The Logistics Benchmark 2025 checked 54 logistics companies against 120 criteria in December. 55 percent offer a customer portal, 20 percent one with real functional value. The number describes portals, but it applies just as well to AI solutions: many have one, few have one that does something in operations.
Here are the seven questions we would ask a solution before it enters dispatch, warehouse or billing. With the answers we give ourselves, and the places where the answer is uncomfortable.
A solution that runs in operations has to be able to say what it did today. Read, checked, booked, held for review, per run and per case. Not as a dashboard with curves, but as a table a dispatcher can read.
Our answer: every run writes an entry with numbers and error texts. For a portal agent that is 39 lines in eight weeks. From them the customer learned that corrections come on Wednesdays.
Uncomfortable: a protocol also shows when the service did nothing. That is the point.
The decisive question. A solution that books when in doubt is dangerous. One that aborts when in doubt is useless. The right answer is a review list: the case, the reason, the proposal, and a person decides.
Our answer: at a wholesaler's order capture, 20 percent of orders land in the review list. PDF and proposal sit side by side. The clerk decides with one click, and the service learns nothing from it on its own. That is deliberate.
"Running" is not proof. A service can start every morning, get stuck on a silently failed prior step and end without a result, for weeks. The question is whether an alarm hangs on the result.
Our answer: the watchdog counts bookings, notices, confirmations. If the number stays at zero on a working day although input arrived, an alarm goes to us and to the customer. We learned that when a service ran for a week and delivered nothing.
Customer portals, slot bookings, telematics screens. The answer "then we need an API" means in practice: no project. The question is whether the solution can operate the screen like an employee, and whether it logs that.
Our answer: API where there is one. Screen where there is not, with a protocol per click. And the switch to the API as soon as it arrives, without dispatch noticing.
Order data, customer masters, vehicle positions. A solution that sends this to a server outside Europe is a contract problem for many of a forwarder's customers, long before it is a technical one.
Our answer: the data stays in the customer's systems. Intermediate storage sits in data centers in Finland and the EU. The language model only receives what it needs to read, and keeps nothing.
A solution only the vendor understands is a dependency. The question is whether the customer could take the service over. Does the code sit with him? Is it documented? Are the blocks interchangeable?
Our answer: every process is a small, tested program with a protocol, not a click workflow in someone else's account. The blocks behind it are interchangeable, and the customer gets a handover with which his IT can keep running the service. Whether they want to is another question, but they should be able to.
Not per license, per user or per month, but per process. Against it, the manual work at the full hourly rate, which Destatis puts at 45 euros per hour on average for 2025.
Our answer: two to four weeks of build per process, break-even with well-chosen processes in one to three months. And the answer "this does not pay at your company" when the manual work is under 5,000 euros a year. We give it in every third first call.
It does not check whether the model behind it is the newest. That is the question vendors like to answer, and for operations it hardly matters. A two-year-old model with protocol, review list and alarm beats the newest one without the three.
A caveat: the seven questions come from our work with forwarders and warehouses of 20 to 500 employees. For a group with its own platform strategy, questions on integration and operating model come on top that are missing here.
Which offer is on your desk right now? Send me the vendor's seven answers, or the places where none stands. I will tell you which of them will hurt in operations. Ask for a process check: 30 minutes, no sales pitch.
