All use cases

Prevent / Care

Returns and refunds, judge-gated end-to-end

The return is handled start to finish in one conversation — and the refund is confirmed only after independent judges verify it against your policy.

Next best thingClose the loopBehavioral triggerSystem triggerChatIn-appEmailE-commerce & RetailSubscription

What it does

Returns are the highest-volume service flow in commerce and the one customers remember: dig out an order number, fill in a form, then wait in the dark until "where is my refund" becomes a second ticket. The agent executes the whole flow — return, refund, or exchange — directly against the commerce APIs. It matches the customer to the order without asking for an order number, checks eligibility against the actual returns policy, and only renders the refund confirmation after pre-send evaluation passes, with a supervision badge stating the decision was verified against your refund policy. From there it keeps the customer updated as the return progresses, so the follow-up ticket never happens.

How it works

  1. 1
    Trigger. the customer asks to return, exchange, or get a refund in chat or in-app, or a return event arrives from the commerce platform mid-flow.
  2. 2
    Decision. the agent matches identity from the account and conversation context — no order-number hunt — then checks the return window, item rules, and refund eligibility against the policy and the order history.
  3. 3
    Action. the refund confirmation dialog renders only after the pre-send evaluation passes. Under the default stand-in policy, three independent judges verify the decision and refunds under £50 auto-approve; anything above the threshold, or anything ambiguous, goes to a human for one-click approval.
  4. 4
    Follow-through. as the return progresses — label issued, package received, refund settled — the agent proactively tells the customer at each step, so the status question never gets typed. Every action and every judge verdict writes back as an audit trail, and unresolved cases escalate with the full transcript.
Example
AgentFound it — the running shoes from your most recent order, still inside the return window. I've emailed you the return label. Once the courier scan confirms drop-off, the full amount goes back to your original card. This refund was verified against the returns policy before I confirmed it.
CustomerGreat. How long until I see the money?
AgentA few business days after the drop-off scan — and you won't need to chase it. I'll message you the moment the refund is issued.

Configuration

How the agent is wired for this use case.

Triggerthe customer asks to return, exchange, or get a refund in chat or in-app, or a return event arrives from the commerce platform mid-flow
Tools & actions
  • CRM / order history · match the customer to the order from account and conversation context, no order-number hunt
  • Commerce platform / OMS returns API · check the return window and item rules, issue the return label, execute the return or exchange
  • Payment provider · execute the refund to the original payment method on file
  • Messaging channel · render the verified refund confirmation and push proactive status updates at each step
  • Ticketing system · write the audit trail and open escalations
Autonomyunder the default stand-in policy three independent judges verify each decision and refunds under £50 auto-approve; anything above the threshold or anything ambiguous goes to a human for one-click approval, with the written refund policy — never the model's judgment alone — as the single source of truth
Channelschat · in-app · email
Escalationunresolved cases escalate to a human with the full transcript and judge audit trail

What you need

The inputs this use case runs on. Your channels stay yours; the agent supplies the judgment.

Signals

return/refund/exchange intent in conversation, commerce return events (label created, return received, refund issued)

Data

order history for identity matching, the returns and refund policy expressed as checkable rules, payment method on file

Guardrails

pre-send judge evaluation on every refund confirmation; a value threshold above which a human approves; the written policy as the single source of truth, never the model's judgment alone; full audit trail of what each judge verified

Metrics it moves

  • time-to-resolutiondown: a return executed in one conversation instead of a form, a queue, and days of silence
  • contact-ratedown: proactive status updates prevent the "where is my refund" repeat contact
  • csatup: no order-number hunt on the way in, no waiting in the dark on the way out

See it on your own customer journey

Bring one drop-off, one churn cliff, or one silent segment. We will show you what a proactive agent with memory and judgment does with it.

Book a demo