What it does
The member enquiry agent classifies an incoming written enquiry, retrieves the member's record and recent activity through MCP servers, finds the applicable policy or product wording, and drafts a reply with citations. A person reviews and sends every reply.
It is designed for regulated service teams: superannuation, insurance, banking and utilities, where a wrong answer is costly and an untraceable one is worse.
Built for contact centres where the research takes longer than the reply. The agent assembles the account, recent activity and the right policy wording before a person opens the enquiry.
Ships in shadow mode first, scored against the replies your team actually sent.
What it may and may not do
| Operation | System | Access | Note |
|---|---|---|---|
| Read the enquiry and thread | Contact platform | READ | |
| Classify topic and urgency | Internal | COMPUTE | |
| Fetch member record | Registry | READ | Agreed fields only |
| Fetch recent transactions | Registry | READ | |
| Retrieve product wording | Document store | READ | |
| Draft reply with citations | Contact platform | DRAFT | Review required |
| Send to member | Contact platform | NEVER | Not built |
How autonomy is earned
Level 1 · Shadow
Drafts against live enquiries, visible only to the evaluation team.
Risk: none
Level 2 · Assist
Drafts appear beside the enquiry. Staff edit and send.
Risk: none to members
Level 3 · Approve
For agreed low-risk topics, a team leader approves batches of drafts.
Risk: low, sampled weekly
Compatibility
- CRM
- Salesforce, Dynamics 365
- Contact platform
- Genesys, Zendesk, Salesforce Service Cloud
- Documents
- SharePoint, Confluence
Probably not for you if
- Enquiries that are mostly phone calls rather than written
- No agreement yet on which member fields the agent may read
- Fewer than a few hundred enquiries a month
Questions we get asked
Can it give financial advice?
No. It drafts factual replies from the record and approved wording. Anything that looks like advice is routed to a qualified person.
How do you measure accuracy?
Against a test set of your own historical enquiries and the replies actually sent, scored by topic.
