Understanding AI
AI in your application — what you should know.
You want the AI solution. But you do not want to explain why the application made a mistake.
This page does not explain what AI can do. It shows which mechanisms make an application reliable — and how you recognize these mechanisms.
What carries it
Reliability is not the model — it is the application around it
As long as AI was a tool operated by a human, safety was a question of operation. Today it sits inside the process — and its uncertainty becomes a property of your application.
The model guarantees nothing
The model is right most of the time, but not always. Nobody can predict errors. The application must be built for exactly this reality.The tenth case decides
A model that is right in nine out of ten cases carries a reliable application — if the system catches the tenth case. The same model ruins the application if an error triggers a payment, a customer email or a contract unchecked.The difference lies in the application
You are not buying a reliable model. You are having a reliable application built around an unreliable model.The Mechanics
The 6 most important protection mechanisms
Security limits what a mistake does. Quality ensures there are fewer mistakes. These 6 mechanisms must be built into every serious AI application.
1. Limited Reach
The application gets its own account with minimal permissions. What it cannot see, it cannot leak. (Analogy: Access permissions)2. Named Approval
Irreversible actions (payments, contracts) stop at an approval gate. A human decides. (Analogy: Signature rules)3. Hard Rule Check
Fixed rules (price bands, discounts) check every proposal before the effect. In doubt, nothing goes out. (Analogy: Four-eyes principle)4. Clean Handover
Below a certainty threshold, the AI does not guess, but hands over to a human. A quality feature, not a defect.5. Predictable Fallback
If the AI fails or hallucinates, a classic process takes over. Nobody has to improvise.6. Continuous Monitoring
Metrics show when behaviour shifts (more handovers, longer answers) before a customer reports it.Checklist
The 5 most important questions for every provider
Ask them of every supplier – including us. Anyone who evades these has not built the mechanisms.
Question 1: Limits
Which part of the application can the AI influence — and which part guaranteed not?Question 2: Errors
What happens to the one case out of ten where it is wrong?Question 3: Approvals
Which actions are irreversible, and who signs them by name?Question 4: Uncertainty
What does the application do when it is unsure: guess, stop or hand over?Question 5: Failure
What continues to run if the AI component fails?No slide
Where we already build like this
We recommend nothing that we have not built ourselves first.
Federal portal naturgefahren.ch
Hazard data from multiple sources becomes public warnings. During a severe weather situation, everyone accesses it at the same time — that is why the check before every change is non-negotiable.Federal portal MeteoSwiss
Almost 100 weather products in real time. Since January 2026, no change bypasses the automatic checks — they are not a byproduct, but the prerequisite for the speed.In-house Quote Generator
Our own product: the AI drafts, fixed rules check price and conditions, approval remains with a human. Exactly the setup this factsheet describes.Questions
What decision-makers ask us about this
Which AI approach fits your case?
This page says what has to be built in. Which kind of AI your case actually needs — rules, prediction, language or an agent — is answered by the technology page, including a free advisor. And whether the effort pays off for you is worked through on a concrete case in The AI potential in your tenders.
Ready for the next step?
Michael — Use our enterprise portal software to build innovative digital products. Michael is happy to advise you.