How It Works

From incoming ticket to closed case

Replylume runs four steps on every ticket: intake, classification, data fetch, and resolution action. Nothing black-box. Every step is logged and every escalation is intentional.

01

Intake from your existing help desk

Replylume connects to your help desk through a read-write API or webhook. New tickets are pulled as they arrive, with full metadata: subject, body, sender, ticket ID, priority, and any existing tags. No email forwarding, no separate inbox.

Setup takes under 15 minutes for supported platforms. For custom ticketing systems, Replylume accepts a standard webhook payload format documented in the API guide.

Incoming ticket payload
POST /replylume/intake
# Webhook from Zendesk on new ticket

{
  "ticket_id": "ZD-48821",
  "subject": "Order didn't arrive, want refund",
  "requester_email": "[email protected]",
  "created_at": "2026-06-11T23:41:17Z",
  "priority": "normal",
  "tags": ["billing", "refund"],
  "body": "Hi, I ordered on Jun 4..."
}
Classification result
# Replylume classifies intent + tier

intent:   refund_request
tier:     1  # eligible for automation
signals:  ["billing tag", "refund keyword x3",
          "no account flag", "recent order"]
confidence: 0.94

# Low-confidence or multi-intent: escalate
escalation_threshold: 0.72

02

Classification and tier routing

Replylume reads the ticket content and classifies the intent: refund request, order status, billing question, account access, or other. Tickets that match a supported Tier-1 pattern with high confidence are routed for automation. Low confidence or ambiguous tickets are flagged for a human immediately.

You control which intent categories are in scope and can adjust the confidence threshold per category from the dashboard.

03

Live data fetch from billing and CRM

Before drafting any response or taking action, Replylume fetches the live record. For a refund request: it queries your billing API for the relevant order, confirms purchase date, amount, and refund window, and pulls any account notes from your CRM.

All API calls are read-only at this step. Nothing is modified until the action step that follows.

Context Fetch 2 sources
Stripe order #RL-48821 FETCHED
HubSpot account record FETCHED
Refund eligibility check ELIGIBLE
Account flags / blocks NONE
Resolution actions taken
# 1. Issue refund via Stripe
PATCH /v1/refunds
charge_id: "ch_3Qx7rMnP"
amount:    8900  # cents
reason:    "customer_request"

# 2. Close ticket in Zendesk
PUT /tickets/ZD-48821
status: "solved"

# 3. Write to HubSpot activity log
POST /engagements/v1/engagements
# elapsed: 8.3s total

04

Resolution action and audit trail

Once the data confirms eligibility and no blocking conditions exist, Replylume executes the action sequence: issue the refund, send a customer confirmation, close the ticket, and write the structured CRM activity entry. All three write operations are logged in the audit trail.

If any step fails (API error, rate limit, unexpected response), the sequence halts and the ticket is immediately escalated to a human agent with full context attached. No silent failures.