Skip to main content
Etienne Intelligence is powered by 9 specialized AI agents that work autonomously across your business, handling thousands of tasks daily without human intervention.

​
Agent Architecture

Each AI agent is purpose-built for a specific domain, allowing for deeper expertise and better performance than a single generalist AI.

Command Center

3 agents managing conversations

Scheduling

3 agents optimizing appointments

Intelligence

3 agents analyzing revenue

​
Agent Status Indicators

Every agent displays a real-time status visible throughout the platform:

​
Online (Green Pulse)

Visual: ● with animated pulsing ring Meaning: Agent is actively processing tasks and responding to events in real-time. What the agent is doing:
  • Monitoring for new tasks (conversations, bookings, alerts)
  • Processing queued work
  • Learning from outcomes to improve performance
  • Communicating with other agents when needed
Typical state: 90%+ of the time for most agents

​
Idle (Yellow)

Visual: ◐ solid yellow dot Meaning: Agent is waiting for new tasks. No current work in queue. What this means:
  • Not an error — agent is healthy and ready
  • Common for batch-processing agents (Demand Forecaster, Report Generator)
  • Will transition to Online when triggered
Example: Demand Forecaster runs weekly analyses, so it’s Idle between runs.

​
Error (Red Pulse)

Visual: ⚠ with animated pulsing ring Meaning: Agent encountered an issue and is attempting to recover. Common causes:
  • Temporary API rate limit (Zenoti, Anthropic)
  • Network connectivity hiccup
  • Unexpected data format
  • High computational load
Auto-recovery: Agents automatically retry every 2 minutes. 95% of errors resolve within 10 minutes. When to contact support:
  • Error persists >30 minutes
  • Multiple agents in error state simultaneously
  • Critical agent (Conversation Analyst, Response Monitor) stuck in error
If you see the Response Monitor in error state during business hours, contact support immediately. This impacts your ability to respond to customer inquiries.

​
Command Center Agents

Three agents work together to manage all customer communications:

​
Conversation Analyst

Type: Analysis
Module: Command Center
Status: ● Online (typical)
Tasks Handled: 1,247 (monthly average) Responsibilities:
  • Real-time speech-to-text conversion
  • Speaker diarization (AI vs. Client vs. Staff)
  • Conversation summarization
  • Sentiment detection (positive/neutral/negative)
  • Booking request detection
  • Question vs. complaint classification
  • Urgency assessment
  • VIP client identification
  • Upsell potential in conversation
  • Package deal suitability
  • Client education needs
  • Referral opportunities
  • Medical concern keywords
  • Negative sentiment patterns
  • Billing dispute language
  • Policy exception requests
Technology:
  • Speech recognition: Deepgram real-time ASR
  • Natural language: Claude 4.5 Sonnet
  • Sentiment: Custom model trained on med spa conversations
Performance metrics:
  • 98.7% transcription accuracy
  • 94% intent classification accuracy
  • 89% escalation prediction accuracy
  • Less than 2 second analysis latency

​
Response Monitor

Type: Monitoring
Module: Command Center
Status: ● Online (typical)
Tasks Handled: 2,381 (monthly average) Responsibilities:
  • Monitors all incoming inquiries across channels
  • Timestamps first response
  • Calculates and reports response time SLAs
  • Alerts on delayed responses (>5 minutes)
  • Routes SMS and web chat conversations
  • Manages conversation state (active/waiting/resolved)
  • Ensures no inquiry goes unanswered
  • Handles conversation handoffs (AI to staff)
  • Recognizes same client across channels
  • Prevents duplicate responses
  • Maintains conversation history context
  • Syncs updates across platforms
  • 24/7 monitoring (no downtime)
  • Instant response to after-hours inquiries
  • Flags urgent issues for on-call staff
  • Tracks after-hours conversion rates
Technology:
  • Message processing: Custom event queue system
  • Response generation: Claude 4.5 Sonnet with RAG
  • Context memory: Vector database for conversation history
Performance metrics:
  • 99.2% uptime (including after-hours)
  • 8-12 second average response time
  • Zero missed inquiries (100% response rate)
  • 78% AI resolution rate (no human needed)

​
Escalation Tracker

Type: Tracking
Module: Command Center
Status: ● Online (typical)
Tasks Handled: 312 (monthly average) Responsibilities:
  • Identifies appropriate team member for each escalation
  • Considers expertise, availability, and workload
  • Provides full context and transcript
  • Tracks handoff completion
  • Assigns urgency levels (urgent/pending/normal)
  • Re-prioritizes based on client responses
  • Escalates further if no staff pickup within SLA
  • Manages VIP client priority queue
  • Tracks time-to-resolution for escalations
  • Follows up if conversation stalls
  • Requests client satisfaction after resolution
  • Feeds resolution patterns back to training
  • Analyzes why escalations occurred
  • Identifies AI knowledge gaps
  • Recommends training data additions
  • Reduces future escalation rate
Technology:
  • Routing logic: Multi-factor optimization algorithm
  • Priority scoring: ML model trained on outcomes
  • Tracking system: Real-time event stream processing
Performance metrics:
  • Less than 15 minute average staff pickup time
  • 92% first-contact resolution
  • 18% month-over-month reduction in escalation rate (AI learning)
  • 4.6/5 average client satisfaction on escalated issues

​
Scheduling Agents

Three agents collaborate to optimize your appointment book:

​
Schedule Analyst

Type: Scheduling
Module: Scheduling
Status: ● Online (typical)
Tasks Handled: 1,893 (monthly average) Responsibilities:
  • Calculates real-time utilization by provider, room, location
  • Identifies underutilized time slots
  • Predicts utilization 7-14 days out
  • Recommends slot consolidation opportunities
  • Matches waitlist clients to cancellations in under 30 seconds
  • Prioritizes by VIP status, wait time, service value
  • Sends instant notifications to matched clients
  • Tracks waitlist conversion rates
  • Routes new bookings to optimize utilization
  • Reserves prime slots for high-value services
  • Balances workload across providers
  • Suggests alternative times for better efficiency
  • Books multiple services in logical sequence
  • Identifies room/provider conflicts
  • Optimizes same-day multi-service bookings
  • Manages couple/group bookings
Technology:
  • Optimization engine: Constraint satisfaction solver
  • Matching algorithm: Weighted scoring with real-time updates
  • Prediction model: Time series forecasting (Prophet)
Performance metrics:
  • 68% waitlist conversion rate
  • 18.5% average utilization improvement
  • Under 30 second waitlist fill time
  • 94% booking accuracy (no double-books)

​
No-Show Predictor

Type: Prediction
Module: Scheduling
Status: ● Online (typical)
Tasks Handled: 647 (monthly average) Responsibilities:
  • Calculates no-show probability for every appointment
  • Updates scores as new signals arrive (confirmations, etc.)
  • Categorizes as low/medium/high risk
  • Triggers prevention workflows for high-risk
  • Sends 24-hour reminders to medium+ risk
  • Sends 2-hour reminders to high risk
  • Personalized messaging based on client preferences
  • One-click confirm/cancel links
  • Identifies client-specific no-show patterns
  • Day-of-week and time-of-day correlations
  • Service type risk factors
  • Lead time impact on no-show likelihood
  • Measures prevention success rate by strategy
  • A/B tests reminder timing and messaging
  • Calculates ROI of no-show prevention
  • Feeds results back to prediction model
Technology:
  • Prediction model: Gradient boosting (XGBoost) with 15+ features
  • Reminder system: Multi-channel message orchestration
  • A/B testing: Bayesian optimization for message variants
Performance metrics:
  • 73% prevention success rate for high-risk appointments
  • 57% overall no-show rate reduction
  • 85% prediction accuracy (correctly identifies high-risk)
  • 24,000−24,000-35,000 monthly revenue saved per location

​
Demand Forecaster

Type: Analytics
Module: Scheduling
Status: ◐ Idle (typical)
Tasks Handled: 89 (monthly average) Responsibilities:
  • Forecasts appointment demand by day/time/service
  • Identifies seasonal patterns
  • Predicts impact of marketing campaigns
  • Projects capacity needs 30-90 days out
  • Suggests optimal provider schedules
  • Identifies overstaffed vs. understaffed periods
  • Recommends cross-training opportunities
  • Calculates staffing ROI scenarios
  • Identifies price-sensitive vs. price-insensitive time slots
  • Recommends dynamic pricing opportunities
  • Suggests promotional timing for maximum impact
  • Models revenue impact of pricing changes
  • Projects when current capacity will be insufficient
  • Models scenarios (new provider, new room, new location)
  • Estimates revenue impact of expansion
  • Identifies optimal timing for growth investments
Technology:
  • Forecasting: Facebook Prophet with custom seasonality
  • Scenario modeling: Monte Carlo simulation
  • Recommendation engine: Multi-objective optimization
Performance metrics:
  • 89% forecast accuracy (within 10% of actual demand)
  • 12% revenue increase from optimized staffing
  • 3-6 month advance warning on capacity constraints
  • Runs weekly (hence typically shows “Idle” status)
The Demand Forecaster runs comprehensive analyses every Sunday night. It’s normal to see “Idle” status during the week. Check the dashboard on Monday mornings for new insights.

​
Intelligence Agents

Three agents power your revenue analytics:

​
Revenue Analyst

Type: Analytics
Module: Intelligence
Status: ● Online (typical)
Tasks Handled: 412 (monthly average) Responsibilities:
  • Tracks revenue by source (AI, staff, online, walk-in)
  • Calculates AI contribution vs. baseline
  • Attributes revenue to marketing channels
  • Measures ROI of AI system
  • Identifies revenue trends (daily, weekly, monthly)
  • Detects anomalies and investigates causes
  • Compares locations and providers
  • Projects future revenue based on trends
  • Calculates response gap (slow/missed inquiries)
  • Measures no-show prevention impact
  • Identifies upsell opportunities missed
  • Quantifies after-hours revenue capture
  • Powers conversational business intelligence
  • Answers natural language questions
  • Generates custom analyses on demand
  • Provides actionable recommendations
Technology:
  • Analytics engine: Custom Python + Pandas + NumPy
  • Attribution model: Multi-touch attribution with decay
  • Chatbot: Claude 4.5 Sonnet with RAG over your data
Performance metrics:
  • 97% revenue attribution accuracy (validated against Zenoti)
  • Under 5 second response time for chatbot queries
  • 92% user satisfaction with chatbot answers
  • 30K−30K-50K monthly revenue gaps identified per location

​
Opportunity Scout

Type: Analytics
Module: Intelligence
Status: ● Online (typical)
Tasks Handled: 238 (monthly average) Responsibilities:
  • Scans for churn risk (clients not returning)
  • Identifies rebooking opportunities
  • Finds upsell gaps (clients who usually bundle)
  • Detects underutilized capacity
  • Estimates revenue potential for each opportunity
  • Calculates probability of conversion
  • Prioritizes by expected value
  • Tracks opportunity decay over time
  • Suggests specific actions for each opportunity
  • Provides messaging templates
  • Recommends optimal timing for outreach
  • Assigns to appropriate team member
  • Monitors opportunity resolution
  • Measures conversion rate by opportunity type
  • Calculates actual revenue captured
  • Improves recommendations based on results
Technology:
  • Discovery: Pattern matching + anomaly detection
  • Valuation: Regression models trained on historical conversions
  • Recommendation: Reinforcement learning (improving over time)
Performance metrics:
  • 40-60 opportunities identified per location per month
  • 58% conversion rate when acted upon
  • 8,000−8,000-15,000 monthly revenue from opportunity actions
  • Hourly scan frequency (always finding fresh opportunities)

​
Report Generator

Type: Reporting
Module: Intelligence
Status: ◐ Idle / ⚠ Error (typical)
Tasks Handled: 156 (monthly average) Responsibilities:
  • Weekly performance summary (sent Mondays)
  • Monthly executive report (sent 1st of month)
  • Custom recurring reports (user-configured)
  • Automated distribution via email
  • Ad-hoc report generation on demand
  • Custom date ranges and filters
  • Comparative analysis (locations, providers, periods)
  • Export to PDF, Excel, CSV
  • Charts and graphs for all metrics
  • Location comparison dashboards
  • Trend visualizations
  • Executive summary slides
  • HIPAA-compliant data handling
  • Audit trail generation
  • Data retention policy enforcement
  • Regulatory reporting support
Technology:
  • Report engine: Custom templating system
  • Data export: Pandas + Openpyxl + ReportLab
  • Scheduling: Cron-based job system
Performance metrics:
  • 98% on-time report delivery
  • Under 5 minute generation time for standard reports
  • Supports 50+ custom report templates
  • Note: Often shows Idle (waiting for schedule) or Error (resource-intensive jobs)
The Report Generator is computationally expensive and may show Error status during large report generation. Jobs are queued and will complete. Only contact support if error persists >30 minutes.

​
Inter-Agent Communication

Agents don’t work in isolation—they communicate to provide seamless experiences:

​
Example: Booking Flow

1

Response Monitor receives inquiry

Client sends web chat message: “Can I book Botox this Saturday?”Response Monitor detects booking intent.
2

Conversation Analyst extracts details

  • Service: Botox
  • Preferred day: Saturday
  • Location: (inferred from IP or asks client)
  • No specific time mentioned
3

Schedule Analyst finds optimal slot

  • Checks Saturday availability
  • Considers utilization (slots other services to balance)
  • Offers 10 AM, 2 PM, 4 PM (avoiding over-booking Dr. Chen)
4

No-Show Predictor assesses risk

  • New client = higher baseline risk
  • Web chat booking = medium risk (vs. phone = lower)
  • Saturday = higher risk
  • Calculates 45% no-show probability
  • Tags as “medium risk” for automated reminders
5

Response Monitor confirms booking

Sends confirmation: “Great! You’re booked for Botox on Saturday at 2 PM at SoHo with Dr. Chen. Total is $450. Confirmation sent to your email!”
6

Opportunity Scout checks for upsell

Analyzes:
  • Botox clients at this location also book Hydrafacial 40% of the time
  • Suggests add-on
Response Monitor: “Many clients love adding a Hydrafacial ($250) to their Botox visit. Interested?”
7

Revenue Analyst tracks attribution

  • Logs booking source: Web chat (AI)
  • Revenue: 450(or450 (or 700 if Hydrafacial added)
  • Attributes to “After-hours capture” if outside business hours
  • Updates Intelligence Hub metrics
Total interaction time: 8-12 seconds
Agents involved: 5
Human intervention: Zero

​
Agent Learning & Improvement

All agents use machine learning to improve over time:

​
What Agents Learn From

Conversation outcomes:
  • Which responses led to bookings vs. abandonment
  • Optimal phrasing for common questions
  • When to escalate vs. continue handling
  • Sentiment patterns that predict client satisfaction
Scheduling patterns:
  • Which no-show predictions were accurate
  • Effectiveness of different reminder timing/messaging
  • Utilization optimization success rates
  • Waitlist conversion factors
Revenue analytics:
  • Which opportunities had highest conversion
  • Accuracy of demand forecasts
  • Attribution model refinements
  • Chatbot answer quality ratings

​
Continuous Improvement

Weekly model updates:
  • Conversation Analyst: Retrained on last week’s conversations
  • No-Show Predictor: Updated coefficients based on actual outcomes
  • Opportunity Scout: Adjusted value estimates based on conversions
Monthly reviews:
  • Response Monitor: Optimize routing logic
  • Schedule Analyst: Refine optimization algorithms
  • Revenue Analyst: Validate attribution accuracy
Quarterly enhancements:
  • New features based on user feedback
  • Integration of latest AI model improvements (e.g., Claude upgrades)
  • Performance benchmarking and tuning
You don’t need to do anything to trigger these improvements. Agents automatically learn from every interaction. The longer you use Etienne, the better it performs for your specific business.

​
Agent Monitoring & Health

​
Dashboard Visibility

Every module (Dashboard, Command Center, Scheduling, Intelligence) shows the agents relevant to that module. What you see:
  • Agent name and type
  • Real-time status (Online/Idle/Error)
  • Tasks handled (monthly count)
  • Last activity timestamp

​
Health Checks

Every agent performs self-checks every 60 seconds:
  1. Connectivity: Can reach all required APIs (Zenoti, Anthropic, etc.)
  2. Performance: Processing tasks within SLA (under 5 second response time)
  3. Accuracy: Recent predictions/actions within quality thresholds
  4. Queue depth: Not backed up with too many pending tasks
If any check fails, agent status changes to Error and auto-recovery begins.

​
Support Escalation

You should contact support if: Critical agents in error >10 minutes:
  • Conversation Analyst
  • Response Monitor
  • Schedule Analyst
These impact live customer interactions. Multiple agents in error simultaneously: Suggests system-wide issue rather than individual agent problem. Agent stuck in error >30 minutes: Auto-recovery should have resolved by now. When contacting support, provide:
  1. Agent name and status
  2. Time error started
  3. Screenshot of agent status panel
  4. Any error messages visible in UI

​
Best Practices

Make it part of your opening routine:
  1. Open Dashboard
  2. Scroll to AI Agents module
  3. Verify all critical agents show Online (green)
  4. If any errors, wait 5 minutes and check again
  5. If error persists, contact support
Takes under 30 seconds and prevents issues.
Agents are highly accurate but not perfect:
  • Spot-check AI conversations weekly (review 10-15)
  • Compare AI bookings to what you would have done
  • Verify no-show predictions after appointments occur
  • Validate opportunity recommendations before acting
This builds confidence and identifies edge cases for improvement.
When you see something wrong:
  • Use “Report Issue” button in conversation transcripts
  • Describe what should have happened
  • Your feedback directly improves the agents
Example: If AI escalated a conversation unnecessarily, report it. The Escalation Tracker will learn.
Track “Tasks Handled” counts month-over-month:
  • Growing counts = agents handling more (good, means business growing)
  • Declining counts = potential issue (less activity, or agent not working)
  • Sudden spikes = investigate (could be error causing false triggers)
Example: If Response Monitor usually handles ~2,400 tasks/month and suddenly jumps to 8,000, investigate why.

Dashboard

View all agent statuses in one place

Command Center

See conversation agents in action

Intelligence Hub

Interact with the Revenue Analyst chatbot

Build docs developers (and LLMs) love

Get started for freeTalk to us