We put AI on the production line, in the invoice tray and in the shared inbox, and measure it in euros.
Every part, checked.
A camera and a vision model at the line inspect each part in 38 ms and send rejects to the PLC. Press “Invoices” or “Shared inbox” to see the office side.
What we build.
Systems that run inside a real process, measured against a baseline, with a person in the loop where it matters.
Visual quality inspection
Every part, at line speed.
Predictive maintenance
Failures flagged days ahead.
Production scheduling
Re-plans in seconds when orders change.
Invoice & order processing
PDF to ERP, exceptions to people.
Inbox triage
Sorted, routed and answered in draft.
Contract review
Risky clauses flagged, with a redline.
Energy optimisation
Heavy loads moved to cheaper hours.
Why pilots stall.
Most AI projects die between the demo and the daily routine. These are the five reasons we see most, and what we do about each.
The demo used clean data.
We pilot on your real, messy data from day one: scanned invoices with stamps on them, parts under your lighting, emails in three languages. If it only works on clean data, we find out in week two, not month six.
Nobody agreed on success.
Before we write code, we write down the baseline and the target: escape rate, straight-through rate, hours per week. The pilot ends with a go or no-go against that number, not a feeling.
It lived in a new screen.
Results land where people already work: the PLC, the CMMS, the ERP, Outlook and Teams. A tool that needs its own login and its own habit gets abandoned by the third week.
No one owned it after launch.
Models drift when suppliers, products, lighting or seasons change. We monitor accuracy, retrain and report monthly, or hand over to your team with the documentation to do it themselves.
AI was the wrong tool.
Sometimes a rule, a report or a better form does the job. When that’s the case we say so in the audit, and you don’t pay for a model you don’t need.
One quarter, end to end.
Works with what you already run.
We connect to ERP and accounting by API, database or file exchange, and to machines through OPC UA, MQTT, Modbus or retrofit sensors. Hosted in the EU or on your own servers.
Audit
For teams who want to know where AI pays back before spending on it.
We interview operations, finance and IT, score every candidate use case on value and feasibility, check your data, and hand over a 12-month roadmap with a business case for the top three.
2 weeksfixed price
Request an auditPilot
For teams with one problem worth solving this quarter.
One use case built on your real data, with the success metric agreed before we start. You get a working system, a baseline-versus-AI report and a clear go or no-go. If you scale with us, the pilot fee is credited, and AI Operations then runs it for a monthly fee you can cancel.
6 weeksfixed scope
Start a pilotStraight answers.
We’re a 50-person company. Is AI really for us?
Often more than for large enterprises. Smaller teams feel every hour of manual work, and modern AI no longer needs a data science department. Most of our work is with companies of 20 to 500 people.
Do we need a lot of data to start?
Less than you think. Document and email automation works from day one on pre-trained models. Visual inspection needs a few hundred example images, which we help you collect. The audit tells you exactly what you have and what is missing.
Where does our data go?
It stays in the EU: in an EU data centre or on your own servers, and fully offline on the shop floor when needed. We sign a data processing agreement and never use your data to train models for anyone else.
What about the EU AI Act?
Most operational use cases (documents, inspection, maintenance) are minimal or limited risk. We record the classification for every system and, where obligations apply, deliver the technical documentation, logging and human oversight the Act requires.
Who makes the final decision, the AI or us?
You do. Our systems propose, people decide. Confidence thresholds send uncertain cases to the right person, and anything that moves money or reaches a customer has an approval step.
How accurate will it be?
We’ll tell you after the audit, not before, because it depends on your data. A single accuracy figure is the wrong question anyway. What matters is how often the system is right when it’s confident, and what happens when it isn’t. Low-confidence cases always go to a person, so a mistake costs a review, not a wrong shipment.
Will it work with our ERP and machines?
Almost certainly. If a system holds data, we can usually reach it by API, database, file export or, for machines, OPC UA, MQTT, Modbus or a retrofit sensor. We read from PLCs; we don’t change your control logic.
What does it cost?
The audit and the pilot are fixed-price, agreed in writing before we start. A production system is a one-off integration fee plus a monthly operations fee. We only recommend scaling when the pilot shows a payback you’re comfortable with.
Who owns what you build?
You do: the models trained on your data, the prompts, the pipelines and the documentation. Everything is built on open standards so you can take it in-house or to another partner.
Tell us what slows you down.
Thirty minutes, no slides. We’ll say honestly whether AI is the right tool, and if it isn’t, what is. Or write to hello@aisensus.eu.