Two colleagues answer the same email while another customer's request sits untouched. A message marked “read” disappears from everyone's attention. An AI-generated reply sounds helpful but promises something the business has never agreed to do.
A shared inbox needs operating rules that connect each conversation to a responsible person and a next action. A trained virtual assistant can resolve routine requests independently within an approved scope, use AI for suitable drafting and bring specific decisions to the manager. Here is a practical playbook to adapt to your business.
Assign the conversation before answering it
Give each active conversation one reply owner. Record its category, status, next action and due time. Start with a few useful statuses: new, in progress, awaiting customer, awaiting internal decision and resolved.
“Awaiting internal decision” still needs an owner who follows up. “Resolved” means the request has reached its agreed finish point, rather than simply that somebody sent an email. A customer reply introducing unfinished work should return the conversation to an active queue.
The VA can check incoming requests, link them to the correct record and take the next permitted action. If a thread contains two requests, track both. Sending the requested brochure does not resolve the pricing question beneath it.
Use a triage matrix with real decision rights
This example separates assistance, routine authority and exceptions. Replace the boundaries with ones your team can actually apply.
| Request | What AI may draft | What the VA may resolve | When another decision is needed |
|---|---|---|---|
| Standard service question | Wording from current approved information | Answer within the published service scope | Conflicting information or a requested exception |
| Routine appointment change | Options from verified availability | Make and confirm changes within booking authority | Capacity conflict, cancellation dispute or unusual commitment |
| Copy of an existing document | A short covering message | Send the correct version to an authorised recipient after required checks | Uncertain identity, access rights or document version |
| Complaint about service | An acknowledgement without a proposed remedy | Record the concern, gather facts and maintain contact | Refunds, disputed promises or remedies beyond delegated authority |
| Payment-detail change or suspicious request | No automated action on the request | Flag and route through the verification process | Authorised finance or security review before changes |
A VA should not need a manager's approval to answer every routine question once trained and authorised. Equally, a manager's job title does not confer technical or professional authority they do not hold. Route specialist decisions accordingly.
Keep drafting permission separate from sending permission
For this playbook, AI produces suggestions in a draft area. The reply owner checks the recipient, attachments, factual basis, commitments and latest conversation before sending. A previously approved draft can become outdated when a new message arrives.
Give the tool an approved answer source and a narrow instruction, such as: “Draft a reply using the supplied service information. Identify unanswered questions. Do not add prices, dates or promises.” Review what it actually produces.
Decide what information the tool may receive before connecting the inbox. The OAIC recommends assessing product suitability, information access, privacy risks and human oversight. It also recommends keeping personal information, particularly sensitive information, out of publicly available generative AI tools as best practice. Read the OAIC's commercial AI guidance.
Start with fictional messages and generic templates. An approved business tool still needs an assessment of its actual configuration and permitted use.
Write templates that leave decisions visible
Keep a small template library with an owner, review date and conditions for use. These examples contain placeholders that must be replaced with verified details before sending.
Routine information: “Our current service includes [confirmed inclusion]. You can find the details in [approved document or link]. To arrange the next step, please [specific action].”
Clarification: “I can help with [request]. To make the correct change, please confirm [specific missing detail] through [approved channel]. We have not changed [booking or record] yet.”
Decision pending: “I have referred your request about [issue] to [responsible role]. The decision is still pending. I will contact you again by [agreed contact time], even if we are still waiting for an answer.”
Only say that a referral happened after it did. Choose a contact time the team can meet. The template should explain the present position without implying an unapproved refund, booking or completion date.
Prevent duplicate replies across people and channels
Use the inbox's assignment or collision-warning features where available. Agree that the assigned owner sends the reply unless responsibility is explicitly transferred. Before sending, check the latest inbound message, sent items and relevant internal notes.
Suppose a customer emails a booking request and then calls. The phone colleague confirms an available slot and records the call against the same case. The VA sees that update, cancels the obsolete draft offering different options and records whether any written confirmation remains due. This is a hypothetical example of one request arriving through two channels.
Link related threads using a reliable customer or case reference. Avoid automatically merging messages solely because names look similar. Preserve the original correspondence and any distinct outstanding requests.
At handover, list active conversations, promised contacts and decisions still needed. The incoming owner acknowledges responsibility so the same message is not answered twice—or left to both people.
Escalate the decision with enough context
An escalation should state the customer's request, verified facts, action already taken, decision required and relevant deadline. Include source links so the reviewer can check the record without searching the whole mailbox.
Define immediate escalation triggers, including suspected fraud, exposed personal information, safety concerns and threats. Use the appropriate incident route instead of leaving those messages in the ordinary queue.
For payment changes, the ACSC recommends approval controls and calling the sender using a known, verified number rather than a number supplied in the email. A familiar-looking sender or existing thread does not establish that a request is legitimate. See ACSC guidance on preventing business email compromise.
Review missed follow-ups, duplicate replies and reopened conversations with the team. If the same question repeatedly reaches a manager, clarify the procedure or delegated authority. Our hybrid team responsibility guide helps assign those decisions. Talk to Nexus about inbox support to build the role around your actual queue.
