Customer Support Agent

​

Overview

​

What the Agent Needs to Accomplish

• Retrieve customer information from multiple sources (CRM, order system, support history)
• Access order details and transaction history
• Review previous support interactions and resolutions
• Query product catalog and documentation
• Provide accurate, context-aware responses
• Track and update support ticket status

​

How Pylar Helps

• Unified Customer View: Creating a single view combining customer data from CRM, orders, and support systems
• Governed Access: Ensuring the agent only accesses appropriate customer data
• Real-time Queries: Allowing natural language queries that translate to efficient SQL
• Multi-Source Integration: Joining data across Snowflake (orders), HubSpot (CRM), and Zendesk (support tickets)
• No API Hassles: Eliminating the need to manage multiple API integrations

​

Without Pylar vs With Pylar

​

Without Pylar

• ❌ Multiple API integrations required (HubSpot API, Order API, Zendesk API)
• ❌ Complex authentication for each service
• ❌ Rate limiting and API quota management
• ❌ Manual data joining across sources
• ❌ No unified customer view
• ❌ Security concerns with direct database access
• ❌ Difficult to maintain and update

• 3-5 different API integrations
• Custom code for data joining
• Complex error handling
• Rate limit management
• Authentication token management
• ~2-3 weeks development time

​

With Pylar

• ✅ Single MCP endpoint for all data sources
• ✅ Unified customer 360 view
• ✅ Governed SQL views ensure data security
• ✅ No API rate limits or authentication complexity
• ✅ Easy to update views and tools
• ✅ Real-time querying across all sources
• ✅ Built-in observability with Evals

• Connect 3 data sources (30 minutes)
• Create unified customer view (1 hour)
• Build MCP tools with AI (30 minutes)
• Connect to agent builder (15 minutes)
• Total: ~2 hours

​

Step-by-Step Implementation

​

Step 1: Connect Data Sources

1. Connect HubSpot (Customer data, deals, interactions)
2. Connect Snowflake (Order history, transaction data)
3. Connect Zendesk (Support tickets, previous interactions)

​

Step 2: Create Customer 360 View

CREATE VIEW customer_360 AS
SELECT 
  h.customer_id,
  h.customer_name,
  h.email,
  h.phone,
  h.company_name,
  h.deal_stage,
  h.total_deal_value,
  s.order_count,
  s.total_spent,
  s.last_order_date,
  s.avg_order_value,
  z.ticket_count,
  z.last_ticket_date,
  z.resolved_tickets,
  z.open_tickets
FROM hubspot.customers h
LEFT JOIN snowflake.order_summary s ON h.email = s.customer_email
LEFT JOIN zendesk.ticket_summary z ON h.email = z.customer_email;

​

Step 3: Create Support-Specific Views

CREATE VIEW customer_order_history AS
SELECT 
  c.customer_id,
  c.customer_name,
  o.order_id,
  o.order_date,
  o.order_status,
  o.total_amount,
  o.items,
  o.shipping_address
FROM hubspot.customers c
JOIN snowflake.orders o ON c.email = o.customer_email
ORDER BY o.order_date DESC;

CREATE VIEW customer_support_history AS
SELECT 
  c.customer_id,
  c.customer_name,
  t.ticket_id,
  t.subject,
  t.status,
  t.created_date,
  t.resolved_date,
  t.agent_name,
  t.resolution
FROM hubspot.customers c
JOIN zendesk.tickets t ON c.email = t.customer_email
ORDER BY t.created_date DESC;

​

Step 4: Create MCP Tools with AI

• Prompt: “Create a tool to fetch complete customer information including contact details, order history, and support tickets”
• AI generates: get_customer_info(customer_email: string)

• Prompt: “Create a tool to retrieve order details by order ID or customer email”
• AI generates: get_order_details(order_id: string, customer_email: string)

• Prompt: “Create a tool to fetch customer support ticket history”
• AI generates: get_support_history(customer_email: string, limit: number)

• Prompt: “Create a tool to search product catalog by name or SKU”
• AI generates: search_products(query: string, category: string)

​

Step 5: Test Tools

• Verify customer data retrieval
• Test order lookup
• Confirm support history access
• Validate product search

​

Step 6: Publish and Connect

1. Publish Tools: Click “Publish” in Pylar
2. Copy Credentials: Get MCP HTTP Stream URL and Authorization Token
3. Connect to Agent Builder: Add to your agent builder (Claude Desktop, Cursor, LangGraph, etc.)

​

Step 7: Monitor with Evals

• Monitor agent performance
• Track common queries
• Identify errors
• Optimize tool usage

​

Example Agent Interactions

​

Scenario 1: Customer Order Inquiry

1. Calls get_customer_info("john@example.com")
2. Calls get_order_details("12345", "john@example.com")
3. Responds: “Order #12345 for John Smith is currently ‘Shipped’ and expected to arrive on March 15th. The order contains 2 items totaling $149.99.”

​

Scenario 2: Support Ticket Creation

1. Calls get_customer_info(user_email)
2. Calls get_order_details(null, user_email) - gets recent orders
3. Calls get_support_history(user_email, 5) - checks previous issues
4. Responds: “I see you purchased a laptop on March 10th. I also notice you had a previous ticket about setup. How can I help you today?”

​

Outcomes

​

Efficiency Gains

• Response Time: 80% faster responses with instant data access
• First Contact Resolution: 60% increase in resolving issues on first contact
• Agent Productivity: 3x more tickets handled per agent
• Customer Satisfaction: 40% improvement in CSAT scores

​

Data Access

• Unified View: Single source of truth for all customer data
• Real-time Updates: Always current data from all sources
• No API Limits: No rate limiting or quota concerns
• Secure Access: Governed views ensure data security

​

Maintenance

• Easy Updates: Modify views without redeploying agents
• Observability: Full visibility into agent queries and performance
• Iteration: Quickly refine tools based on Evals insights

​

Best Practices

1. Data Governance: Create views that only expose necessary customer data
2. Error Handling: Use Evals to identify and fix common query errors
3. Performance: Optimize views for common query patterns
4. Security: Ensure views filter sensitive data appropriately
5. Monitoring: Regularly review Evals dashboard for insights

​

Next Steps

• Sales Assistant Example - Build a sales pipeline agent
• Customer Churn Predictor Example - Predict and prevent churn
• Marketing Campaign Optimizer Example - Optimize marketing campaigns