Trending Topics
Trending
Zendesk’s refund agents turn customer support into a fraud control question
It’s official. Zendesk’s specialised AI agents can do more than explain a retailer’s returns policy. They can act on it, handling refunds, exchanges and order changes across connected systems. That development puts customer service automation in direct contact with the company’s money.
Announced in mid September, the launch combined Industry Agents, initially targeting commerce, with Custom Agents built around individual businesses’ processes. Connections include Shopify, Narvar, Stripe and Riskified, all allowing agents to retrieve information and complete work without service teams switching between systems.
According to Zendesk, the agents can automate up to 80% of workflows. It is a bold claim, and extremely consequential if every retailer can eliminate four-fifths of its support workload. The company also reports that customers run more than one million Custom Agent executions within seven weeks of launch.
A refund needs more than a convincing explanation
The partnership with Riskified addresses a crucial part of the proposition: distinguishing legitimate requests from fraud and policy abuse. Zendesk says it is embedding Riskified’s risk intelligence into returns, refunds and exchanges.
For retailers, this makes authority an important deployment decision.
Which orders can an agent refund?
What value requires approval?
When should suspicious account activity halt the transaction?
Zendesk’s no-code Agent Builder begins with a quiz where you answer a few questions about how the team works and Zendesk then tells you what a Custom Agent could take off your plate. This approach makes it easy to define an agent’s job, permitted systems, available actions, and points requiring human approval.
Knowledge matters too.
As Stonly co-founder Alexis Fogel previously told us, “AI will only be as good as the content it relies on”. An outdated returns policy is a poor foundation for automated decisions.
Count the outcomes, including the mistakes
Zendesk’s Andrew Lawson has argued that lower ticket volume does not necessarily mean customers received good answers.
Refund automation deserves similarly careful measurement: legitimate requests completed, fraudulent claims rejected, customers wrongly refused and errors requiring human repair.
A retailer could start with low-value, straightforward returns, reserving exceptions for staff. Giving an agent access to the refund button need not mean handing it an unlimited spending allowance.
Related articles
About The Author
