Reports on request, for management decisions
“Which product group had the best margin last autumn?” – managers ask, the AI answers from your own data with a table and a recommendation. This is what clients ask it most.
→ no report building
Artificial intelligence
AI that works. Not hype.
Reports for management decisions, warehouse proposals, document generation – inside your existing systems, with AI receiving only the anonymised data it needs.
Demo
One question. Anonymised data. A ready proposal.
A manager asks. Before anything reaches the AI, we mask names, addresses and anything the decision doesn’t need. The answer comes from your own stock and usage data.
- 1. Question
- 2. Masking
- 3. Analysis
- 4. Proposal
Privacy: AI only receives what the decision needs
- Smith & Co Ltd.→Customer #A17
- 12 High Street, London→[address removed]
- Peter Brown (buyer)→[name removed]
- + stock, daily usage, shelf, lead time ✓
Based on the last 8 weeks of usage and a 3-day lead time:
| Item | Shelf | Stock | Use/day | Order |
|---|---|---|---|---|
| Mineral water 1.5 l (6-pack) | A-03 | 42 | 18 | 110 |
| Raspberry syrup 1 l | B-11 | 9 | 4 | 24 |
| Soda water 19 l bottle | C-01 | 60 | 7 | – |
Refill shelf B-11 first: current stock lasts ~2 days, lead time is 3 days.
Use cases
Where most of the manual work is today.
Warehouse refill & ordering
What should I order? What is on which shelf, and what to refill first based on usage? Proposals come from real stock, usage and lead times.
→ fewer stock-outs and overstock
Non-invoice documents
Waybills, delivery notes, replenishment slips – what you can’t download from the tax authority, AI helps bring into the system.
→ less manual typing
Company document generation
Contracts, quotes, internal policies and letters from your templates – here you knowingly provide the required data.
→ draft in minutes
Searchable company knowledge
Ask internal documents and policies in plain language – the AI only sees what the user is allowed to.
→ with sources
Early warnings
Unusual changes in stock, receivables or usage – before they become problems.
→ early signal
Data security
AI only sees what the decision needs.
Data security comes first. We have never lost data since we started – and we introduce AI the same way.
Masked data
A report doesn’t need customer names or addresses – the AI never receives them. Only for document generation do you knowingly provide what’s required.
On-premise model
If needed, the model runs on your servers – data never leaves the company.
Hungarian data centre
For cloud models: a data centre in Hungary, no training on your data.
Inherited permissions
The AI only sees and reveals what the user is allowed to access.
GDPR & EU AI Act
Risk classification, logging, human oversight.
How to start
Start small. Measure everything.
No need for a big, uncertain AI project. An assessment and a prototype will show whether it pays off.
011–2 weeks
AI assessment
Workshop with leaders and key users. We map where AI would bring measurable return – and where it wouldn’t.
- Prioritised use cases with ROI estimates
- Data and security review
022–4 weeks
Prototype
We build a working prototype on your own data for one chosen process and measure accuracy.
- Working prototype
- Measured accuracy and time saved
03per project
Rollout & operations
We integrate it into your existing systems with permissions, logging and monitoring – and keep measuring.
- Production rollout
- Ongoing quality tracking
From the inside too
We work this way too.
AI already helps with testing, code review and documentation – so we ship faster without cutting quality, because every change is approved by a person.
Tech radar$ git log --oneline -4
a41f2c9 feat(invoice): AI ledger suggestions
9be03d1 test: 142 new cases (AI-generated, human-reviewed)
5c7e880 docs: API reference updated
e02b4a7 fix(nav): retry token on timeout
✓ 1,284 tests passed · review: 2 people · deployed: 11:42
GYIK
AI questions
Does our data go to an AI provider?
Only if you decide so. We can use models hosted in Hungarian data centres with no training on your data, or run a model fully on your own servers.
What does the AI see of our data?
Only what the question needs. For a report it doesn’t receive customer names, addresses or contacts – we replace them with IDs before anything reaches the model. When generating a document (e.g. a contract), you knowingly provide the required data.
What if the AI gets it wrong?
We keep a human in the loop: uncertain cases are flagged, extracted data is offered for approval and every step is logged. Accuracy is measured continuously.
How does this comply with the EU AI Act?
We classify each use case by risk category at the start and align documentation, transparency and oversight requirements accordingly.
Ask your own data.
After a short call we’ll tell you which process to start with – or that it’s not worth it yet.
+36 70 626 5550 · iroda@frederik-technologies.com
A real team answers – not a call centre.