Support · Returns and refunds
MixedFashion & D2COne queue for returns, refunds and complaints from WhatsApp, email and calls
A size exchange on WhatsApp, a damaged-item email in Zendesk and an angry call about a late refund are the same kind of work. These notes cover how to put them in one queue, draft the answer from your policy, and keep refunds with a person.
Valnora & Elric team5 min read
In a fashion brand’s support inbox, the same week brings a hundred "can I exchange for a size M" messages, a dozen damaged-item photos, refund chasers from people who returned something three weeks ago, and a few genuinely angry customers. They arrive on WhatsApp, by email and by phone, and each channel has its own tool and its own backlog.
These notes describe a setup in Agent Serve that puts them in one queue and leaves the money decisions with people.
Where do the cases come from?
Three channels into one support queue
Channels in
- 01WhatsApp triggerMessages to your WhatsApp Business number.
- 02Zendesk or IntercomTheir triggers start a run when a ticket or conversation is created.
- 03Agent Inbound & OutboundAnswers the support line and creates a support query; it can be exported or synced into the queue during setup.
Workflow · Triage and draft
- 04ClassifyReturn, exchange, refund status, damaged item, late delivery or complaint.
- 05Check the orderShopify order lookup: items, delivery date, payment method, tags.
- 06Returns policySearches the Knowledge Base for the window, condition rules and exceptions.
- 07Support queueOne row per case with channel, category, order, draft and status.
People · Decide
- 08Refund approvalA Human block pauses the run until a support lead approves or declines.
- 09Refund or cancelShopify cancel with refund to the original payment method, or the gateway API.
- 10Complaints dashboardVolume by reason, SKU and channel, and time to resolve.
Three channels feed one support queue in Tables:
- WhatsApp. Messages arrive through the WhatsApp webhook trigger of a workflow built in the workflow builder.
- Email. If you run support in Zendesk or Intercom, their triggers start the same workflow when a ticket or conversation is created. The workflow can reply, assign or close back in that tool.
- Calls. Agent Inbound & Outbound, a separate Valnora & Elric product available by invitation, answers the support line and creates a support query with a summary of the call. Those queries can be exported or synced into the same queue during setup.
How is each case classified and checked?
An agent block reads the message, or the call summary, and picks a category: return, exchange, refund status, damaged or wrong item, late delivery, or complaint. It pulls the order number from the text or matches the customer’s phone or email.
Then it checks facts instead of trusting the message. The Shopify block gets the order: items, delivery date, payment method and tags such as cod-confirmed. A return requested 40 days after delivery is treated differently from one requested on day three.
How does the policy get applied?
Upload your returns and refund policy to the Knowledge Base: return window, condition rules, non-returnable categories, exchange rules, and what counts as damaged. The workflow searches it for the rule that fits the case and quotes it in the case record, so the reviewer sees why the draft says what it says.
The agent drafts a reply in your tone. A rule flags cases that need extra care: a second complaint from the same customer in 30 days, words like "consumer court" or "fraud", or an order above a set value.
What the solution produces
Case drafted from a WhatsApp message (illustrative)
Record- Channel
- WhatsApp · 11:08
- Customer said
- "Kurta came with a torn seam on the sleeve. Photo attached. I want my money back."
- Category
- Damaged item · refund requested
- Order
- #50214 · delivered 2 days ago · prepaid · 1 item ₹1,890
- Policy
- Damaged items: refund or replacement within 7 days of delivery, photo required. Met.
- Draft reply
- Apology, confirms refund to original payment method once approved, reverse pickup booked.
- Needs
- Approval for ₹1,890 refund
Who approves refunds and cancellations?
A person, every time. When a case needs money back or an order cancelled, a Human block pauses the run and waits for a support lead to approve or decline. Approved actions then run:
- Shopify orders are cancelled with "Refund to original payment method", restock and customer notification, all from the Shopify block.
- Payments outside Shopify’s refund flow, for example a gateway such as Razorpay, are not a native integration. The workflow calls the gateway’s refund API with the API block, using a key stored as a workspace secret.
- COD returns need a bank or UPI refund your finance team makes; the workflow records the case and tells the customer when to expect it.
Set a limit above which a manager, not a lead, approves.
What does the complaints dashboard show?
A dashboard over the queue shows cases by reason, SKU, channel and courier, time to first response and to resolution, and refunds approved by week. The useful moment is when one SKU’s "damaged" count climbs: that is a packaging or supplier issue, not a support issue, and the dashboard makes it visible to the people who can fix it.
What protects customer data?
Controls and governance
Run logs
Every workflow run records each block’s inputs, outputs and errors, so a reviewer can trace how a result was reached.
PII detection
A Guardrails block detects personal data and blocks or masks it before text reaches the next step.
Deployed versions
Workflows run from a deployed version, so a change is reviewed before it reaches production runs.
Roles and permission groups
Admin, write and read roles per workspace, plus permission groups that limit which models, blocks and integrations a team can use.
Row-level access
Row-level policies narrow what each person sees in tables, dashboards, exports and scheduled reports.
Audit trail
Proposals, approvals, declines and executed actions are written to the workspace audit log, which admins can export.
More on how Agent Serve handles access, data and audit on the security page.
Every run is traced in Logs. PII redaction masks phone numbers, addresses and emails in logs, workflow input and block outputs. Workflows run from a deployed version, so a policy wording change is reviewed first.
What should you start with?
What you need
- Your returns, exchange and refund policy as documents in the Knowledge Base
- WhatsApp Business account, plus Zendesk or Intercom access
- Shopify app token with order read and cancel permissions
- For prepaid refunds outside Shopify, your payment gateway API key stored as a workspace secret
- Named support leads who approve refunds, and a limit above which a manager approves
- Agent Inbound & Outbound access if calls are in scope (by invitation)
Start with one channel and two categories, usually WhatsApp exchanges and refund status. Read a week of drafts with the team before any reply goes out without a person.
Questions teams ask
Can Agent Serve refund through Razorpay?
Razorpay and similar gateways are not native integrations. You can call the gateway’s refund API with the API block, with the key stored as a workspace secret, after a person approves. For Shopify orders, the cancel operation can refund to the original payment method.
Do replies go out without a person seeing them?
You choose. Most teams start with every reply drafted and sent by a person, then let low-risk categories such as "where is my refund" go out automatically once the drafts are reliably right.
What does Agent Inbound & Outbound do on a support call?
It answers the call, handles what it can, and creates a support query with a summary for anything that needs your team. It is a separate Valnora & Elric product, available by invitation.