resolved before a complaint arrives.
Fleet maintenance
Where telemetry exists, anticipating failures from usage patterns. The saving isn't only the repair: it's avoiding a vehicle stopped mid-route, with a committed delivery and the cost of urgent replacement.
Logistics customer service
Enquiries about shipment status, rescheduling and complaints. High and repetitive volume. Automating status enquiries is among the first things worth doing, with clear routing for complaints and exceptions.
What you need before starting
Three conditions determine whether the project will work:
Normalized and geocoded addresses. The number one problem in the region, and the one that takes longest to fix.
Real service times. How long a delivery actually takes at each type of customer, not how long it should take.
Documented constraints. Receiving hours, access requirements, specific conditions at each delivery point. This usually lives in planners' heads rather than any system.
That last point deserves attention: much of a logistics operation's knowledge is unwritten. Making it explicit is a prerequisite, and it usually has value in itself.
What to measure
- Kilometres per delivery and deliveries per vehicle per day.
- Compliance with committed time windows.
- Cost per delivery, with all components included.
- Anticipated incidents versus complaints received.
- Administrative hours spent processing documentation.
If the first two improve and window compliance holds or rises, the project is working. If kilometres drop but compliance worsens, you optimized the wrong variable.
Frequently asked questions
How much can route optimization with AI save?
In urban distribution with dozens of stops per vehicle, 8% to 20% reductions in distance and time are common results. The range depends on how manual the previous planning was.
What data is needed to start?
Normalized addresses, real time windows, vehicle capacities and service times per stop. What most affects results are badly recorded addresses and unrealistic unloading times.
Is it worth it for small fleets?
From around five vehicles with many stops there's a gain. With two or three vehicles on stable routes, manual planning is usually sufficient.