Answering emails with AI: how it works with Outlook and IMAP
An AI email assistant reads incoming customer emails, finds the answer in your own data and writes a draft. Whether it sends on its own is up to you. How it works with Outlook and IMAP, and what to watch for.
The short answer
Yes, you can have AI answer customer emails, and do it safely. What matters is that the AI answers only from your own data (website, knowledge base, shop data), can back every statement and, when in doubt, writes a draft for your team instead of sending. You connect your mailbox via Outlook / Microsoft 365 or IMAP and start with drafts.
What an AI email assistant does
- Read: a new customer email arrives in the connected mailbox.
- Search: the assistant looks for the answer in your sources, such as shipping and return rules, product data or, for a shop, the real order.
- Write: it drafts a reply in your team's tone.
- Check: every statement is checked against its source. Anything not backed stays out of the email.
- Decide: either a draft waits for approval, or the email goes out automatically if you allowed that for this mailbox.
Outlook, Microsoft 365 or IMAP?
Outlook and Microsoft 365 connect through the Microsoft sign-in. You grant access, nobody shares a password, and you can revoke access in your Microsoft account at any time.
IMAP and SMTP work with almost any other provider. You need server, username and password, with many providers a dedicated app password. Incoming mail arrives via IMAP, replies go out via SMTP from your own mailbox, so the customer gets them from your address as usual.
Three stages instead of all or nothing
Don't let the AI send on its own on day one. A start in three stages works well, as in the Alpheris mail agent:
- Observe: the AI learns from your resolved tickets in the background and sends nothing.
- Suggest: a checked draft is ready for every email. You read it and send with one click.
- Autopilot: the AI sends on its own, but only what passes every check. You switch it on per mailbox and off again at any time.
What the AI should never send on its own
- Complaints: an upset customer needs a human. In Alpheris, complaints never go out automatically.
- Promises without proof: "arrives tomorrow" only if the tracking status shows it.
- Legal and health topics: individual warranty questions or product effects belong to your team.
- Placeholders and half answers: if information is missing, a draft beats an email with a gap.
In Alpheris, a reply has to pass 20 checks before the autopilot sends it: every statement backed, every question answered, independent checkers agreeing, order questions only with live Shopify data. If one fails, it stays a draft with the reason. If the AI is unavailable, nothing is sent.
Get started in an hour
- Connect the mailbox (Outlook / Microsoft 365 or IMAP).
- Let it read your website and knowledge base; for Shopify, connect the store.
- Work in "Suggest" mode for a week and correct the drafts.
- Check which questions were always answered correctly (usually shipping, delivery time, returns, order status).
- Only then switch on the autopilot for this mailbox.
What about data protection?
Emails contain personal data. Look for processing in the EU, a data processing agreement and a promise that your data isn't used to train third-party models. Alpheris is hosted in Germany and the DPA is public. More in GDPR-compliant chatbots.
What does it cost?
In Alpheris, the mail agent is included from the Business plan (€79 per month excl. VAT, AI included, no per-seat pricing). You can try it free for 14 days without a payment method. All plans are on the pricing page.
