Overview
Hiro’s AI Assistant brings conversational intelligence to your CRM. Ask questions in natural language and get instant insights about reservations, customers, revenue, and operations—without navigating through menus or building reports.Powered by: OpenAI GPT-4 with hospitality-specific training and your real-time operational data.
Key Capabilities
Natural Language Queries
Ask questions in Spanish or English just like talking to your team:
- “¿Cuántas reservas tenemos hoy?”
- “Show me VIP customers who haven’t visited in 3 months”
Real-time Data
Responses use live data from:
- Today’s reservations
- Customer database
- Revenue metrics
- Visit patterns
Context-Aware
Understands hospitality context:
- “Detractors” = customers with NPS ≤ 6
- “VIP” = top-tier loyalty customers
- “Revenue” = calculated from reservations × avg ticket
Actionable Insights
Not just data—get recommendations:
- Recovery strategies
- Re-engagement tactics
- Operational optimizations
How It Works
Query Router
When you ask a question, Hiro’s AI follows this process:Intent Classification
Analyzes your question to determine intent:
- Reservations query?
- Customer lookup?
- Revenue analysis?
- Operational help?
Context Loading
Fetches relevant real-time data:
- Today’s reservations across all locations
- Weekly revenue trends
- VIP customer status
- Inactive customer lists
- General statistics
Response Generation
Two modes depending on setup:
- OpenAI Mode: Full GPT-4 response with conversational tone
- Structured Mode: Pre-formatted responses with data tables
Query Types
Reservation Queries
Ask about today’s reservations:Example: Daily Reservations
Example: Daily Reservations
Query: “¿Cuántas reservas tenemos hoy?”Response:
Example: VIP Reservations
Example: VIP Reservations
Query: “Show me today’s VIP reservations”Response:
Customer Queries
Find and analyze customer segments:Example: Inactive VIPs
Example: Inactive VIPs
Query: “¿Qué clientes VIP no han visitado en 3 meses?”Response:
Example: Customer Lookup
Example: Customer Lookup
Query: “Find customer named García”Response:
Revenue Queries
Analyze financial performance:Example: Weekly Revenue
Example: Weekly Revenue
Query: “¿Cuál ha sido el revenue esta semana?”Response:
Analytics Queries
Get operational insights:Example: General Statistics
Example: General Statistics
Query: “Dame un resumen general del negocio”Response:
Example: No-Show Analysis
Example: No-Show Analysis
Query: “Analyze no-show rate and give recommendations”Response:
Help & Recommendations
Example: General Help
Example: General Help
Query: “What can you help me with?”Response:
Example: Optimization Recommendations
Example: Optimization Recommendations
Query: “Give me recommendations to increase occupancy”Response:
OpenAI Configuration
Setup Requirements
To enable full AI Assistant capabilities:Enter API Key
Input your OpenAI API key:
- Get key from https://platform.openai.com
- Requires paid OpenAI account
- Recommended: Create dedicated key for Hiro
Configure Model
Choose model:
- gpt-4-turbo (Recommended): Best quality, slower
- gpt-4: High quality, balanced speed
- gpt-3.5-turbo: Faster, lower cost, less capable
Set Parameters
Adjust settings:
- Temperature: 0.7 (default) for balanced creativity
- Max Tokens: 1000 (default) for detailed responses
Fallback Mode
If OpenAI is not configured:- AI Assistant still works with structured responses
- Uses pre-formatted templates
- No conversational depth
- Limited to predefined query types
Cost Estimate: With gpt-4-turbo, typical query costs €0.02-0.05. For 100 queries/day, expect ~€3-5/day in API costs.
Privacy & Security
Data Handling
What's Sent to OpenAI
- Your natural language query
- Aggregated statistics (counts, sums)
- NO customer emails or phone numbers
- NO payment information
- NO personal identifying data
What Stays in Hiro
- Full customer profiles
- Contact information
- Payment details
- Reservation notes
- All sensitive data
OpenAI Data Policy
Per OpenAI’s Enterprise Agreement:- API data NOT used for model training (as of March 2023+)
- Data NOT retained after 30 days
- GDPR compliant
- ISO 27001 certified
Important: Never ask the AI Assistant to “send” customer data externally. It has no email/export capabilities and won’t expose sensitive data.
Best Practices
Be Specific
“VIP customers inactive 3 months” is better than “show me customers”
Use Time Ranges
“This week” or “last 30 days” gives more relevant results than no time frame
Ask Follow-ups
After getting a list, ask “tell me more about the first one” for details
Request Actions
Ask for recommendations: “How should I handle this?” or “What’s the best approach?”
Advanced Features
Semantic Customer Search
Find customers by description, not just name: Query: “High-spending customers who love wine and haven’t visited recently” AI Process:- Identifies criteria: high spend + wine preference + inactive
- Searches customer tags for “wine” or “wine lover”
- Filters by spend (top 25%)
- Filters by last visit (60+ days)
- Returns ranked results
SQL Generation (Advanced)
For complex queries, AI generates SQL: Query: “Show me average party size by day of week for last month” AI Process:- Recognizes need for grouping and aggregation
- Generates SQL query (executed safely)
- Formats results in readable table
- Provides interpretation
SQL generation is sandboxed and read-only. No data modification possible.
Trend Analysis
AI identifies patterns: Query: “Are we getting busier or slower?” Response:Troubleshooting
AI Not Responding
AI Not Responding
Possible Causes:
- OpenAI API key invalid or expired
- Rate limit exceeded
- OpenAI service outage
- Check Settings → Integrations → OpenAI
- Verify API key is active in OpenAI dashboard
- Check OpenAI status page
- Falls back to structured mode automatically
Slow Responses
Slow Responses
Causes:
- GPT-4 models slower than 3.5
- Complex queries require more processing
- Large datasets (many customers/reservations)
- Switch to gpt-3.5-turbo for speed
- Reduce max_tokens to 500
- Be more specific in queries (narrows data set)
Inaccurate Answers
Inaccurate Answers
Causes:
- Ambiguous question
- Missing context in data
- AI hallucination (rare)
- Rephrase question more specifically
- Ask for data source: “How did you calculate that?”
- Cross-reference with manual reports
- Report persistent issues to support
Technical Details
Query Processing:- Average response time: 2-5 seconds (GPT-4)
- Timeout: 30 seconds
- Concurrent queries: Unlimited (rate-limited by OpenAI)
- Loads last 7 days of reservations
- Top 100 VIP customers
- Last 30 days of revenue data
- Ensures fast responses with relevant data
- Follows OpenAI tier limits (usually 10,000 requests/day)
- No Hiro-side limits
- Automatically retries on transient errors