Pre-sale questions
Sizing, availability, shipping windows, return policy. Sonar reads intent, AI answers fast, cites the source, escalates if confidence drops below your threshold.
Solution · Ecommerce
Ecommerce support that doesn't drop the cart
OrcaLinq answers pre-sale and order-status questions across web, WhatsApp, Instagram, Messenger, and Shopify — and pages a person the moment a refund, complaint, or VIP order needs one.
Why this fits
Pre-sale, post-sale, and order-status — three different routing policies
Ecommerce conversations cluster around three intents. The semantic PBX treats each one differently: routine pre-sale and order status get AI answers from approved knowledge; refunds, complaints, and VIP orders go to humans with full context.
Order status
Live tool calls into Shopify (or your OMS) at reply time. Tracking, fulfillment, ETA. No memorized inventory, no stale numbers.
Refunds and complaints
Forced handoff regardless of confidence. The agent gets the order context, transcript, AI suggested response, and a one-tap accept-or-edit interface.
VIP and high-cart customers
Tagged routing in Matriarch. High-LTV customers reach a human first. The AI doesn't gate them, it pages the right agent immediately.
Conversation example
An order-status reply that closes itself
A typical D2C post-purchase conversation. AI answers, fetches live data, and only asks for a human when the policy demands it.
Customer (WhatsApp, 11:14 PM): "Where's order #SF-48201? Said 3 days, it's been 5."
AI (after live tool call to Shopify): "Hi Aishwarya — your order shipped on the 3rd, picked up by Delhivery, and is currently at the Mumbai sorting facility. The latest scan was 6 hours ago. ETA on the tracker is now Saturday afternoon. Tracking link: …"
Customer: "I won't be home Saturday. Can you reroute to my office?"
AI: "Reroute requests need our team to confirm with Delhivery. Paging Riya now — she'll get the new address and update you within the next hour. Could you share the office address while we wait?"
Customer: "Sure, [office address]"
Riya (WhatsApp, 11:16 PM): "Hi Aishwarya, Riya here. I have your office address — putting through the reroute now and will share the updated tracker as soon as Delhivery confirms."
The conversation lives in Wake. No silent failure, no fake ETA, no "let me check and get back to you" that never comes.
Recommended setup
What a typical D2C tenant configures in week one
- Website widget on every PDP and the cart page, opening pre-filled with the SKU in context.
- WhatsApp Cloud API with your verified number for post-purchase, tracking, and follow-up.
- Shopify connector for live order, fulfillment, cart, and customer data — fetched at reply time.
- Approved knowledge: shipping, returns, exchange, COD, sizing, materials, care, warranty.
- Matriarch routing: pre-sale → support; refund/complaint → human; VIP/high-cart → senior agent; out-of-stock → marketing.
- Wake retention: full transcript, reasoning, source citations, and confidence scores logged on every conversation.
Features used
What's doing the work
Sonar
Reads intent, urgency, and sentiment on every inbound message. Refund and complaint language are flagged before any reply is composed.
Matriarch
Routes by intent, customer tag, cart value, and channel. Out-of-hours and unavailable-agent flows are part of the policy, not a hack.
Knowledge base AI
Sized and policy documents are the source. AI cites them in every answer; gaps surface in the learning queue for admin approval.
AI human handoff
Refund, complaint, and VIP triggers force handoff. Agent sees order context, transcript, suggested reply, and one-tap accept.
Agent-native endpoints
Replies go from agent's WhatsApp, Telegram, or PWA. No new dashboard for the team to learn.
Wake
Every routing decision, source consulted, confidence score, and handoff reason is logged. Auditable from day one.
Metrics to improve
What changes after the semantic PBX is live
- First-reply time — typically drops from minutes-to-hours to under 30 seconds for AI-handled questions.
- Order-status ticket volume — falls sharply when AI fetches live tracking instead of asking the team to check.
- Refund handling time — agents arrive with order, transcript, and a suggested reply; resolution time compresses.
- Cart-abandonment recovery — honest, policy-controlled outreach reaches the customer on a channel they read.
- Support-team headcount needed per 1,000 daily conversations — drops without losing answer quality.
- Wrong-answer incidents — go to zero on tenant facts, because AI never answers from general model memory for store data.
Ecommerce FAQ
Frequently asked questions
Does it integrate with Shopify?
Yes. Shopify event webhooks (order created, fulfillment update, cart abandoned, customer created) feed directly into Sonar. Live order data is fetched via tools at reply time, not memorized — so tracking, ETA, and order status are always current.
What about WhatsApp for ecommerce?
WhatsApp Cloud API is the primary post-purchase channel for many ecommerce teams. Order updates, tracking, and follow-up route through the same semantic PBX with respected templates and pricing tiers. BYOK — your number, your token, your sender identity.
Can the AI handle returns and refunds?
It can collect details and propose a refund, but most teams hold the refund initiation for human approval. That's a tenant-configured policy in Matriarch, not a platform default. Refund language always forces handoff regardless of confidence.
What about cart-abandoned recovery?
Honest, tagged outreach is supported on channels that allow it. Aggressive spam is not — every outbound is policy-driven, customer-context aware, and respects WhatsApp's template tiers.
What about COD verification?
COD verification flows are a common Indian-D2C pattern. The AI can confirm the order, address, and intent, page a human if anything looks off, and write the verification result back to your OMS via tool call.
Can it answer product-fit questions?
Yes — sizing, compatibility, ingredient, and care questions are answered from your approved policy and product documents. If the question is outside the approved knowledge, the AI hands off rather than guessing.
What about product reviews and social proof?
If your store exposes reviews via API, the AI can surface average rating and recent quotes when relevant. It will never invent a review or claim a rating it cannot fetch.
Related pages
Keep exploring
See ecommerce routing on your real Shopify catalog.
Bring a sandbox store. We'll wire the webhook, fire a real order event, and route a customer message live in your tenant.