If you get to choose one place in your business to start, choose your email. Not because it's the most exciting place (it isn't), but because it's the place where you'll know within two weeks whether this works.

That's a bigger advantage than it sounds. An AI project without a yardstick can't show whether it works. With email you have that yardstick straight away: you see per draft whether it's right, and you know exactly what you would have written yourself.

01 / the definitionCan AI write my business emails?

Yes, in draft form. The system reads the incoming email, extracts the question, and writes a reply based on how you answered similar messages before. That draft is ready in your own email programme. You review it, adjust it and send it. The final word stays yours.

What it doesn't do is reply on its own. A reply to a client is a statement with your name under it, and that belongs in front of your eyes first. Why that boundary isn't an option but a design decision is explained in why we don't publish anything automatically.

02 / the choiceWhy start with email and not with content or sales?

Because the inbox combines three qualities that come together nowhere else. The volume is high: email arrives every day, unasked, whether you feel like it or not. The patterns are strong: most of your replies resemble earlier replies. And the yardstick is sharp: you see per draft within ten seconds whether it's on target.

Content and sales don't have those three. One blog post a week gives a system little practice material and you few reference points: you only find out in a quarter whether it did anything. Fifteen emails a day give both, straight away.

The inbox isn't the most spectacular place to start. It's the most instructive.

There's another reason, and it's less cheerful. Email is the kind of work that gets pushed into the evening: not because it's important, but because it's left over. That's exactly the kind of work a system should sit in: high in volume, low in judgement, and it eats the attention you wanted to save for something else.

03 / the sourceWhere does such a system get your tone from?

From two sources that complement each other. The first is your sent folder: it shows how you actually write, not how you think you write. The second is a conversation in which you explain what you do and why: that provides the reasoning that's only visible as an outcome in your emails.

That second step is often skipped, and that's where the weak spot of "read in my last two hundred emails" lies. Sent mail gives you tone. It doesn't give judgement. Your archive shows that you gave that one client a discount, not why you did that at the time and why you wouldn't do it for the next one. A system that only knows your archive copies your style and guesses your reasoning.

That's why we record it the other way round. Not with a questionnaire but with a recording:

1. Recording what you'd say anyway. An hour in which you explain how you work, what you sell to whom and where your boundaries lie. You don't need to prepare anything: you record what you'd say regardless.
2. Transcribing. A transcript turns a fleeting conversation into a text you can read back and organise.
3. Organising. Your way of working, your tone, your offering, your standard replies. Four drawers everything falls into.
4. Deduplicating. You say many things more than once. That repetition is exactly what shows what the core is. One clear version remains.

What you end up with isn't a folder of files but a knowledge layer: the place from which every draft is written afterwards. Your sent mail determines how it sounds, the conversation determines what's in it.

04 / the wordWhy don't we call this "self-learning"?

Because it doesn't learn on its own. In this category, "self-learning" has become a common word, and for this kind of system it isn't accurate. What happens is that the system improves through your corrections. Without corrections, nothing happens. "Self-learning" suggests it happens by itself, and that's exactly the expectation that makes people drop off.

The honest word is correctable. If you adjust the same phrasing three times, that adjustment should become the standard. That isn't magic and it isn't a mistake by the model either, but the way your judgement enters the system. The correction loop is the learning mechanism, and you are its engine.

That difference matters when you're choosing. A system that learns by itself doesn't need maintenance from you. A system that's correctable asks for a fixed moment each week where you talk or type into it what's changed about your offering, your target audience or your way of working. Without that moment, the knowledge layer quietly goes out of date, and you only notice once the drafts start to grate.

05 / the catchWhat goes wrong in the first weeks?

You correct a lot. Drafts are then too formal or too loose, they miss context that only you have, or they neatly answer the wrong question. Expect a few weeks in which reviewing sometimes takes as long as writing it yourself. That's not a malfunction, that's the period in which the system gets to know you.

This is the caveat a sales pitch would rather leave out, and it's the reason people drop off in week two. Whoever knows the dip in advance gets through it. Whoever buys a system that's supposed to be perfect from day one buys a disappointment with an invoice attached.

After that it tips over. The drafts get sharper, your role shrinks to reading, changing a sentence here and there, sending. When that turning point arrives depends on how much mail you have and how consistently you correct. We don't attach a number of weeks to it, because anyone who does can't actually know.

06 / honestlyWhat shouldn't be part of the promise here?

A few things, and we'd rather name them ourselves. This is precisely the point where a blog post can easily promise more than a supplier delivers.

• No hours promise. "Save eight hours a week" is a common promise framework in this market, and nobody can know it in advance. It depends on your email volume and on how similar your replies genuinely are to each other.
• No replacement of your judgement. The system suggests what you'd probably reply. Whether you give that discount, take on that job or have that conversation remains your work.
• No fix for an unclear offering. If you yourself aren't clear about what you sell to whom, a system will only speed up that unclarity. That's not a reason to wait, but it is a reason to look at that first.
• No guarantee that every message is suitable. Email with a lot of context that only exists in your head, or a sensitive matter where tone determines everything, remains manual work. The system is good at the pattern, not the exception, and it should show you which of the two it thinks it's dealing with.

For us, the mailbox falls under the Back Office: one of the fixed roles in the system, not a separate project bolted on. So what you get isn't a mail assistant but one crew member set up on your knowledge layer, alongside the roles carrying your content, your meeting notes and your lead follow-up.

The mailbox remains yours throughout. The system works in your account, not in an environment of ours that you get a login for. If you cancel, you keep your mail, your archive and everything that's happened in it.

This text was produced with AI assistance and checked and approved by a human before publication.