Let AI make the routine decisions. Your people keep the hard ones.
We build AI decision-making systems that approve, reject, route and moderate on your behalf: refunds, claims, supplier checks, user content, expense reports, support tickets. Each decision comes with its reasoning. When the AI isn't sure, or the case is one you've said a person must handle, it goes to a named person on your team. Every decision is logged, so you can always show why it was made.
It's human-in-the-loop AI built the way the GDPR and the EU AI Act expect, connected to the systems you already run, and operated by us for a fixed monthly fee.
Trusted since 2004 by the public sector, energy and heavy industry: buyers who audit their suppliers before they sign. See the case studies.
Decision automation for high-volume, rule-heavy work
Content moderation
Reviews, listings, comments, forum posts and uploaded documents checked against your own policy, not a generic one. Clear cases are published or removed, borderline ones go to a moderator.
Refunds, claims and returns
Checks the order, the policy, the photos and the customer's history, then approves, rejects or asks for more. Large amounts and repeat claimants always go to a person.
Supplier and customer checks
Documents, public registers, sanctions lists and expiry dates. Approves the clean files and explains exactly what's missing from the rest.
Expense and invoice approval
Matches spending to policy and budget, approves what's in line and stops what isn't, with a one-line reason your finance team can check in seconds.
Ticket and request triage
Reads every incoming email, ticket or form, decides what it is, how urgent it is and who should handle it, and answers the simple ones directly.
Access and permission requests
Grants routine access that matches someone's role, flags requests that don't, and keeps the record your NIS2 auditor will ask for.
How human-in-the-loop AI works in practice
- A confidence threshold you set. Above it, the AI decides. Below it, the case goes to a person with the AI's reasoning and the evidence attached.
- Always-human rules. Amounts over a limit, certain customers, legal complaints, anything you name. These never get an automatic decision.
- Random spot checks. A sample of automatic decisions goes to a reviewer every day, so you measure accuracy instead of assuming it.
- A right to appeal. Anyone affected by a decision can contest it and gets a human review, as the GDPR requires.
- A safe failure mode. If the model is down or behaving oddly, everything goes to your people. Work slows down, but nothing wrong gets approved.
- A full decision log. Input, decision, reasoning, model version, who reviewed it and when. Searchable and exportable.
Automated decision-making and the GDPR
Article 22 of the GDPR gives people the right not to be subject to a decision made solely by a machine when it has legal or similarly significant effects on them. Where it applies, you need a person who can step in, a way to contest the decision, and an explanation of the logic behind it.
We design that in from day one: which decisions can be fully automatic, which need a person to confirm, and what you tell the people affected.
Human oversight under the AI Act
Some uses, such as credit scoring, recruitment and access to essential services, are classed as high-risk under the EU AI Act and need documented human oversight. Most moderation and back-office decisions aren't high-risk, but buyers and auditors increasingly ask for the same evidence anyway.
We classify each use case during the assessment and write the oversight records while we build, not afterwards.
What it costs
A fixed-price assessment, a pilot, then one fixed monthly fee.
We don't bill per decision or per model call. The assessment usually takes a week and is credited toward the build. The pilot runs next to your team so you can compare the AI's decisions with theirs before anything goes live. After that, we run it under SLA as part of managed AI operations.
Start with one decision
Pick the decision your team makes most often and likes least. On a free 30-minute call we'll tell you how much of it AI could take, where the human fallback should sit, and roughly what it would cost. It's the same approach behind our AI agents for invoices, tenders and contracts.
Common questions
Can AI make decisions on its own?
For routine, well-defined cases, yes. You set a confidence threshold and a list of cases that always need a person. Above the threshold the AI decides; below it, or in any case on the always-human list, a named person on your team decides, with the AI's reasoning in front of them.
What does human in the loop mean?
It means a person can review, change or overrule what the AI does. In our systems that happens three ways: uncertain cases are escalated, a random sample of automatic decisions is checked every day, and anyone affected can appeal to a human.
Is automated decision-making allowed under the GDPR?
Yes, with conditions. Article 22 restricts decisions based solely on automated processing that have legal or similarly significant effects. Where it applies, you need a legal basis, the right to human intervention, a way to contest the decision and meaningful information about the logic. We design the process around those requirements.
What happens if the AI makes a wrong decision?
It's caught by the daily spot checks, by the person who appeals, or by the escalation rate changing. Every decision is logged with its input, reasoning and model version, so you can find and reverse similar ones. If quality drops below the agreed threshold, more cases go to people until it's fixed.
Can you moderate content in languages other than English?
Yes. We regularly work in English, Romanian and German, and the models handle most European languages. Your moderation policy is written once and applied in every language you receive content in.
How is it priced?
A fixed-price assessment, usually one week and credited toward the build, then a pilot, then a fixed monthly fee to run it. We never bill per decision or per model call.
Find out in 30 minutes if it's worth doing.
Tell us about one process that eats your team's week. We'll tell you on the call whether an agent can take it over, roughly what it costs and what it saves. If it's not a fit, we'll say so.