AI for logistics

Somebody in your office is reading a screen out loud to a customer who wants to know where a pallet is. It is the most common inbound message in the industry, the answer is already in your own system, and that is where AI for logistics pays first — not in the route optimisation every pitch leads with, which matters at forty vehicles and is noise at six.

The four holes, in cost order

The same data typed three times. It arrives by email, goes into the transport system, gets re-entered for the customer's portal, and appears again on the invoice. Four copies, four chances to disagree, and a person doing it who could be selling capacity.
"Where is my delivery?" It is the single most common inbound message in the industry, the answer is already in your system, and a person is reading it out. That is a full-time job in a mid-sized operation, and it is a job nobody wanted.
Exceptions found late. A delivery that failed at 11:00 and is discovered at 16:00 has cost a redelivery and a phone call from an angry customer. The information existed at 11:00; nobody was watching.
Proof of delivery chased by hand. Missing paperwork is unpaid money. The chase is boring, repetitive and always at the bottom of somebody's list.

What it must not decide alone

Re-planning a route mid-day. It can propose. A traffic office knows things about that driver and that customer which are in nobody's database.
Telling a customer a delivery will be late without a person seeing it first, where the relationship is worth more than the notification.
Accepting a job outside your operating licence, weights or hours. Those are hard limits and they are coded as refusals.
Anything touching driver hours compliance. It surfaces the risk. A transport manager decides.

What gets built, and why that order

One clean record per consignment, assembled from wherever the data arrives — email, portal, spreadsheet, EDI. Everything downstream depends on this and nothing else works without it.
Status answered automatically, in whatever channel the customer already uses, from that record. With a rule about when it should hand over to a person instead.
Exception watching: a delivery that has not moved, a driver hours problem, a temperature reading out of range — surfaced within the hour, not at end of day.
Proof of delivery collected, matched to the job and chased when missing, so the invoice can go out on time.

Too early if

Consignments live only in email threads. Getting to one record per job is the whole first project, and it is worth doing before anything with AI in the name.
Under a few hundred movements a month. At that volume the office is the system and a build will not beat it.
Your customers all use their own portals and will not take a message from you. Then the status work has a ceiling, and the exception work is where to start instead.

The same machinery, running on other freight

Data platform — the one-record problem solved on somebody else's data: several sources reduced to one true record per item, with an AI layer answering from it. A consignment is the same shape as a product.
New-arrivals radar — the honest counter-example: it makes no judgement at all, it pulls and ranks on a schedule. Half of logistics automation is the same — knowing which work only needs a schedule saves the budget for the work that needs judgement.

Questions from the traffic office

Is AI for logistics mostly route optimisation?

That is what gets marketed. Under about forty vehicles the payback is usually somewhere else: status enquiries answered without a person, exceptions surfaced within the hour, and consignment data entered once instead of three times.

Can it answer 'where is my delivery' properly?

Yes, provided there is one record per consignment to answer from. Without that you get a confident answer from the wrong copy of the data, which is worse than no answer. That is why the record comes first in every build.

Will it work with our transport management system?

It works alongside it. Where the system offers a proper interface, that is used. Where it does not, a scheduled two-way file exchange is honest and it works — and it can be upgraded later without redoing the parts built on top.

What about EDI and customer portals?

Both are just another way data arrives and leaves. The point of one clean record is that adding a fifth channel stops being a project and becomes a configuration.

We are a freight forwarder, not a haulier. Does that change it?

The paperwork gets heavier and the case gets stronger. Forwarding is document work — bookings, customs, proof of delivery — and document work is exactly what this handles best.

Send me a week of status enquiries

Just the messages asking where something is, however they arrive. I will tell you what share could be answered from your own data without a person, what it would take to get there, and whether the exception watching is worth more to you than the answering. The first hour is free.

Book that hour →