AI is a copilot
Iron Watch AI may interpret commands, summarise information, extract proposed fields, rank eligible workers and explain recommendations. It does not replace workforce supervision.
A human should review decisions materially affecting employment, earnings, work opportunities, discipline, legal rights, health or safety.
Deterministic facts remain authoritative
Objective matters such as stored expiry dates, shift overlap, configured credentials and recorded unavailability are determined by application rules where technically supported. AI is not authoritative for licence validity, credential validity, tenant boundaries or final assignments.
- Licence and credential validity
- Availability and recorded conflicts
- Site requirements and inductions
- Existing assignments and tenant facts
Human review
Material staffing decisions remain subject to authorised human review. Iron Watch recommendations depend on accurate underlying data. Final staffing responsibility remains with the authorised scheduler.
Workers should have a reasonable way to query or correct information used in significant decisions.
Document onboarding
AI may propose fields extracted from authorised credential documents. Proposed information requires authorised confirmation and does not mark a credential verified. If a field cannot be extracted reliably, it should be entered manually.
Fair and lawful use
Customers must not configure or use Iron Watch to discriminate unlawfully. Sensitive characteristics must not be used unless lawful and genuinely necessary. Iron Watch should not rank workers primarily by who historically says yes fastest.
Operational resilience
AI assistance is designed not to become a single point of operational failure. If AI is unavailable, deterministic Iron Watch functions can continue to provide eligibility, compliance checks, availability, conflicts, Rapid Cover and manual assignment.
When this occurs, the platform may state: “AI assistance temporarily unavailable. Core rostering remains available.”
Privacy and security
AI requests should be tenant scoped, limited to required context, schema validated and handled through server-side credentials. OpenAI API credentials must never appear in frontend bundles or ordinary logs. No cross-tenant AI context is permitted.