Email triage that sorts the shared inbox and drafts replies for approval.
AI email triage reads each message in a shared inbox, labels it, routes it to the right person and drafts a reply from your approved information. A person reviews and sends the draft, so the inbox moves faster without anything going out that nobody has checked.
What does email triage do with a shared inbox?
A shared inbox such as info@ or hello@ collects everything: new enquiries, customers chasing an answer, supplier invoices, job applications, newsletters and the occasional complaint. Someone has to open each one, decide what it is and pass it on. Triage takes over that first pass.
- Classifies each message against categories your team already uses, and applies a label or moves it to a folder.
- Routes it to the person or team who owns that kind of request, with a one-line summary.
- Pulls out details such as names, order or booking references, dates and amounts, and records them where they belong.
- Drafts a reply using your approved answers and the customer’s history, saved as a draft in the same thread.
- Flags urgency, for example a complaint, a deadline mentioned in the text or a customer writing for the third time.
Is email triage a job for an agent?
Tick what is true of your email triage work.
Tick the statements that apply and the suggestion appears here: an AI agent, ordinary automation, or a person with software support.
Start with the categories your team already uses
The categories decide almost everything else, so write them down before anything is built. Five to ten is usually enough. Each one needs an owner and a default action.
| Category | Default action |
|---|---|
| New enquiry | Label, extract contact details and request, create a CRM record, draft a first reply. |
| Existing customer request | Find the customer, link the thread to their record, route to their account owner. |
| Invoice or supplier message | Forward to finance or into [invoice processing](/use-cases/invoice-processing/); no reply drafted. |
| Complaint | Mark urgent, notify a named manager, draft a holding reply only if your policy allows one. |
| Newsletters and notifications | Label and archive, so they stop crowding the inbox. |
| Unclear | Leave unlabelled for a person. Guessing here causes more work than it saves. |
Messages that contain two requests, such as a booking change plus a billing question, are common. Decide whether they go to one owner who handles both, or are split.
Routing also needs a rule for absence. If the owner of a category is on leave, messages should go to a named cover person rather than sit in a folder nobody is watching. The same applies to urgent flags sent out of hours: decide whether they wait for the morning or reach someone on call.
Why replies should wait for a person
Email is permanent and easily forwarded. A reply that quotes the wrong price or promises the wrong date is hard to take back. That is why the first version saves drafts rather than sending them: your team reads, edits if needed and presses send.
The human-in-the-loop guide covers how to design approval steps that people actually use, rather than click through.
How much time could email triage give back?
Put your own numbers in.
The starting values are an illustrative example, not a benchmark or a neu figure. Use your own volume, the time each one takes today and the share you expect the system to handle without a person.
Time saved is only part of the case: response times, missed work and errors matter too. The ROI guide covers how to set a baseline, and the cost guide what the build and running costs include.
What it needs access to, and what to protect
Triage connects to your mailbox through the provider’s API, for example Google Workspace or Microsoft 365. An administrator grants access to the specific mailbox, ideally the shared one rather than personal accounts. To draft useful replies it also needs read access to your approved answers and, often, your CRM.
Email carries a lot of personal data, including things people should not have sent. Decide what the AI model is allowed to see, how long anything is stored outside the mailbox and which provider processes it. Our guide to AI agents and GDPR sets out the questions to settle with your advisor.
What usually goes wrong
- Long forwarded chains, where the real request is buried three messages down. The triage should read the latest message first and summarise the rest.
- Attachments that matter, such as a photo of damage or a signed form. Decide whether attachments are read, and route anything unreadable to a person.
- Auto-reply loops, where an out-of-office triggers a draft that triggers another out-of-office. Filter automated mail early.
- Tone drift, where drafts become too formal or too casual. Keep a few approved example replies in the instructions.
- Moved goalposts, where categories change but nobody updates the rules. Give one person ownership of the category list.
How to scope and measure a first release
- Export a few weeks of inbox subjects and sort them by hand into categories. This shows what the volume really is.
- Pick the two or three categories that take the most time and have clear answers.
- Run triage in labelling-only mode for a week and check its labels against your team’s.
- Turn on drafts for those categories, and measure time to first reply, drafts sent unchanged and messages misrouted.
- Widen the categories once the numbers are steady.
If your inbox is mostly predictable notifications, plain mail rules may do the job without AI. Where triage pays off is mixed, free-text mail that needs reading. For budgeting, see the AI agent cost guide.
Common questions
Will the AI read every email in our mailbox?
Only the mailbox and folders you grant it. Most setups connect a shared inbox, not staff accounts. You can also exclude senders or labels, such as HR or legal, so those messages are never passed to the model. Keep a written record of what is in scope for your data-protection documentation.
Can it reply in Irish as well as English?
Current language models can read and write Irish, but quality varies more than in English. If you receive mail in Irish, test drafts with a fluent speaker before relying on them, and keep those replies in the approval queue. The same applies to any other language your customers use.
Does it work with Outlook and Gmail?
Yes, both Microsoft 365 and Google Workspace let approved software read mail, apply labels or categories and save drafts through their APIs. Your administrator grants the permissions. Older on-premises mail servers can be harder and are worth checking before you scope anything.
What if it labels something wrongly?
Your team moves it, and that correction is worth logging. A weekly look at corrections shows which categories are confused and whether the instructions need examples. Misroutes are why urgent categories, such as complaints, should notify a person rather than rely on a label alone.
Bring us the task.
Tell us what happens today, which tools are involved and where it goes wrong. On a free 30-minute review, we’ll work out whether an agent, an automation or neither is the right fix, and what a sensible first release would be.
Start a project.

Book a call or send us a message. We read everything.
tash@neu.ie
Working with teams in Ireland and worldwide.
Free 30-minute AI review.