RoutineSmithHow to Run a Shared Inbox Workflow in a Small Team
A shared address becomes an observable workflow when every message has a visible owner, state, response target and duplicate check. This guide shows a practical operating model for a small team—before automation or AI is allowed to prepare anything.
What a shared inbox workflow needs
Start with four controls that a teammate can see without asking around: who owns the conversation, what state the work is in, when the next action is due and which conditions require human judgment. Together, they make ownership and next actions observable.
Define ownership before adding automation
Ownership should be singular. An on-call rotation can decide who receives new work, and an escalation path can name a backup, but one person remains accountable for the current conversation. When ownership changes, record the new owner and time. “The inbox team” is not an owner because it does not tell anyone who must decide or act.
Automation should never invent the accountable person. A deterministic assignment rule may nominate the owner, but the team needs a visible way to accept, transfer and escalate responsibility. Only after that path works should AI classify or prepare a draft.
Use two layers: queue and lifecycle state
A queue answers what kind of decision does this need? A lifecycle state answers where is the work? Mixing them produces labels that cannot explain whether a message is waiting, being reviewed or already finished.
URGENT
COMMITMENT
ROUTINE
REFERENCE
MATCH CHECK
NEW → OWNED → WAITING FOR INFO / DRAFTED / REVIEW → COMPLETE
Duplicate branch: NEW → DUPLICATE → CLOSED
Keep the taxonomy small enough for two people to classify the same message consistently. If a label does not change the owner, required decision or next action, it probably does not deserve to be a queue.
For a commercial intake example, use the quote request intake checklist to define complete-enough facts before a request reaches estimation.
Set response targets the team can measure
A useful target defines five things: the business time zone, covered hours, the event that starts the clock, conditions that pause it and the person responsible when the target is missed. Without those definitions, two team members can report different results from the same conversation.
Track acknowledgment and resolution separately. An acknowledgment confirms receipt without making a delivery promise; resolution records that the agreed action or answer is complete.
A practical ownership, target and deduplication table
The times below are a synthetic example, not an industry benchmark or contractual SLA. Adapt them to actual volume, staffing and risk.
| Queue / state | Accountable owner | Example internal target | AI may prepare | Human decision | Deduplication rule |
|---|---|---|---|---|---|
URGENT · NEW | Duty manager | Acknowledge within 30 business minutes | Known facts and risk flags | Priority, promise and response | Search open conversations first; link any match and stop the new route |
COMMITMENT · OWNED | Named coordinator | First review within 2 business hours | Summary, missing fields and draft | Price, refund, deadline, exception or uncertain intent | Keep one canonical ID; no second outbound draft |
ROUTINE · OWNED | Assigned operator | Review within 4 business hours | Low-risk draft from complete facts | Factual and tone check, then send | Compare sender, thread and recent open request before drafting |
ROUTINE · WAITING FOR INFO | Existing owner | Recheck next business day | Missing-information checklist | Whether and how to request information | Reply only on the canonical thread |
REFERENCE · NO ACTION | Inbox steward | Daily sweep | Label and short summary | Confirm no action is required | Merge or label repeated notices; never delete automatically |
MATCH CHECK · DUPLICATE | Owner of original | Link within 1 business hour | Match explanation | Confirm it is a true duplicate | Preserve source, link to original, create no second owner, draft or send |
ROUTINE · COMPLETE | Final owner | Close after evidence is stored | Closure summary | Confirm the commitment was fulfilled | New information reopens or creates a related case; history remains |
Prevent duplicate ownership and replies
Duplicate prevention is a workflow rule, not a claim that a platform will find every match. Use a stable message or conversation ID, normalize only the fields required for comparison and search open work before assigning or drafting. Preserve every source message so the team can reconstruct what happened.
- Identify.Keep a stable message or conversation ID.
- Compare.Search active conversations using sender, thread and relevant request fields.
- Link.Preserve the new source and connect it to the canonical case.
- Stop.Create no second owner, draft, reminder or outbound reply.
- Review.Send ambiguous matches to a person instead of guessing.
Northstar Field Services: one case, one owner
Northstar Field Services is a fictional 12-person maintenance company using service@northstar.example. This scenario illustrates control logic; it is not a customer result or performance benchmark.
An inspection request arrives without a location. The team records case NS-104, Maya becomes owner and the case enters REVIEW. AI extracts known facts and prepares a missing-information list. Nothing is sent.
The sender forwards the request with a different subject. The duplicate check finds the open case, preserves the new source and links it to NS-104. No second owner or draft is created. Maya confirms the match.
The location arrives. The case moves to DRAFTED. AI prepares a factual acknowledgment; Maya checks the address and confirms there is no price or date promise, then sends and closes with evidence.
Observed operating result: one case, one owner, one visible exception and one reviewed response. The example does not claim that AI answered the inbox.
Map the policy to Google Workspace or Microsoft 365
Google documents how a Google Group can be used as a Collaborative Inbox, including taking or assigning conversations and marking them complete, duplicate or no action. Gmail delegation is a different model that gives delegates access to read, send and delete within the delegated account.
Microsoft describes shared mailboxes as a way for a group to monitor and send from a common address; they are not intended for direct sign-in. Microsoft also notes that sent-item behavior must be configured if the team needs replies visible in the shared mailbox rather than only in an individual sender’s Sent Items.
These products provide platform capabilities. Ownership, response targets, deduplication controls and the human approval gate in this guide are RoutineSmith operating recommendations, not guarantees from Google or Microsoft.
Draft the supervised AI triage layer.
Choose the queues, approval boundary, stop conditions and seven-day measure for your own team. The planner runs in your browser and does not connect to an inbox.
Draft your supervised AI triage policyRun a seven-business-day control test
Record a baseline before changing the workflow. During the test, use representative messages and retain every exception. Measure actionable messages, assignment on time, conversations with exactly one owner, duplicate candidates, confirmed duplicates, duplicate drafts or replies, missed review targets, reopenings, review minutes and exceptions by reason.
Seven business days provide only an initial signal, and only when enough messages arrive. The test does not prove ROI or universal readiness for production. If review effort grows, exceptions remain unclear or the team cannot explain a decision, revise the policy before adding automation.
Make a GO, REVISE or STOP decision
Use the evidence collected by your team; there is no universal threshold. Record who made the decision, the observed measures and the unresolved exceptions.
- GO: continue only when owner, state and response targets were used consistently, duplicate and outbound errors remained controlled, reviewer effort was acceptable, and the chosen outcome improved against the baseline without shifting hidden work to someone else.
- REVISE: change the policy and test again when ownership gaps, exception handling, the true review cost or useful improvement remain unclear.
- STOP: do not connect or expand automation when sending is uncontrolled, the task includes sensitive or high-consequence action, evidence is missing, the underlying process is unstable, or the test creates no useful improvement worth the operating cost.
A seven-day signal and a calculator estimate can support this decision, but neither proves ROI or production readiness.
Common shared inbox mistakes
- Giving everyone access but assigning no accountable owner.
- Using labels that mix urgency, department and lifecycle state.
- Calling a fast acknowledgment a completed resolution.
- Closing a possible duplicate without preserving and linking the source.
- Allowing AI to send prices, refunds, promises or uncertain answers.
- Measuring time saved without recording corrections and reviewer effort.
- Expanding the workflow before the exception path is understood.
Shared inbox workflow questions
What is a shared inbox workflow?
A shared inbox workflow is a documented process for receiving, assigning, reviewing and closing team email. It gives each actionable conversation one accountable owner, a visible lifecycle state, a response target and an exception path. The shared address is only the entry point; the operating rules make ownership, state and next actions observable.
How many people should own one email?
One person should be accountable for an active conversation at any given time. A backup or escalation owner can be named, but responsibility should never be left with “everyone.” When ownership changes, record who took it and when so the team can see the current decision maker.
What statuses should a small team use?
Start with a small lifecycle: NEW, OWNED, WAITING FOR INFO, DRAFTED, REVIEW and COMPLETE. Add DUPLICATE as a closed branch. Keep queues separate from states: a queue describes the decision required, while a state describes where the work currently sits.
Does a shared inbox need an SLA?
It needs measurable internal response targets, but they do not have to be contractual service-level agreements. Define business hours, time zone, when the clock starts, what pauses it, and who responds to a miss. Measure acknowledgment and resolution separately because they answer different operational questions.
Can AI reply automatically from a shared inbox?
AI can prepare a classification, summary or low-risk draft, but automatic sending is a poor first pilot. Keep prices, refunds, commitments, deadlines, account changes and uncertain intent behind human approval. Expand automation only after the team has measured exceptions and demonstrated that the review process works.
How do you prevent two people from replying?
Assign one canonical conversation ID, search active work before creating a new case and allow only the current owner to move an approved draft to send. If a possible duplicate appears, preserve it, link it to the canonical case and require a person to confirm the match before closing it.
Is Google Collaborative Inbox the same as Gmail delegation?
No. A Google Group can be configured as a Collaborative Inbox with conversation assignment and status features. Gmail delegation gives another person access to read, send and delete within a delegated mailbox. They are different operating models, so choose the one that matches your ownership and audit needs.
Where are replies from a Microsoft shared mailbox stored?
Microsoft documents that sent-item behavior depends on configuration. By default, a reply sent from a shared mailbox may be saved in the individual sender’s Sent Items rather than the shared mailbox. Configure shared sent-item visibility deliberately so the team can verify what was communicated.