AI projects are the ones we most want to hear about.
If your operation already uses AI, or you are deciding whether it should, talk to us. We build the part that makes a model useful in daily work: clean data in, a checked action out, and a person who stays in charge.
Four kinds of project, one common problem: the model is the easy part.
Documents that people still retype
Orders, delivery notes, contracts, invoices, service reports. A model reads them, your system receives structured records, and a person approves the ones that matter.
Decisions that wait for someone to look
Which vehicle takes the job, which machine goes to service first, which alert is real. AI proposes and gives its reasons; the dispatcher or operator decides.
Questions buried in operational data
Telemetry, work orders and costs that nobody has time to read. We make them answerable in plain language, with the source records attached.
A pilot that never reached production
The demo worked, daily use did not. We find out whether the gap is data, integration or trust, and close it.
What AI does today in EFFI, the logistics system we build.
EFFI is our logistics operations system, in daily use at road freight companies. These parts of it run on language models.
Reading documents
An uploaded paper is read by its content, not by its file name. The system proposes the type, the owner (driver, vehicle or company) and the expiry date; for contracts also the notice period. A person approves the proposal.
Orders from email
Incoming transport orders and cost documents are read automatically and turned into a draft transport or cost entry, ready to be checked.
AI co-workers
Roles such as secretary, dispatcher and controller prepare proposals inside the system. Every proposal waits for human approval, and each company has its own cost limit.
What we add to an AI project
The data path
A model is only as good as what reaches it. Collection, normalization and integration are our daily work.
A place in the system people use
The result lands on the dispatcher’s screen or in the ERP, not in a separate chat window.
Controls
Approval steps, permissions, an audit trail, and a spending limit per company or team.
Operation
Monitoring, a fallback when a model is unavailable, and the option to change provider without rewriting the product.
How we build AI features
- A person approves what matters
- AI proposes, people decide. Nothing that changes money, contracts or schedules happens without a human step.
- The model sees only what it needs
- We send the text in question, not the customer list. Names and records are matched inside your system.
- Costs are capped and visible
- Every call is metered against a budget, so a busy month cannot turn into a surprise invoice.
- It works without the model too
- When a provider is down, the work continues by hand or by rules, and the system says so clearly.
We do not train models and we do not write AI strategy papers. We build the software around a model so that it holds up in daily operation.