Monday, 9.15am. The shared inbox at a fictional six-person office fit-out firm in Ubi shows 83 unread messages. A supplier asking to rebook a loading bay sits between two newsletters. A client's note about a cracked glass partition is on the second screen. And nowhere at all is the quote the firm sent a prospective client twelve days ago, because nobody has written back.
That last item is the expensive one. An inbox records what arrived; it keeps no track of what you are waiting for, a gap that a playbook on AI inbox triage and follow-ups identifies as the reason follow-ups quietly go cold. The quote awaiting a decision, the invoice awaiting payment, the client who said they would check with their director: none of these produce new mail, so they drop out of sight.
A triage routine has to do two jobs: sort what arrives, and hold on to what has not. The routine below works with a person doing every step. An AI tool can take on the sorting, drafting and remembering, but it needs the same rules written down first, and the rules are most of the work.
Write down the rules that live in your head
Your team already triages; the rules are just unwritten and applied differently by each person. Put them into three short lists.
Labels. Six to eight categories your mail actually falls into, named for what happens next. For the fit-out firm: New enquiry, Active project, Defect or complaint, Supplier, Building management, Invoice or payment, Admin, Noise.
Priority rules. What jumps the queue, whatever the label. For the firm: any mention of a defect, site access or a cancellation; anything from a client with a project on site this month; any payment reminder addressed to the firm.
Routing. Who owns each label. New enquiries go to Kumar in sales. Active projects go to the project lead named in the subject line. Suppliers and building management go to Siti in operations. Defects and invoices go to the owner, Daniel. Noise is archived, and someone unsubscribes from the worst offenders once a month.
Test the rules before trusting them. Ask a colleague to label 20 messages from last week using only the lists, then compare with your own labels. Where you disagree, the rule is unclear, not the colleague.
Make triage end in a draft, not a to-do
For every message that needs a reply, triage should produce a draft, ready for its owner to approve, edit or reject. Labelling and drafting happen inside the business, so they can be done at any point. Sending is the step that leaves the building, so it waits for a person.
Keep one queue, such as a "To approve" label or a shared drafts folder, and review it at fixed times rather than all day. Two sittings, mid-morning and mid-afternoon, suit many small teams. Approving a batch of finished drafts is faster than composing replies one interruption at a time.
Without AI, the person triaging writes a two-line draft or picks a saved template and fills it in. With AI, the tool reads the thread, checks the relevant documents, such as the standard lead-time sheet, and places a draft in the queue. Either way, nothing sends until its owner approves it.
Set limits on what any draft may say. At the fit-out firm, no draft commits to a completion date, quotes a price that is not on the rate card, or accepts responsibility for a defect. Those replies are written by Daniel.
Keep a waiting-on list the inbox cannot
Whenever a thread ends with the firm waiting on someone, it goes onto a waiting-on list. A spreadsheet is enough. Each line needs:
- who you are waiting on, and for what;
- the date you last sent something;
- the next nudge date;
- the owner of the thread;
- attempts so far;
- the stop rule.
Then set a cadence and write it at the top of the list. The fit-out firm chose, for illustration: a first nudge after three working days, a second after a further five that asks one specific question, and a final note after that which makes it easy to say no. After three attempts the line is closed, or handed to Daniel for a phone call. Some threads are never chased by anyone but Daniel, such as a client in a dispute.
A nudge should do more than "just check in". Our guide to a sales follow-up that helps explains how to answer the question the other person left open.
With AI, the tool can add a thread to the list automatically when the last message went out from the firm with a quote or a question in it, and draft each nudge on its due date. The draft joins the approval queue like any other reply.
The daily and weekly rhythm
Each morning, 15 to 20 minutes:
- Deal with anything that matched a priority rule overnight.
- Label everything new; archive the noise.
- Approve, edit or reject the drafts in the queue.
- Check the waiting-on list for nudges due today, and approve them.
- Give anything left over an owner and a date.
Each Friday, 20 minutes:
- Sort the waiting-on list by days outstanding and decide, line by line, whether to nudge, call or close.
- Look at the week's rejected drafts. If several failed for the same reason, fix the rule or the document behind them.
- Check the labels. Merge any that went unused, and pull out anything hiding in Admin that deserves its own label.
An AI tool can also produce a morning summary of what arrived overnight, sorted by priority with drafts attached, and alert the owner straight away when a message matches a priority rule rather than waiting for the summary. In the fictional firm's first Friday review, the list surfaced four quotes that had gone more than ten days without a reply. Those figures are illustrative, not a result.
Adding AI without losing control
Add a tool to the step that takes the most time once the rules have stopped changing, usually labelling and drafting. Start with the shared inbox, not personal mailboxes, and give the tool access only to what triage needs.
Keep sending, commitments and chasing decisions with people until reviewed history shows that drafts in a category rarely need edits. Keep a simple record of each message's label, draft, approver and send time, so that "did we ever reply to them?" has an answer. The same record tells you which categories are ready for less review, and which never will be.
Common questions
Does this routine need an AI tool?
No. Labels, a drafts queue, a waiting-on list and a fixed daily rhythm work with a person doing every step; a tool can take on the sorting, drafting and remembering once the rules are written down.
How often should we chase a quote that has gone quiet?
Set a cadence and stop rule in advance. The fictional firm nudges after three working days, again after a further five with a specific question, then sends a final note and closes the line or hands it to a person to call.
Should AI-drafted replies send automatically?
Not at the start. Keep every send in an approval queue until reviewed history shows that drafts in a category rarely need edits, and keep commitments such as dates, prices and defect responses with a person.
Sources
Something changed or worth adding?
Send a correction or suggestion