Skip to main content
Pazit is PazGas Gas Order Agent, Israel’s gas company. She guides existing customers through ordering 1–2 gas tanks via a structured flow covering authentication, pricing, payment, and safety verification with a friendly and efficient approach.

View Agent

Open in Wonderful Platform

Overview

FieldValue
MarketIsrael - Energy
VerticalUtilities
CompanyPazGas (פזגז)
ModelGPT Realtime 28/28/2025
PersonaFriendly, Helpful, Professional
LanguageHebrew only
ChannelVoice

Skills

  • customer_authentication - Authenticates existing customers using customer number, phone, or address; retrieves account details and pricing.
  • credit_card_payment - Handles secure credit card collection and payment processing with installment options (payment tool implemented in backend – GO).
  • direct_debit - Processes orders for customers with existing direct debit authorization.
  • cash_and_check - Finalizes orders for customers paying cash or check upon delivery.

Key Tools

ToolPurpose
authenticate_customer()Verifies customer identity via customer number, phone, or address.
get_customer_details()Fetches account data, validates entitlement, checks for existing orders.
quick_authentication()Fast phone-based lookup for address retrieval.
check_allowed_payments()Determines available installment options.
create_debit_order()Finalizes order using direct debit.
create_check_or_cash_order()Finalizes order for cash/check payment.
check_customer_safety_check()Validates mandatory safety inspection post-payment.
send_sms()Sends SMS confirmations to customer.

Prompting Techniques

  • Structured conversation flow with explicit steps: The agent follows a clear 7-step process to ensure no steps are missed and the customer experience is consistent.
  • Pronunciation rules for voice clarity: Voice-specific formatting ensures numbers are spoken correctly and clearly understood by Hebrew-speaking customers.
Example Prompt Snippet:
Structured conversation flow (explicit steps):
1. Start and confirm order intent  
2. Authenticate customer  
3. Get customer details and confirm address  
4. Collect order details  
5. Price and approval  
6. Payment decision logic  
7. Order processing & completion  

Pronunciation rules for voice clarity:
- **Phone** (e.g. 054-1234567): "0, 5, 4... 1, 2, 3..."  
- **Credit card** (16 digits): Read in groups of 4  
- **Prices & quantities:** Speak exact numbers (not spelled out)

Lessons Learned

What worked
  • Modular skill architecture: Breaking the flow into distinct skills by payment method keeps logic clean, maintainable, and easier to debug.
  • Central validation: Implementing validation in get_customer_details() catches edge cases (existing orders, agency customers, entitlement) in one place rather than scattered throughout the flow.
  • Clear pronunciation rules: Voice-specific number formatting significantly improved customer comprehension and reduced errors in data collection.
Challenges
  • Backend payment integration: The implementation of the credit card payment method on the backend, caused by an issue with the “Pelecard” version, poses a difficulty for supporting agents when handling escalations related to payment flows.
  • Multiple authentication paths: Handling authentication via phone, customer number, or address needed clear fallback logic and retry limits to prevent customer frustration while maintaining security.
  • Safety check compliance: Ensuring customers understand and comply with mandatory safety inspection requirements while keeping the conversation flowing naturally.