A sentence I hear in first calls in variations: "We spent a lot of money on an AI project last year. What came out is a chatbot nobody uses." The amount changes. The story stays the same.
The answer is uncomfortable: what got automated was the visible problem, not the expensive one. Customers call, so the website got a chatbot. Meanwhile dispatch retypes orders, accounting checks freight invoices by hand, and someone asks the driver for arrival times. That costs money every day, and it never had a project.
Mid-sized companies do not have an AI problem. They have a selection problem.
The chatbot wins because it can be shown. The managing director can present it to the board. The vendor can demo it in an hour. And it touches no system that hurts. Order capture, by contrast, sits in the TMS, in the mailbox and in the habits of two dispatchers. It is invisible and expensive.
BVL surveyed over 200 specialists and executives in February 2026. Two thirds want to introduce or expand AI. The obstacles are missing expertise, data quality and resources. None of these shrinks if you start with the chatbot. All three shrink if you start with a process whose data you know afterwards.
At the forwarders we work with, it is always the same four. Not because they are exciting. They run daily, have rules and can be measured.
Order capture from mail and PDF. Five to fifteen minutes per order by hand. A service reads mail and attachment, checks against the customer master and creates the order in the TMS. The dispatcher only sees the cases that need a decision. At a wholesaler with 50 orders a day, processing time went from three and a half hours to under one.
Customer portal without an API. An agent operates the portal like an employee, checks each order against the TMS and confirms. At a contract logistics provider that is 900 to 5,300 orders per run. The customer's API had been announced for two years. Now nobody needs it.
Arrivals from telematics. Read positions every 15 minutes, calculate against time slots, notify dock and customer. At a forwarder with 40 trucks: zero calls to drivers, about 45 minutes per dispatcher and day saved.
Reconciling toll, trip data and invoice. Arithmetic that never happens by hand, because nobody lays 1,000 bookings against 1,000 trips. The service does it per tour and outputs a list of deviations, before payment.
What the four share: no new system, live in two to four weeks, and a protocol that shows whether they delivered.
One process costs two to four weeks of work plus operations. Against it stands the manual work, priced at the full hourly rate. According to Destatis, an hour of work cost 45 euros on average in 2025. For the four processes above, break-even sits between one and three months. A chatbot nobody uses never has one.
(In defense of the chatbot: there are companies where 200 customer requests a day ask for the same shipment status. There it is the right first process. At a forwarder with regular customers it is not.)
Three things that appear in offers and rarely help:
A caveat: the pattern comes from forwarders and warehouses with 20 to 500 employees. In a company whose customers are consumers, the order shifts and the customer dialog moves forward.
At the wholesaler from the order capture case study, the first service was order capture from the mailbox. Two weeks of build, then a review list with a handful of cases a day instead of 50 retyped orders. A chatbot never made the list, because nobody missed it afterwards.
Which process did you automate last, and which one does your dispatch still retype? Name the second. I will show you, on our example, what it looks like after four weeks. Ask for a process check: 30 minutes, no sales pitch.
