Agent responsibility
Check order & policy
Assemble a case summary, check the required facts, and propose either a policy-aligned response or an escalation.
Refund recommendation
Customer operations / Customer service / Solution blueprint
Customer refund reviews
Prepare a refund decision with the order history and policy in view.
Business problem & expected outcomes
The challenge behind
customer refund reviews.
Support teams move between orders, conversations, and policy documents. Approving a refund without all three can create inconsistent treatment.
What this workflow is designed to deliver
Refund recommendation
Assemble a case summary, check the required facts, and propose either a policy-aligned response or an escalation.
How to measure the improvementEvidence coverage; escalation accuracy; refunds executed only with the required authorization.
Start with the right information.
Order details, payment references, support history, and the current refund policy.
Interactive workflow diagram
From refund request
to a clear next step.
Assemble a case summary, check the required facts, and propose either a policy-aligned response or an escalation.
Customer operationsSelect a stage to explore
Context
Refund request
Order details, payment references, support history, and the current refund policy.
Prepared outputRefund recommendation
Designed around the work
A focused agent. The right capabilities.
Give the agent the context and tools this process needs, with an explicit output and an accountable owner.
Connected tools & context
Bind only the operations this workflow needs. Keep credentials in ART and choose the permitted tools explicitly.
Connectors & MCP
Scoped read accessRead the source records needed for customer refund reviews. Pass record identifiers and retrieved evidence to the agent.
Knowledge Base
Project & environment scopeRetrieve approved guidance and attach source references to the prepared answer.
The decision boundary
Validate the amount and policy conditions. Require the configured approver before a refund tool executes.
ART product mapping
Components used.
Chosen for this case.
Agent Lab prepares the work described below. These are the capabilities that support this workflow.
Agent Lab
Connectors & MCP
Retrieve the required records through scoped tools; begin with read-only access.
Agent Lab
Knowledge Base
Retrieve approved guidance and attach source references to the prepared answer.
Agent Lab
Execution governance
Validate required facts and evaluate explicit rules before an action can proceed.
Agent Lab
Human approvals
Pause the proposed action for the designated owner to review its evidence.
The implementation, step by step
From source context
to a reviewed result.
- 01
Refund request
Order details, payment references, support history, and the current refund policy.
Agree the trigger or manual request, record identifiers, and the process owner. - 02
Check order & policy
Assemble a case summary, check the required facts, and propose either a policy-aligned response or an escalation.
Define instructions, permitted tools, and the expected output. Keep missing information explicit. - 03
Authorized review
Validate the amount and policy conditions. Require the configured approver before a refund tool executes.
Configure the selected policy or approval capabilities before enabling write actions. - 04
Refund recommendation
Evidence coverage; escalation accuracy; refunds executed only with the required authorization.
Compare outputs with reviewed examples and test incomplete, conflicting, and unauthorized inputs.
Before you build
What this implementation needs.
Connectors & MCP
Source-system APIs or MCP tools, scoped credentials, and an integration owner.
Knowledge Base
An approved document set, content owners, and retrieval access boundaries.
Execution governance
Registered actions, required facts, and explicit allow, block, or approval rules.
Human approvals
Named reviewer roles, escalation ownership, and timeout and rejection behavior.
Also prepare an ART project and environment, the model configuration appropriate to your agent, an accountable process owner, and approved sample data.
Authority stays with your team
Define the boundary
before enabling the action.
Validate the amount and policy conditions. Require the configured approver before a refund tool executes.
Explore execution governanceConnect deliberately.
Map the source material and responsible owners first. Where the selected components require system access, configure scoped APIs or tools. A document-based pilot can start with approved material already available in the workspace.
Make each handoff inspectable.
Keep source references with the prepared output and review representative results before expanding scope. If the implementation includes executable actions, define their required facts and authority first. Apply retention to any stored records, files, or memory used by the case.
Build evidence from your own work
Start small.
Agree what good looks like.
What to measure for this case
Evidence coverage; escalation accuracy; refunds executed only with the required authorization.
Establish a baseline
Choose a bounded sample of completed cases, record the current process, and agree review criteria with the process owner.
Run in review mode
Use approved sample data and read-only connections. Compare prepared outputs with reviewer expectations, including incomplete and conflicting inputs.
Expand with approval
Enable a limited set of actions only after the team accepts the results. Check approvals, failures, and handoffs before increasing scope.
Make it your workflow
Try this approach
in Agent Lab.
Start with refund request. Connect your approved tools, configure the review boundary, and adapt the output to your team's process.
Try Workflow in Agent Lab
Open your ART workspace to configure this use case.
Discuss the setup with usIndustry & department library
Find your next use case.
3 use cases in Customer operations
Customer operationsWorkflow design
- ContextCase history
- Agent LabSummarize prior work
- Your teamSpecialist handoff
Prepared outputContext-rich case brief
Solution blueprint
Support case handoffs
Give the next specialist the context to continue the conversation.
Used in this case
Explore the implementation - Knowledge Base
- Agent memory
- Agent Skills
Customer operationsWorkflow design
- ContextService question
- Agent LabRetrieve relevant knowledge
- Your teamCheck source coverage
Prepared outputGrounded service answer
Solution blueprint
Service knowledge answers
Prepare grounded answers from approved support guidance.
Used in this case
Explore the implementation - Knowledge Base
- Agent Skills
Customer operationsWorkflow design
- ContextActive service case
- Agent LabShare current context
- Your teamSpecialist collaboration
Prepared outputConnected case handoff
Solution blueprint
Live service collaboration
Keep the service team aligned while a complex case is being resolved.
Used in this case
Explore the implementation - Secured channels
- Agent memory
- Agent Skills
Customer operations / Let’s explore the fit
Discuss your
industry use case.
Tell us where work slows down. We’ll discuss how ART could fit your process, your systems, and the decisions your team needs to own.
