Clara is Shipsy’s autonomous customer-experience agent. She does not deflect tickets — she resolves them. Clara ingests the consignee’s query, pulls the live shipment state from Shipsy’s tracking graph, decides whether the issue is a real exception or a communication gap, and either replies conversationally with the answer or triggers the corrective operational action (reschedule, reroute, address correction, refund). Across deployments, Clara autonomously closes 60–80% of CX tickets that used to sit in a human queue.
Why we built this
Logistics CX is structurally broken in three ways. First, volume scales linearly with shipment count — every incremental million parcels brings incremental tens of thousands of “where’s my package” tickets. Second, the answer usually isn’t written anywhere — the CX agent has to triangulate the order system, the tracking page, the driver app, and sometimes the carrier network. Third, the resolution often requires an operational action (reschedule a slot, reroute to a neighbor, push an address fix) that a traditional chatbot simply cannot execute.
We built Clara to collapse that triangulation and execution into a single autonomous loop — reasoning grounded in the actual shipment state, with the permission to take action inside Shipsy’s operational stack.
How it works
Clara runs as a multi-step reasoning agent with four explicit components:
Component 1 — Intent classifier. Incoming messages (WhatsApp, email, chat widget, voice-to-text) are classified into the dominant intent taxonomy: status inquiry, reschedule request, address correction, NDR rescue, refund/return, complaint. Clara uses few-shot prompts tuned per language and locale — she ships with production coverage for 20+ languages.
Component 2 — Shipment-state retrieval. Clara pulls the live state of the shipment from Shipsy’s tracking graph: current leg, predicted ETA, driver assignment, prior attempt history, NDR reason codes if any, and the consignee’s own history. This is the step that a generic chatbot cannot do — Clara has first-class access to the operational substrate, not a scraped tracking page.
Component 3 — Decision layer. For each intent + state combination, Clara picks from a decision tree of responses and actions. A status inquiry with an on-track ETA returns a conversational answer with the predictive ETA. A reschedule request checks Astra’s slot availability and books a new slot. An NDR “incomplete address” triggers the address intelligence pipeline and asks the consignee one confirming question. A high-severity complaint is escalated to a human CX lead with a pre-drafted response and full context.
Component 4 — Action execution. Clara writes back into Shipsy — updating consignee-provided address fields, triggering reschedule, creating refund tickets, tagging shipments for retry. Every action is audit-logged with the Clara session ID, the reasoning trace, and the human approval step if configured.
Clara also runs proactive outreach loops. She watches the live ETA and exception feeds; when she detects a slip or an NDR risk, she reaches out to the consignee before they reach out — with a personalized message (not a template blast) containing the specific reason and next-action options.
Here’s the sequence at a glance:
Early results
Enterprises deploying Clara typically report, within 90 days:
- 60–80% of CX tickets resolved without a human — the rest escalate with full context, so human agents handle only high-complexity cases.
- First-response time under 10 seconds for status, reschedule, and address correction.
- NDR rescue rates up 15–25 percentage points — Clara’s proactive loop converts soft NDRs into successful next-attempt deliveries.
- Cost per ticket falls 50–70% — and CSAT typically rises, because consignees are getting real answers, not deflection.
A leading Western European parcel operator with 50%+ national market share uses Clara across 8M+ shipments to handle the long tail of CX load without expanding the human team.
What’s next
Three expansions: voice-native Clara for inbound calls with under-2-second response latency, multi-shipment context for customers with repeat or subscription deliveries, and deep merchant-policy integration — Clara learning the specific refund and retry policies of each merchant she serves on a 3PL or marketplace network.