The best first customer service automation is often unglamorous. Order status, delivery questions and membership details arrive repeatedly, have a known source and can be handed to a person when the case stops being routine.
A service flow with a safe exit
A real Australian example
Wesfarmers says it is implementing AI agents in contact centres for high volume routine enquiries such as order status, delivery queries and membership questions. The stated goal is better response times and consistency, alongside support for staff handling more complex conversations.
The public story describes a programme in progress. It does not publish a resolution rate or a customer satisfaction study, so it should be treated as an example of intended workflow design rather than proof of a guaranteed result.
Draw the boundary before you automate
List the information the system may use, the answer it can safely provide and the event that must trigger a handoff. A delivery estimate may be low risk if pulled from a live order system. A refund dispute, vulnerability disclosure or suspected fraud needs a more careful route.
Define how the system identifies itself, how a customer reaches a person and what conversation context follows the handoff. Test unusual wording and incomplete records, not only the clean demonstration prompt.
Measure the whole experience
Track whether the issue was resolved, whether the customer had to repeat information, how often the answer needed correction and whether the handoff arrived with useful context. A deflection rate alone can reward the system for making it hard to reach a human.
- Start with one queue and a human fallback.
- Keep answers linked to an approved source.
- Review complaints and failed handoffs every week during the pilot.
Want to explore how this could apply to your organisation?
Explore Private AI ↗