How We Automated Our Email Inbox With AI

Every morning started the same way. Open Gmail. Scroll through 40–60 new emails. Mentally sort them: junk, junk, supplier invoice, junk, patient enquiry, newsletter we forgot to unsubscribe from, important letter from the health fund buried under three marketing emails.

Thirty minutes later, the inbox was sorted. Nothing was replied to yet — just triaged. And that was on a good day. On a busy Monday after a long weekend? Closer to 45 minutes of pure inbox housekeeping before any real work started.

We run a small dental practice in Darwin with 4 staff. We don't have a dedicated admin person whose only job is email. The office manager handles it, and every minute she spends sorting emails is a minute she's not managing the practice.

So we built an AI to do it instead. We call it the Smart Inbox, and it's one of the most useful tools we've made.

What we tried first

Before building our own solution, we went through the usual options:

• Gmail filters — we had about 30 of them. They caught some things, missed others, and needed constant maintenance as new senders appeared. A filter can match a sender or a keyword. It can't understand what an email is actually about.
• SaneBox (~$7/month) — better than raw filters, but still just sorting into folders. It couldn't take action on emails, save attachments, or unsubscribe from things.
• Clean Email (~$10/month) — good for bulk cleanup, but it's a reactive tool. You still have to go in, review, and decide. It doesn't handle the daily triage.

Between those subscriptions and the remaining manual effort, we were paying roughly $17/month and still spending 20–30 minutes a day on email. The tools helped at the margins but didn't solve the actual problem: someone still had to read every email and decide what to do with it.

What the AI does now

Our Smart Inbox connects to the Gmail API and processes every incoming email automatically. Here's the workflow:

1. Read and classify — the AI reads each email and determines its intent. Is it a supplier invoice? A patient enquiry? A marketing email? A notification from our practice management system? An insurance reminder? It understands context, not just keywords.
2. Label by category — each email gets automatically labelled: supplier, patient, admin, marketing, financial, clinical. No more manually dragging emails into folders.
3. Prioritise — the AI assigns a priority level. Anything that needs a human response gets flagged. Routine notifications get filed quietly. The office manager's morning inbox shows only what actually matters.
4. Save attachments — invoices, statements, lab reports, referral letters — they all get saved to the right folder automatically. No more downloading PDFs and figuring out where they go.
5. Unsubscribe from junk — marketing emails we never read? The AI identifies persistent junk senders and unsubscribes. Not archives — actually unsubscribes, so they stop coming.

The whole thing runs in the background. By the time the office manager opens her inbox in the morning, the work is already done. She reviews the flagged items — usually 5 to 8 emails that genuinely need human attention — and gets on with her day.

Before and after

The numbers speak for themselves:

• Time on email triage: dropped from 30 minutes/day to about 5 minutes reviewing flagged items
• Important emails missed: zero. Before the AI, things occasionally slipped through the cracks — an insurance deadline buried under newsletters, a patient complaint that sat for two days. That doesn't happen anymore.
• Response time: improved significantly. When the only emails in your "needs attention" view are genuinely important, you respond to them faster.
• Subscriptions cancelled: SaneBox and Clean Email — about $17/month we no longer pay. Not a huge saving on its own, but it adds up. We wrote about the full subscription picture in our cost breakdown post.
• Attachment filing: fully automatic. This used to eat 5–10 minutes a day on its own — downloading invoices, renaming files, saving them to the right folder.

That's roughly 25 minutes a day given back to the office manager. At a loaded cost of around $45/hour, that's about $75 per week in staff time — over $3,900 per year — on just email triage.

How it works (without the jargon)

The system is straightforward. A Python script connects to Gmail via Google's API, which gives it secure access to read and manage emails. When a new email arrives, the script pulls it and sends the content to a language model — the same kind of AI that powers tools like ChatGPT, but running locally on our own server.

The AI reads the email and returns a structured classification: category, priority, whether it needs a response, whether there are attachments worth saving, and what action to take. The script then executes those actions — applying Gmail labels, moving to folders, saving files, or hitting the unsubscribe link.

Because the AI runs locally, no email content leaves our network. That matters when you're handling patient information. Nothing gets sent to OpenAI or any third-party cloud service. The AI processes everything on our own hardware and the data stays where it should.

What it's bad at

We believe in being honest about limitations. The Smart Inbox is not perfect, and we don't pretend it is.

• Tone-sensitive emails — if a supplier sends something that's technically a routine update but the tone suggests a relationship issue, the AI might not catch that. Humans are better at reading between the lines.
• Clinical judgement — if a patient emails describing symptoms, the AI can flag it as high-priority, but it absolutely should not be making clinical assessments. A human always reviews anything clinical.
• Novel situations — the first email from a brand-new type of sender might get miscategorised. The system learns over time, but the first encounter with something genuinely new can be a miss.
• Multi-intent emails — an email that's half invoice, half question about a policy change is harder to classify cleanly. The AI picks the dominant intent, which is usually right, but occasionally it'll label something as "financial" when the important part was the question buried in paragraph three.

That's why we designed it to flag rather than act on anything ambiguous. When in doubt, it puts the email in front of a human. The goal isn't to replace human judgement — it's to make sure human judgement is only needed where it actually matters.

This isn't just a dental practice thing

Email overload is universal. We built this for ourselves, but the pattern applies to almost every business we've talked to:

• Law firms — drowning in client correspondence, court notifications, and opposing counsel emails. Triage is critical and time-consuming.
• Real estate agents — enquiries from portals, vendor updates, settlement correspondence, marketing. A busy agent might get 80+ emails a day.
• Accountants — client documents, ATO notifications, bank statements, receipts. Half of their inbox is attachments that need to be saved in the right client folder.
• Trades businesses — quote requests, supplier invoices, job updates from subcontractors. The admin side of a trades business is relentless.
• Medical practices — pathology results, referrals, insurance correspondence, patient enquiries. Same problem we had, different speciality.

The underlying challenge is identical: too many emails, not enough time, and the important stuff gets lost in the noise. If your team spends more than 15 minutes a day on email triage, an AI inbox system would save you time and money.

The bigger picture

The Smart Inbox is one piece of a larger system we've built for our practice. It works alongside our AI phone receptionist, our voice dictation tool, and a dozen other automations that handle the admin so our team can focus on patients.

None of these tools are magic. They're practical, targeted solutions to specific problems. The email AI doesn't try to write replies or manage our calendar or predict the future. It does one thing well: it reads every email, sorts it, saves what needs saving, and tells us what needs our attention. That's it. And that's enough to save 25 minutes a day.

If your inbox is a source of daily dread — or if you suspect your team is spending more time on email than they should — get in touch. We'll walk through how your email workflow currently works and where an AI system could take the pain away. No obligation, just a practical conversation.