Home

Permission design for customer-facing AI agents

Decide what your AI agents are allowed to do.

PermissionByDesign helps CX leaders, ecommerce operators, and service teams translate business policy into practical boundaries: what an AI agent may do, when it needs approval, and what evidence should be retained afterward.

Helpful AI, clear boundaries graphic showing a return-refund example on a phone beside a package.

What should your AI agent be allowed to do?

Learn how to define actions, set limits and decide when human approval is required.

A simple example

An order-support agent might be allowed to check delivery status, propose a refund, and draft a customer message. It might need a manager’s approval before issuing a refund above a set amount, overriding a return window, or changing the customer’s account record.

Example permission plan

  • May check order status.
  • May draft a refund recommendation.
  • Must request approval for exceptions.
  • Must retain the order record, policy rule, approval result, and customer message.

Start with the use cases that carry real business risk.

Refunds and returns

Set limits around amounts, windows, exceptions, and evidence.

View the refund example

Account changes

Decide which profile, billing, and access changes require stronger verification.

View the account example

Purchases

Separate recommendations, cart changes, payment actions, and exception approvals.

View the purchase example

Useful before it becomes complicated.

Use the guides, examples and planning tools to define an agent’s limits and discuss them with your implementation team.

Download the Permission Design Playbook

Use the free 12-page guide to turn one customer task into a permission plan with clear limits, approval rules and a tested path to resolution.

Build a permission plan

Work through one customer task, define what the agent may do and identify where approval, evidence or human recovery is required.