Amazon Connect Cases + Knowledge Base for Tier-1 Deflection

Not every customer issue needs a live agent. Password resets, order status checks, and policy FAQs are tier-1 volume that inflates staffing costs without adding customer value. Amazon Connect Cases combined with Amazon Q in Connect (formerly Wisdom) creates a self-service and async resolution layer that deflects 25–35% of inbound volume before it reaches a queue. This guide covers the architecture, knowledge base design, and case management workflows we deploy for AWS-centric contact centres.

The tier-1 deflection stack

Amazon Connect's deflection capabilities operate at three layers, each handling a different resolution path:

• Amazon Q in Connect (Wisdom) — AI-powered knowledge retrieval for agents and self-service. Answers customer questions from indexed content in real time
• Amazon Connect Cases — structured case management for issues that can't be resolved instantly. Tracks status, assignments, and SLA without leaving Connect
• Lex V2 + Lambda — conversational AI that handles structured lookups (order status, account balance) and creates cases when resolution requires human follow-up

The three components share a common knowledge base and customer context layer. A customer who starts in chat, gets a partial answer from Wisdom, and escalates to an agent should never repeat their issue — the case record carries the full transcript.

Step 1: Build the knowledge base

Amazon Q in Connect indexes content from S3, Salesforce, ServiceNow, Zendesk, or Microsoft SharePoint. For tier-1 deflection, curate content deliberately:

1. Export your top 50 support articles by page view and ticket deflection rate
2. Rewrite each article as a concise Q&A pair (question, answer, related articles)
3. Tag with intent categories that map to your Lex bot intents
4. Upload to an S3 bucket with the Connect knowledge base connector

Content quality matters more than content volume. Twenty well-structured articles outperform two hundred outdated PDFs. Schedule a monthly content review cycle — stale knowledge erodes deflection rates within a quarter.

Step 2: Configure Connect Cases

Connect Cases provides native case management without a third-party CRM for async workflows. Define your case fields and templates:

# Connect Cases field template (CloudFormation excerpt)
CaseField:
  Type: AWS::Connect::CaseField
  Properties:
    DomainId: !Ref CasesDomain
    Name: issue_category
    Type: SingleSelect
    Options:
      - billing
      - technical_support
      - account_management
      - product_enquiry

CaseTemplate:
  Type: AWS::Connect::CaseTemplate
  Properties:
    Name: tier1_async_resolution
    Fields:
      - issue_category
      - customer_id
      - priority
      - resolution_notes

Configure case assignment rules: billing cases route to the billing team queue, technical cases to tier-2. Set SLA targets per case type — 4 hours for billing, 24 hours for technical. Connect Cases tracks SLA breach and sends notifications via SNS or EventBridge.

Step 3: Wire the contact flow for deflection

The contact flow orchestrates the deflection logic. A typical tier-1 flow:

1. Customer initiates chat or calls inbound
2. Lex bot handles intent recognition and structured lookups
3. On FAQ intent → invoke Wisdom API for knowledge retrieval → return answer
4. On resolution → close contact with deflection flag
5. On unresolved → create Connect Case with transcript, notify customer of case ID and expected response time
6. On explicit escalation request → transfer to agent queue with case pre-populated

# Lambda: create case on unresolved interaction
import boto3

def create_async_case(contact_id, customer_id, transcript, category):
    connect_cases = boto3.client('connectcases')
    response = connect_cases.create_case(
        domainId=CASES_DOMAIN_ID,
        templateId=TIER1_TEMPLATE_ID,
        fields=[
            {'id': 'issue_category', 'value': {'stringValue': category}},
            {'id': 'customer_id', 'value': {'stringValue': customer_id}},
        ]
    )
    return {
        'caseId': response['caseId'],
        'message': f'Case {response["caseId"]} created. '
                   f'We will respond within 4 hours.'
    }

Step 4: Agent assist with Wisdom

For contacts that do reach an agent, Wisdom provides real-time knowledge recommendations on the agent desktop. As the agent types or the customer speaks, Wisdom surfaces relevant articles ranked by relevance score.

Configure Wisdom with:

• Agent-facing recommendations — top 3 articles displayed in the CCP (Contact Control Panel) sidebar
• Auto-suggest — Wisdom proactively surfaces content based on conversation context without agent search
• Feedback loop — agents mark recommendations as helpful or not; feed back into content prioritisation

Agent assist reduces average handle time by 15–25% for knowledge-intensive contacts. Combined with tier-1 deflection, the total volume reduction often justifies the Amazon Q licensing cost within the first quarter.

Measuring deflection effectiveness

Build a Connect analytics dashboard tracking:

• Self-service deflection rate — contacts resolved without agent or case
• Async case deflection rate — contacts resolved via case without live agent
• Knowledge article hit rate — which articles are surfaced and acted upon
• Case SLA compliance — % of async cases resolved within target
• Re-contact rate — customers who deflect but call back within 24 hours

A healthy deflection programme shows declining re-contact rates over time — customers trust the self-service path because it actually resolves their issue.

What's next

Connect Cases and Wisdom are the async resolution layer beneath your conversational AI. Once tier-1 deflection is stable, extend with Amazon Q agent assist for complex tier-2 contacts and integrate Cases with your CRM for closed-loop reporting. See our AI Support Agent Quickstart or contact us for a deflection assessment.