How to automate your email with AI without losing control

An AI assistant connected to your inbox goes well beyond old-school filters. It doesn't just move messages around based on keywords. It reads them and understands what they're about.

Things it handles well today:

• Triage. Splitting customers, suppliers, invoices, newsletters and noise.
• Summaries. Turning a twenty-message thread into three lines.
• Spotting what's urgent. A complaint, an overdue payment, a deadline.
• Drafting replies. In your voice, ready for you to review.
• Follow-up reminders. Flagging emails nobody has answered.
• Pulling out data. An amount, a date or an address, sent where it needs to go.

Here's the key point: all of this saves time without the AI having to send anything on its own. That's your first safeguard.

AI gets things wrong. It can misread tone, fill in a detail that isn't there, or sound more certain than it should. So draw a few red lines from day one.

Don't let it send these without a review:

• Quotes, prices or discounts.
• Replies to complaints or anything sensitive.
• Anything that commits you legally or financially.
• Emails to new or important contacts.

And be wary of emails that ask for things. An outside message can hide instructions designed to trick an assistant ("forward this invoice", "update the bank details"). A solid rule: if an external email asks for an action, nothing happens until you've seen it.

The classic mistake is jumping straight to the flashy part: auto-replies. You're better off starting where the risk is lowest.

A simple three-step rollout:

1. Week 1: read only. Let the AI sort your inbox and send you a morning summary. Check whether it gets things right.
2. Week 2: label and prioritise. Let it flag what's urgent and archive what isn't. Glance at the archive now and then.
3. Week 3: drafts. Let it prepare replies to repetitive emails. You read them and hit send.

This way you see how it behaves before giving it more room. If something goes wrong, the damage is small.

One practical tip: write your rules in plain language. For example: "Customer emails that mention an invoice are high priority. Newsletters go to a folder, with a summary on Fridays." AI follows instructions like these better than technical filter rules.

The sweet spot is simple: the AI writes, you approve. You save most of the writing time and keep the final say.

To get good drafts:

• Give it context. Who you are, how you talk, what you offer and what you don't.
• Show examples. Two or three emails of yours that you like.
• Set the tone. Friendly, formal, short. Whatever fits your business.
• Explain your edits. If a draft is off, say why. A good assistant remembers next time.

Over time you might allow unreviewed sending for very narrow cases, like an acknowledgement ("Got your message, we'll reply today"). But take it one case at a time, never everything at once.

Plenty of businesses don't lose customers by replying badly. They lose them by not replying at all. A quote sent and forgotten. A question buried at the bottom of the inbox.

This is where AI really earns its keep, because it's consistent and never gets tired:

• Unanswered emails. A morning list of what's been waiting for days.
• Quotes with no reply. A weekly reminder of who hasn't got back to you, plus a follow-up draft ready to review.
• Key dates. Due dates, appointments or renewals mentioned in an email that you shouldn't miss.

Notice that none of these require sending anything automatically. It alerts you and does the prep. You decide.

Connecting AI to your email means giving it access to sensitive information. Before you do, check a few basics:

• Permissions. Look at what the tool asks for: read, draft, send, delete. Grant only what's needed.
• Activity log. You should be able to see what it did and when. No history, no control.
• Approvals. Sensitive actions should go through you before they happen.
• Your AI provider. Find out which model it uses and what that provider's terms say about your data. Check their website.
• Team access. If several people use it, give each their own permissions.
• An exit plan. You should be able to disconnect it in a minute if something feels off.

And one common-sense rule: never share passwords by email, and never let an assistant act on credentials from a message it received.

Qiu is an AI coworker that takes action, not just answers questions. It connects to your email through Google and Microsoft 365 and works by the rules you set. See how its AI coworkers split the work by area.

What matters for staying in control:

• Approval before action. Sales emails, anything that moves money and anything an outside message asks for are shown to you first, with a Confirm button. More on approvals.
• Everything is logged. Every action is recorded, so you can see what it did and when.
• Runs 24/7. It can run routines and watches in the background, and picks up again after a restart. See how it runs 24/7.
• Memory. It remembers across chats, so you don't have to explain your tone every time.
• Your own AI. Connect your Claude or ChatGPT (or an API key). You pay your provider directly for AI usage.

To get going fast, there are templates like Unanswered emails every morning and Unanswered quotes every Monday. Browse them in templates.

You can talk to Qiu in the web app or on Telegram. Plans start at €29/month (Personal), with Pro at €99/month and Business at €299/month. No lock-in.