ORIN RESERVATIONS
Designing the Voice of AI in High-Pressure Hospitality
ORIN is an AI-powered reservation platform designed for restaurants to manage guest communication and bookings in real time. The system uses conversational messaging to automate replies, confirm reservations, and capture guest preferences.
UX Writer
3 months
Hospitality / SaaS
Conducted a linguistic audit of over 200 historical guest enquiries to identify high-frequency "Anchor Words" and recurring request patterns.
Observed restaurant staff during peak shifts to map the "Social Translation" friction, the mental effort required to pivot from floor service to professional digital messaging.
Synthesised research findings into a strategic Affinity Map, defining the core tonal pillars for the orinAI messaging voice.
Designed a decision-based messaging framework that transitions the user task from "composition" to "selection," significantly reducing cognitive load.
Authored all AI prompt logic and microcopy for the "Intent Toggles" to ensure a consistent, brand-aligned British hospitality tone.
Problem
In the high-pressure environment of a busy restaurant, communication is often the first casualty of a hectic shift. Staff members are forced to switch constantly between physical service and managing digital guest enquiries.
The Linguistic Barrier: Manual responses are often inconsistent, overly blunt, or prone to typos due to time constraints.
Cognitive Friction: The mental effort required to draft professional, personalised replies mid-service creates a significant "bottleneck," leading to delayed responses and lost booking opportunities.
Brand Risk: Incoherent or slow communication damages the restaurant's reputation for high-standard hospitality.
Objectives
My goal as a UX Writer was to design a communication framework where AI acts as a "Co-pilot" rather than a replacement for human hospitality.
Minimise Cognitive Load: Design a "decision-based" UI that eliminates the need for staff to compose text from scratch.
Standardise Voice & Tone: Ensure every AI-generated message feels professional, welcoming, and aligns with British hospitality standards.
Ensure Accuracy: Create a system that dynamically injects specific guest data (dates, times, special requests) into drafts to prevent manual errors.
Solutions
Engineered a "Mirroring" prompt logic that automatically echoes guest-specific needs (e.g., "high chair") to build immediate trust.
Introduced Intent-Based Toggles ("Reply Positively" / "Reply Negatively") to act as cognitive shortcuts for staff mid-service.
Developed the "Three-Sentence Hierarchy" for AI drafts to ensure messages are scannable in under 3 seconds.
Integrated dynamic data injection to automatically pull booking dates and times into drafts, ensuring 100% accuracy.
Designed the "Improve" feature logic to polish raw staff notes into professional, welcoming brand-aligned greetings.
Established a "Human-in-the-Loop" editable draft system, ensuring staff maintain full control over every AI-generated response.
Discovery
To build a functional and empathetic AI voice, I conducted a multi-layered research process focusing on the linguistic habits of both guests and staff.
Linguistic Audit: Patterns of Inquiry
Method: I performed a thematic analysis of hundreds of historical guest messages within the platform. I categorised these enquiries by intent and urgency.
Finding: I identified that 85% of enquiries were "Conditional Requests". Guests were not just asking for availability; they were seeking confirmation for specific needs like "baby chairs" or "window tables" before committing to a booking.
Key Insight: AI responses must acknowledge specific keywords (e.g., "baby chair") to ensure the guest feels "heard". A generic "Yes" is insufficient.
Contextual Inquiry: The "Mid-Shift" Pressure
Method: I conducted observational research during "peak hours" in active restaurant environments. I tracked the time taken for staff to switch from floor service to digital messaging.
Finding: While staff knew the factual answer (Yes/No), the cognitive friction occurred during the "social translation" phase. They struggled to pivot from physical tasks to composing a polite, brand-aligned response.
Key Insight: The solution shouldn't just provide an answer; it needs to provide a complete social script. The speed of drafting is more critical than the complexity of the message.
Prompt Iteration: Refining the AI Tone
Method: I prototyped three distinct AI voice tiers Functional (Direct), Concierge (Verbose), and Peer-to-Peer (Casual) and tested them with real users.
Finding: "Functional" felt too robotic and dismissive (e.g., "Request Confirmed"). "Concierge" was too verbose, requiring staff to spend too much time reading and verifying the draft.
Key Insight: The "Hospitality Sweet Spot" is a balance of efficiency and warmth. The AI must be concise enough for the staff to scan in 2 seconds, but polite enough for the guest to feel welcomed.
Key Insights Summary (The "Aha!" Moments)
Impact on Design
Linguistic Audit
Guests use specific "anchor words" (e.g., window, high chair) as a prerequisite for booking.
I designed the AI to repeat these "anchor words" back to the guest to build immediate trust.
Contextual Inquiry
Composing a polite reply mid-service takes 3x more mental effort than a simple "Yes/No" decision.
I introduced Intent-Based Toggles (Positive/Negative) to turn a writing task into a simple selection task.
Prompt Iteration
Long AI drafts increase cognitive load rather than decreasing it.
I restricted AI outputs to a maximum of 3 sentences, prioritising clarity and scannability for the staff.
Affinity Diagram
I categorised the findings into four strategic pillars:
Guest Intent & "Anchor Words": My analysis revealed that guests use specific terms such as "window table," "high chair," or "allergy" as prerequisites for their booking.
Insight: For an AI response to feel authentic, it must mirror these specific keywords back to the guest to build immediate rapport.
Staff Operational Barriers: I identified a significant "Tonal Anxiety" among staff. While they knew the technical answer (Yes/No), they struggled with the mental effort of phrasing it politely mid-service.
Insight: Staff do not need a blank text box; they need Intent-Based Shortcuts that handle the social etiquette for them.
The Scannability Mandate: Observations during peak hours showed that staff have less than 3 seconds to review a message.
Insight: AI drafts must follow a strict "Three-Sentence Hierarchy" to ensure they can be verified at a glance.
Safety & Control: There was a recurring fear of AI "hallucinations" or incorrect data.
Insight: To ensure adoption, the AI must function as an editable draft co-pilot, never as an autonomous sender.
1. Trust is Built Through Echoing
The linguistic audit proved that guests feel "ignored" if a response is too generic. I discovered that repeating the guest's specific request (e.g., "We've noted your request for a window table") was the most effective way to provide psychological validation without human intervention.
2. Transitioning from Composition to Selection
The primary cause of delayed responses was "Social Translation Fatigue." Staff lacked the mental energy during a rush to craft polite sentences. This insight led me to prioritise Intent-Based Toggles (Positive/Negative) over a standard chat interface, reducing the task from "writing" to "confirming."
3. Precision Over Poetry
Through prompt iteration, I learned that "flowery" or long AI drafts increased cognitive load rather than decreasing it. In a high-intensity environment, brevity is hospitality. We established a rule to limit AI outputs to three punchy, professional sentences that prioritise clarity.
Results
Ideating solutions
The analysis of my Affinity Map led to a series of high-impact linguistic and structural decisions. My goal was to move beyond "generic AI" and create a voice that feels native to high-end hospitality.
Echoing
Decision 1: Implementing "Echoing" as a Trust Metric
The Logic: Based on the "Anchor Words" identified in research, I established a rule: the AI must mirror the guest's specific nouns in the first five words of the response.
The Writing Action: If a guest asks about a "high chair," the AI is prompted to start with: "We have reserved a high chair for you..." instead of a generic "Your request is confirmed."
Impact: This reduces the guest's need to follow up, as they feel their specific needs have been accurately processed.
Decision 2: Transitioning from Free-Text to Intent-Based Toggles
Decision 3: The Scannability Hierarchy (The 3-Sentence Rule)
The Logic: Staff need to verify AI drafts in under 3 seconds. Long paragraphs were a friction point.
The Writing Action: I enforced a strict linguistic template:
Greeting & Echo: (Personalised validation).
Core Data: (Date/Time confirmation).
The Hook: (A warm brand-aligned closing).
Impact: This ensures the draft is scannable even during the busiest dinner rushes.
Prototype




