We gave our inbox to an AI. Here's what happened.
Every business we work with has the same inbox. One address — info@, sales@, support@ — where everything lands: real enquiries, supplier invoices, complaints, newsletters, and an ocean of spam. Someone checks it "when they get a chance". Enquiries sit unanswered for days. And the ones that do get answered never make it into any system — they live and die in the mailbox.
Ours was no different. Our website's contact form started collecting convincing, human-typed spam — money-mule scams dressed up as business enquiries, SEO pitches with a friendly name attached. Filters that catch robots don't catch humans. So we did what we do for clients: we built software for it.
Meet MailBridge
MailBridge is a module we built into Vali, our business platform. The idea is simple to say and was very satisfying to build: AI checkers that read a mailbox the way a good assistant would — each with its own instructions, schedule and job.
- The firewall checker reads every incoming mail first and quarantines the junk — including the human-typed kind, because it reads the content, not just the headers. Nothing is deleted; every decision is logged with its reason.
- The responder checkers answer routine mail — an AI-drafted reply in our voice, following our written instructions, sent from the same mailbox. Enquiries get acknowledged in minutes, not days.
- The alert checkers forward what matters to the right person, with the original mail intact, and raise an in-app notification so nothing urgent sits unseen.
The part that surprised us: the inbox is a lead machine
The checker we now consider the most valuable does something else entirely. When an enquiry comes in, it extracts the actual person — name, email, phone, company, and what they asked about — and creates a contact with that interest recorded.
Real example from our own testing: an enquiry arrived from a website form robot (the sender was just accounts@...). Buried in the body was the real customer — name, direct email, phone number, and what he wanted to buy. MailBridge pulled him out and filed him as a lead with "pricing for a CNC router bundle" attached. No human read that email first.
Do that for every enquiry, for months, and your inbox quietly becomes something most small businesses never have: a clean list of real people and exactly what they wanted.
And then you can act on it
That list feeds the next piece we built: ask in plain English — "everyone who asked about routers this year" — and the AI builds the audience from captured interests and purchase history. Draft a campaign from a short brief, preview it, approve it, and it trickle-sends from your own mailbox at a controlled rate, unsubscribe link included, opens tracked.
Enquiry → answered → captured → followed up. The whole loop, without anyone retyping anything.
The honest bits
Three things we'd tell any client considering this:
- Keep a human in the loop at first. Our checkers started in watch-and-notify mode; auto-replies came after we trusted the decisions. The audit log — every action, every reason — is what makes that trust possible.
- AI reads content, so it catches what filters miss — but it costs a little per email (ours runs at roughly a cent a mail) and it needs decent instructions, the same way a new assistant needs a proper handover.
- Fail open. If anything errs, the mail stays exactly where it was. An assistant that occasionally does nothing is fine; one that loses mail is fired.
Where to see it
MailBridge and the marketing tools are release-candidate features in Vali — running on our own inboxes today, exactly the way we'd run them on yours.
Got an inbox that eats enquiries? Tell us about it — we'll be straight about whether this fits your business, and what your rules would need to say.