AI and Heuristic Functions
As of 2026-05-06
Inventory of all automated evaluations, AI components and rule-based heuristics in the platform — pursuant to Art. 50 AI Act (EU) 2024/1689 and § 12 Usage Agreement.
Active rule-based heuristics (NOT AI under the AI Act)
| Function | Description | Method | Risk class |
|---|---|---|---|
| Filename auto-categorisation for PO documents | Detection of typical Mittelstand filenames (AB_*.pdf → order confirmation, Zeugnis_3_1*.pdf → mill cert 3.1, EMPB* → first-article inspection) | Regex on filename + extension | not AI (rule-based) |
| Filename auto-categorisation for RFQ neutral data | Analogous detection for RFQ attachments (Zeichnung* → 2D drawing, *.step → 3D model) | Regex | not AI |
| BOM filename match | Auto-assignment of files to BOM positions via drawing number or position pattern (Pos010_*) | String match against rfq_items.drawing_number / position pattern | not AI |
| ABC classification supplier (Pareto 80/15/5) | Statistical classification based on 12-month spend | Aggregate SQL with cumulative percentages | not AI |
| IBAN validation (mod-97 check) | Check-digit per ISO 13616 | Mathematical formula | not AI |
| Revision bump (DIN 199-4) | Auto-increment of change index (A → B → … → Z → AA, "01" → "02") | Finite state machine | not AI |
These heuristics make no decisions with legal effect or significant impact on the user (Art. 22 GDPR). They are pure suggestions which the user can override at any time.
Planned AI functions
| Function | Status | Expected risk class | Compliance requirements |
|---|---|---|---|
| WebLLM-based chat assistant | not productive | AI system (Art. 3) — risk class depends on context | Art. 50 transparency (UI banner), not high-risk per se |
| ML-based supplier scoring | not planned | High-risk under Annex III No. 5(b) if it affects the business relationship | Before deployment: conformity assessment, EU DB entry, human oversight |
| Automated document classification (replacing regex heuristic) | not planned | Low-risk, suggestion-only | Transparency + opt-out |
Training data exclusion
Nexus M does not use customer data to train own or third-party AI models. This commitment is part of:
- Usage Agreement § 8(2) + § 12
- Data Processing Agreement § 2(2)
- NDA module § 4
Opt-out
When AI-supported analyses with bearing on a user's business relationship are introduced, an opt-out toggle will be provided in organisation settings.
Updates
This inventory is updated before any AI function goes live, and users are informed (Art. 50 AI Act).