roomkey24Suite
Legal · AI Act Art. 50 / GDPR Art. 22

AI Disclosure

Last updated: 3 June 2026

Several Roomkey24 Suite features use generative-AI models. This page lists every AI-assisted feature, which model receives data, what data is sent, and the human-review safeguard that applies before any output takes effect.

Models we use

  • Google Gemini 2.5 Flash / Flash-Lite — OCR, classification, suggestion and extraction tasks. This is the model behind every AI feature listed below except the RMS revenue advisor.
  • OpenAI GPT-5.6 — used only for the RMS revenue-advisor chat, where pricing judgement needs a reasoning-grade model.

All AI calls are routed through the Lovable AI Gateway, which forwards prompts to the chosen model and returns the response. We do not train or fine-tune any model on your data, and we contractually prohibit our providers from training their foundation models on inputs sent via the API.

Features that use AI

  • Past-rota analysis (Teams · Gemini 2.5 Flash). Inputs: an uploaded past/legacy rota file (any format) plus the employee/template list. Outputs: per-employee suggested recurring pattern (preferred template, days per week, recurring days off). Suggestions pre-fill employee records for a manager to review and edit — nothing is applied without confirmation.
  • Menu photo/PDF import (F&B · Gemini). Inputs: uploaded menu photo or PDF. Outputs: proposed categories, items and prices. Staged for review before anything is imported.
  • Menu → ingredient & recipe suggestions (F&B · Gemini). Inputs: menu item names and categories. Outputs: suggested stock ingredients and recipe quantities per item. A manager reviews and edits every suggestion before it's saved.
  • Delivery-docket import (F&B · Gemini). Inputs: photo or PDF of a supplier delivery docket. Outputs: proposed stock delivery lines. Staged for review before anything is recorded.
  • Legacy stocktake import (F&B · Gemini). Inputs: photo, scan or PDF of a paper/spreadsheet stock count sheet. Outputs: proposed counted stock lines. Opens a draft stocktake for a manager to review and confirm before it's finalised.
  • Restaurant floor-plan photo import (F&B · Gemini). Inputs: a photo of the restaurant room. Outputs: proposed table positions, shapes and capacities. A manager drags, edits and confirms every table placement before saving — this remains a prototype feature and review is mandatory.
  • Work-order AI diagnosis (Facility · Gemini). Inputs: a guest or staff fault report (text plus optional photo) and an anonymised PMS snapshot (room occupancy/arrival/departure status). Outputs: fault category, severity, likely cause, first action, and a priority score. Feeds a work order for staff to action — nothing is dispatched automatically.
  • Purchase-order drafting (Facility · Gemini). Inputs: a vendor and the parts at or below reorder level. Outputs: a drafted purchase-order email. A manager reviews and explicitly approves before it sends, and only to vendors flagged as approved.
  • Parts import (Facility · Gemini). Inputs: a supplier parts list (CSV or photo). Outputs: proposed parts-catalogue rows. Staged for review before anything is imported.
  • Fiscal Z-report photo extraction (Finance · Gemini). Inputs: a photo or PDF of the daily Z report (fiscal closing receipt) printed by a certified cash register. Outputs: extracted totals — gross, tax, net, void count/amount, payment-method split. A manager reviews every figure against the paper report and confirms before it's saved at till close; nothing is written from the photo unreviewed.
  • RMS revenue advisor (RMS · OpenAI GPT-5.6). Inputs: your property's own performance history and pace snapshots, plus the GM's typed question. Outputs: written pricing and strategy advice. Read-only and advisory — the advisor cannot change settings, apply rates or push to channels.
  • Employee CSV import (Teams · Gemini). Inputs: header row and a small sample of the spreadsheet rows the user uploaded. Outputs: a column-mapping suggestion. Mapping is shown for confirmation before the import runs.
  • Compliance document OCR & classification (Teams / Facility · Gemini). Inputs: image/PDF of the uploaded document. Outputs: extracted text, document-type label, expiry date guess, confidence score. Confidence scores below 0.75 are flagged for mandatory human review before the document is treated as valid evidence.
  • Bank-statement categorisation (Finance · Gemini). Inputs: transaction description, amount, date. Outputs: suggested category and counterparty. Suggestions never auto-post; an accountant reviews them in the reconciliation screen.
  • Rate recommendations (RMS · Gemini). Inputs: anonymised occupancy, historical ADR, market signals, lookahead window. Outputs: suggested nightly rate per room-type per date. Review-then-apply is the default: a manager commits each rate. A workspace owner can opt in to an "auto-apply recommendations daily" setting, which lets the scheduled sync apply and push rates without per-rate review; it is off unless explicitly enabled. The hotel operator is solely responsible for the rates they publish to OTAs and direct channels.
  • Mailroom document classification (Teams · Gemini). Inputs: contents of HR emails ingested via the dedicated mailroom address. Outputs: document type (payslip / tax notice / certificate / other), suggested employee match, suggested filing category. Same 0.75-confidence rule applies; low-confidence items land in a review queue.
  • Live front-desk translator (PMS · Gemini 2.5 Flash). Inputs: the transcript of the guest's spoken message (captured by your browser's own speech recognition and sent as text), plus extracts of your property's uploaded SOP documents used to ground the answer. Outputs: an English translation of what the guest said, a suggested reply drafted from those SOPs and general hospitality knowledge, and the translation of the reply into the guest's language. Staff review and can edit every suggested reply before it is translated and shown to the guest — nothing reaches the guest automatically. The conversation is ephemeral: no transcript, translation or reply is written to Roomkey24's database, and everything is cleared when the conversation is ended or the page is left. To be precise, ephemeral does not mean the text never leaves the browser — the transcript and reply do transit the Lovable AI Gateway to Gemini for processing, exactly like the other AI features listed here; they are simply not stored afterwards.

Human-in-the-loop guarantee

No AI output in the Roomkey24 Suite is used for automated decision-making about individuals within the meaning of Art. 22 GDPR without human oversight. Every output is either:

  • presented to an authorised user for explicit approval before it produces an effect, or
  • treated as a non-binding suggestion that the user can accept, edit or discard.

Document-classification outputs with confidence below 0.75 are flagged in the review queue and cannot be auto-accepted by the system.

The one exception is the optional RMS "auto-apply recommendations daily" setting described above. It is opt-in per property, disabled by default, applies only to commercial pricing (never to decisions about individuals), and can be switched off at any time. Every automatically applied rate is recorded in the recommendation history.

What we do NOT do with AI

  • We do not generate your weekly rota with AI — the scheduler is a deterministic algorithm that applies your templates, contracted days, availability and time-off rules.
  • We do not screen, score, hire or fire employees with AI.
  • We do not deny service to guests based on AI output.
  • We do not feed your data into model training.
  • We do not publish rates to channels automatically unless a workspace owner has explicitly enabled auto-apply for that property.

Reporting AI errors

If an AI feature returns an output that is offensive, discriminatory, or factually harmful, report it to ai-safety@roomkey24.com. Workspace owners can also disable individual AI features in their settings.