Capability / Service agents
Customer-service agents that check before they answer
AI that drafts or sends replies from your real order data and policies, with refunds, complaints and anything unusual handed to a person.
The problem with a generic chatbot
Most online stores have seen the chatbot that cheerfully tells a customer something untrue: a delivery date it invented, a returns window that doesn’t exist, a discount nobody approved. That happens when the bot answers from general knowledge rather than your data. A service agent built properly does the opposite: it looks things up first and refuses to guess.
What a service agent does
What it handles, and what it hands over
| Enquiry type | Typical handling |
|---|---|
| Where is my order? | Looks up tracking and drafts or sends a reply with the latest status |
| Product questions (size, materials, care) | Answers from product data; flags if the data doesn’t cover it |
| Change of address or order before dispatch | Drafts the change for a person to action |
| Returns and exchanges | Explains your policy, collects details, and hands to a person to approve |
| Faulty or damaged goods, refunds, complaints | Gathers details and photos, summarises, and hands to a person straight away |
| Wholesale, press, partnership enquiries | Tags and routes to the right person |
Consumer law is not optional
Under the Australian Consumer Law, customers have consumer guarantees that a store’s own returns policy cannot take away. An AI agent that tells a customer “no refunds on sale items” when the item is faulty creates a real problem for you. Our agents are built so they do not make statements about refund entitlement on their own; those conversations are drafted for a person, who stays responsible for them. If you want help with your policy wording itself, the ACCC’s guidance for businesses and your own legal adviser are the right sources.
Measuring it
Before the build we sample your enquiries for a few weeks: volumes by type, time to first reply, time spent per reply. After the pilot we measure the same things, plus how often staff edited the agent’s draft and by how much. If edit rates on a category stay high, the agent keeps drafting there rather than sending.
Peak season
For many Melbourne online stores the busiest stretch runs from late-November sales through to Christmas, when message volume can jump while the team is stretched. A service agent built and trusted before that period helps most. A build started in the middle of it is risky, so we usually recommend a pilot earlier in the year.
Questions
Will the agent reply to customers on its own?
Only if you choose that, and only for categories you have seen it handle well, such as order-status questions with tracking data. Everything else is drafted for a person to send.
What about refunds and faulty goods?
Those go to a person. The agent can gather the order details and photos and draft a summary, but it does not decide outcomes or describe consumer guarantees on its own.
Which help desks does it work with?
We design around the inbox or help desk you use now, including shared mailboxes in Microsoft 365 or Gmail. Integration options vary by platform and plan, and the audit confirms what is possible for yours.