Ayra AI
|Docs
Fundamentals

Understanding Ayra AI

This section covers the fundamental concepts you need to understand to use Ayra AI effectively.

Click to unmute

Platform Architecture

Three-Tier System

Ayra is built on a three-tier architecture:

1

Platform Level (Ayra)

Super Admin

  • Manages infrastructure and servers
  • Provides core voice AI technology
  • Handles billing and subscriptions
  • Offers support to agencies
2

Agency/Workspace Level (You)

Your master account

  • Manages all clients
  • White-label customization
  • Sets pricing for clients
  • Creates and deploys voice agents
3

Client Level (Your Customers)

Individual client dashboards

  • Uses voice agents you created
  • Views their analytics only
  • Sees your branding (not Ayra's)

Multi-Tenancy

What is Multi-Tenancy?

What it means:

  • Each client gets isolated environment
  • Data is completely separated
  • Clients can't see each other
  • You manage all from one dashboard

Benefits:

  • Scale from 1 to 1,000 clients easily
  • No additional infrastructure needed
  • Centralized management
  • Consistent experience across clients

Voice Agent Concepts

How Voice AI Works

When a customer calls:

1
Audio Input:Customer speaks → Microphone captures audio
2
Speech-to-Text (STT):Audio → Text transcription
3
Natural Language Understanding (NLU):Text → Intent detection
4
AI Processing:GPT-4 generates appropriate response
5
Text-to-Speech (TTS):Text → Natural-sounding audio
6
Audio Output:Audio plays to customer → Sounds human
7
Loop Continues:Customer responds → Process repeats

Key Components

System Prompt

Instructions that define agent personality, what agent knows and can do, how agent should respond, when to escalate.

Knowledge Base

Facts the agent knows, FAQs and answers, product/service information, business policies.

Tools & Functions

Actions agent can take: schedule appointments, look up information, create tickets, update CRM.

Conversation Memory

Remembers what was said earlier, maintains context throughout call, references previous statements, tracks conversation state.

Voice Agent Lifecycle

CREATE
CONFIGURE
TEST
DEPLOY
MONITOR
OPTIMIZE

(Continuous improvement loop)

Conversation Management

Conversation States

Every conversation goes through states:

Initiated

Call connected, greeting played, waiting for customer input

Active

Customer and agent conversing, processing requests, taking actions

Waiting

Paused (customer thinking), on hold, executing tool/function

Escalating

Transfer to human requested, handing off to agent

Completed

Issue resolved, call ended naturally, goodbye said

Abandoned

Customer hung up mid-conversation, call dropped, system error

Turn-Taking

A conversation turn is one complete exchange between agent and customer.

  • Agent knows when to stop talking
  • Detects when customer wants to speak
  • Allows interruptions naturally
  • Prevents talking over customer

Context Preservation

Short-Term Context (Current Call)

  • Customer name
  • What they asked for
  • Preferences mentioned
  • Previous answers

Long-Term Context (Customer History)

  • Previous calls
  • Purchase history
  • Preferences
  • Account information

Escalation System

Why Escalate?

AI is powerful but not perfect. Escalate when:

AI doesn't understandComplex situationCustomer explicitly requestsEmotional situationPolicy exception neededHigh-value customerCompliance requirement

Escalation Types

Automatic Escalation

Triggers without human decision

Customer says "speak to a person"AI confidence < thresholdCall duration > limitTool failuresNegative sentiment detected

Manual Escalation

Agent decides to escalate

Recognizes complexityKnows limitationDetermines human needed

Smart Escalation

Context preserved during transfer

Full conversation historyCustomer informationWhat was triedWhy escalatingRecommended next steps

Escalation Best Practices

Set Clear Rules

  • Define when to escalate
  • Specify who to route to
  • Provide context to human
  • Track escalation reasons

Make It Smooth

  • Don't apologize excessively
  • Frame positively ("specialist can help better")
  • Provide estimated wait time
  • Offer callback option

Learn From Escalations

  • Review why each happened
  • Could AI have handled it?
  • Update knowledge base
  • Refine system prompts

Analytics & Metrics

Key Performance Indicators (KPIs)

Call Volume Metrics

  • Total calls
  • Calls per day/hour
  • Peak times
  • Average call duration

Quality Metrics

  • Resolution rate (% resolved without escalation)
  • Customer satisfaction score
  • Average sentiment
  • First-call resolution

Efficiency Metrics

  • Average handle time
  • Calls per agent (if hybrid)
  • Cost per call
  • ROI calculation

Technical Metrics

  • Uptime percentage
  • Average response time
  • Error rate
  • Tool success rate

Understanding Sentiment

+0.8 to +1.0Very Positive
+0.3 to +0.7Positive
-0.2 to +0.2Neutral
-0.7 to -0.3Negative
-1.0 to -0.8Very Negative
Identify at-risk customersMeasure customer satisfactionTrain agents on problem areasPrioritize improvements

Integration Concepts

API Integration

Application Programming Interface - allows systems to communicate.

Data Sync:Keep customer info synchronized between Ayra and CRM
Action Triggers:Call ends → Create ticket, Appointment scheduled → Add to calendar
Data Retrieval:Customer calls → Look up order in database → Show order status

Webhook vs Polling

Webhook (Push)

Event happens → System immediately notifies Ayra

Fast, efficient, real-time

Polling (Pull)

Ayra checks every X minutes → "Any updates?"

Slower, more resource-intensive

Prefer webhooks when available.

Security & Privacy

Data Security

Encryption

  • In transit: HTTPS/TLS
  • At rest: AES-256 encryption
  • End-to-end for sensitive data

Access Control

  • Role-based permissions
  • Principle of least privilege
  • Multi-factor authentication
  • IP whitelisting

Compliance

  • GDPR compliant
  • HIPAA available
  • SOC 2 Type II
  • Regular security audits

Data Privacy

Data Ownership

  • You own your data
  • Can export anytime
  • Can delete anytime
  • No data sharing without permission

Data Retention

  • Configure retention periods
  • Automatic deletion
  • Comply with regulations
  • Customer data requests honored

Privacy by Design

  • Minimize data collection
  • Purpose limitation
  • Data minimization
  • Transparency

Common Terms Glossary

Agent:AI voice assistant that handles calls
Conversation:Complete interaction from greeting to goodbye
Escalation:Transferring from AI to human
Intent:What customer wants to accomplish
Knowledge Base:Information the agent knows
System Prompt:Instructions defining agent behavior
Sentiment:Emotional tone of conversation
Transcription:Text version of spoken words
Turn:One exchange in conversation
Webhook:Automated notification between systems
White-Label:Removing Ayra branding, using yours
Workflow:Automated sequence of actions

Next Steps

Now that you understand the key concepts, you're ready to:

Build your first agentExplore FeaturesSet up IntegrationsRead Guides

Ready to transform your agency?

Start building with Ayra today. No credit card required.