Make the policy visible before making the promise.
A merchant workspace for eligibility, evidence, customer responses and the decisions behind every return.
Built by Tern & LoomIllustrative operating scenarioThis describes a Tern & Loom product implementation using fictional business data. It is not a client engagement, testimonial, or measured customer result. The website walkthrough uses authored outputs; the operational product contains the agent orchestration and review workflow.
Returns that respect the customer. And the policy.
Returns teams switch between orders, delivery dates, policy documents and customer messages. Consistency breaks down when an item is outside the return window, condition evidence is missing, or an agent promises a refund before the merchant has approved it.
Order and fulfillment evidence in one review
Item-level eligibility and policy recommendations
Exception decisions with recorded rationale
Customer response drafts and receipt tracking
Follow the exception.
Compare three requests from the seeded merchant queue: an eligible return, a request outside the policy window, and a damaged-item request that still needs condition evidence. Each reaches a different next step, while the customer-facing response avoids promises the evidence cannot support.
Return analyst
Establish the record and the available evidence.
Policy analyst
Check domain context, sources and constraints.
Resolution planner
Prepare a proposed next step with uncertainty visible.
Commerce reviewer
Challenge the proposal before human review.
The product implements these roles with LangGraph orchestration, evidence snapshots and a review gate. This website presents a deterministic walkthrough of that structure; it does not run a model.
- Policy exceptions require a merchant decision and a recorded reason.
- Missing condition evidence routes back for clarification.
- A response draft is separate from refund authorization and settlement.
Define success before rollout.
Establish a baseline from the team’s existing process, run a controlled pilot, and compare like-for-like work. Track exceptions alongside speed.
- Time to the first accurate return decision
- Share of requests requiring evidence follow-up
- Policy exceptions and their documented reasons
No improvement percentages or ROI figures are claimed for this illustrative case.