AI Email Triage: Automatic Classification, Summarisation & Routing

What is AI email triage?

Most businesses have at least one shared inbox — info@, accounts@, support@, sales@ — that receives a stream of emails every day. Someone has to read each one, figure out what it is, decide who should handle it, and forward it along. Often with attachments that need to be opened, read, and understood.

That process is slow, inconsistent, and mind-numbing. It's also a significant chunk of admin time in many businesses — especially in professional services, trades, property, and any industry that does business over email.

AI email triage automates the sorting. AI reads each inbound email, classifies it by type and urgency, summarises the content and any attachments, and routes it to the right person or team — automatically.

Why it matters

The cost of manual email triage is real but often invisible:

• Response time — emails sit in shared inboxes for hours (or days) because nobody is assigned to them. Enquiries go cold. Urgent requests wait behind routine ones.
• Admin hours — a receptionist or admin assistant spending 2–3 hours a day reading, sorting, and forwarding emails is a significant cost. And it's not the best use of their skills.
• Inconsistency — different people classify the same email differently. One person forwards a client complaint to the manager; another responds directly and makes it worse.
• Dropped balls — emails that don't obviously belong to anyone get ignored. Important requests fall through the cracks, especially during busy periods or staff absences.

How it works

AI email triage has three components:

1. Classification

AI reads the email subject, body, and metadata and assigns a category: new enquiry, existing client request, invoice/billing, complaint, job application, spam/irrelevant, urgent matter, and so on. Categories are customised to your business.

2. Summarisation

AI generates a one-paragraph summary of the email and its attachments. If someone sends a 3-page PDF quote with a brief email saying "please review," the AI extracts the key details — vendor, amount, scope — so the recipient knows what they're looking at without opening the file.

3. Routing

Based on the classification, the email is routed to the right person or team. A new sales enquiry goes to the sales team in your CRM. An invoice goes to accounts. A complaint goes to the relevant manager. Routing rules are configured to match your org structure.

Practical use cases

General enquiry inbox

The info@ inbox that receives everything — sales leads, partnership requests, job applications, supplier pitches, and spam. AI classifies and routes each one, ensuring genuine enquiries get a fast response and spam gets filtered.

Accounts inbox

Invoices, payment reminders, credit notes, and queries arrive in accounts@. AI extracts vendor name, amount, and due date from attached invoices, matches against purchase orders where possible, and routes exceptions (mismatches, duplicates, unknown vendors) for review.

Support inbox

Customer support emails are classified by issue type and urgency, summarised with customer and account context, and routed to the appropriate support tier. Urgent issues are escalated immediately rather than waiting in the queue.

Property management

Tenant communications, maintenance requests, lease queries, and compliance documents. AI classifies by property and issue type, routes to the relevant property manager, and flags urgent maintenance (water leaks, safety issues) for immediate attention.

Professional services intake

New client enquiries, document submissions, referrals, and follow-up communications. AI classifies by service type, extracts key details from attached documents, and routes to the relevant partner or team with a summary.

Risks and limitations

• Misclassification happens — AI gets it right 90–95% of the time, but some emails are genuinely ambiguous. Build a "needs review" category for borderline cases rather than relying on perfect classification.
• Email isn't always the right channel — if you're getting 500 emails a day because you don't have a proper client portal, intake form, or ticketing system, AI email triage is treating a symptom. Consider whether a better input channel would solve the root problem.
• Attachment complexity — AI handles standard documents (PDFs, Word, images) well. Highly technical drawings, encrypted files, or unusual formats may not be summarised accurately.
• Privacy and confidentiality — email content can contain sensitive information. The AI system must be deployed with appropriate security, access controls, and data residency — especially for industries with regulatory requirements.
• Human override — always provide a way for staff to reclassify and reroute. AI is a first pass, not the final word.

Getting started

1. Pick one inbox — choose the shared inbox with the highest volume and the clearest pain point. Info@ or accounts@ are common starting points.
2. Define your categories — work with the people who currently sort the inbox. What categories do they use? What's the routing logic? Document the rules they follow (even the unwritten ones).
3. Run a baseline — process a week of historical emails through the AI and compare its classifications against what humans did. This tells you accuracy before you go live.
4. Deploy in shadow mode — AI classifies and routes, but staff still see everything and can override. Run this for two weeks and refine based on misclassifications.
5. Go live — switch to AI-first routing with human override. Monitor accuracy weekly for the first month.

Frequently asked questions

Does it work with Outlook/Gmail?

Yes. AI email triage connects to Microsoft 365 and Google Workspace via their standard APIs. Emails are processed in real-time or near-real-time.

Can it send replies?

It can send automated acknowledgements (e.g., "Thanks for your enquiry, we'll be in touch within 24 hours"). For substantive replies, AI drafts a response for a human to review and send — keeping a person in the loop for anything that matters.

What about emails in different languages?

Modern AI handles multiple languages well. For Australian businesses, most inbound email is in English, but the system can classify and summarise emails in other languages too.

How much does it cost?

A single-inbox triage system typically costs $15K–$25K to build and deploy, with ongoing costs of $200–$500/month for hosting and API usage depending on volume.

How fast is it?

Most emails are classified, summarised, and routed within 30–60 seconds of arrival. Emails with large attachments may take slightly longer for the summarisation step.