How the AI assistance works

This page describes what happens between your click on an AI button and the proposed report: which data is processed, what is kept and where the limits are. For the buttons themselves, see AI Report Assistant.

What you use this for

• Explaining to a customer or manager what happens to the data.
• Deciding which AI service you configure for your environment.
• Knowing what the assistant does and does not do before you rely on it.
• Checking in the log how heavily it is used.

Who can use it

Administrators who configure the service, and managers who read the log.

Prerequisites: an AI service configured by your administrator. Without that, the AI buttons return no result.

The flow

User clicks an AI button
            │
            ▼
   Service order: the raw note
            │
            ▼
   AI service  ──(fails)──▶  Fallback model
            │                      │
            └──────────┬───────────┘
                       ▼
              Proposal as a draft
                       ▼
        A human reads and approves
                       ▼
       Saved on the service order
                       │
                       ▼
                    Logged

Step by step

1. You click. Nothing happens without an explicit action by a logged-in internal user. There is no automatic processing and no background job rewriting your reports.

2. What leaves. The raw note and the report fields of that one service order. No customer database, no price list, no invoices, no data from other customers.

3. The service processes the request. Your administrator chooses which service is used.

4. Fallback on failure. If the chosen service doesn’t respond, a second model is used automatically so your team isn’t blocked.

5. You get a draft. The result lands in an editable field on the service order — not in an email, and not with the customer.

6. A human approves. Nothing is sent without someone having read it.

7. It is logged. Every request is kept, noting which service was used. That log is visible only to managers.

Supported services

The available AI services are configured by your administrator. Several providers are possible; you choose which one is used for your environment. Which one, and with which fallback model, is visible in the settings.

Choosing a service

Sectors and customers have different requirements about where data may be processed. That is why the service is a setting, not something baked in. If you have a specific requirement, discuss it with your administrator.

Security

• Access keys stay on the server and are never sent to the browser.
• There are no public addresses: only logged-in internal users can use the assistant.
• Nothing is ever sent automatically to a customer.
• No data leaves the system without an explicit action by a user.
• Every request is traceable in the log.

The limits

Limits of the assistant

• The assistant does not write technical truth. It rephrases what a technieker entered. If the note contains an error, the polished report contains it just as much.
• It does not invent facts: where information is missing, it stays neutral rather than making up something plausible.
• It does not assess safety and replaces no statutory inspection.
• Translations are working translations. For a legal or contractual document, use a sworn translation.
• It is not deterministic: the same note twice can give two slightly different phrasings. That is normal.

This stands in sharp contrast to the health score of an asset: that uses no AI and always gives the same figure for the same data. That is deliberate — a figure you base an investment decision on should be traceable.

Best practices

• Let techniekers just write in their own words. Keywords are enough; the assistant handles the form.
• Read every proposal before it goes to a customer. It is a draft.
• Use it mainly for form, not for substance.
• Agree internally what is drafted with AI and what isn’t, so your reports stay consistent in tone.
• Check the log if you want to know how heavily it is used.

Troubleshooting

Does my customer data go to an external service? The raw note and the report fields of that one job are processed. Not your customer database, your prices or your invoices.

Can I switch the AI features off entirely? Yes. Your administrator can turn them off; the system carries on working without them.

Who can see what was asked? Managers, via the log.

Can the assistant email a customer by accident? No. It creates a draft. Sending is always a separate, human action.

Next step

• AI Report Assistant — the buttons and the usage
• Integrations — where AI connections are configured
• Permissions & security — who may do what
• Asset Intelligence — deliberately without AI
• Report & signature — where the note comes from