Send segmented newsletters
and above all decide who does not get them.

01The task, as it is done today

Take a shop somewhere between Arlon and Differdange. The monthly newsletter goes out on a Friday afternoon. You reopen last month’s campaign, swap the image and the three featured products, tick the whole list, and send. On Monday, two clients write to say they bought the same item at full price a fortnight ago. A third person unsubscribes for the second time.

The trouble is not what the letter said. It is who it went to.

And the list itself is barely a list any more. It grew in layers: orders from the shop, records from the till, a spreadsheet filled in at a trade fair. The same person appears twice under two addresses, and a few of them never said yes to anything. A service business has the same problem without having sold a single item: its history is made of quotes, appointments and exchanges rather than purchases, and it sits untouched in the client records.

02Can AI personalise my newsletters?

It writes one subject line and one text per group of clients, based on what that group did with you, but it never sees your list, it does not decide who receives, and it sends nothing.

Your sending tool already offers this, inside the subscription, in three clicks: generated subject lines, predictive segments, optimal send time. What it does not do is check that the person it files under “loyal clients” actually said yes, that they did not buy the item the day before yesterday, and that the price quoted is still the price.

The full circuit, as it is built in a company of three to fifty people, has twenty-four stations. At five places it hands control back to a person, and only three of the twenty-four stations involve a model.

03From a dormant list to a campaign someone has read

Eight zones, and a final node where nothing goes out until someone has pressed send. Click a zone to open its stations; otherwise, the sequence plays on its own.

04The twenty-four stations, and who does what

The circuit unfolded, station by station, with whoever owns each one. The writing takes up a single zone, the fifth, between what prepares it and what checks it.

05What the AI does, and what it never does

Three stations out of twenty-four. The model judges whether your history says something about a client or merely that they came once, it writes one subject line and one body per segment, and it flags what it is unsure about. On the very task every tool claims a model will write for you, that is little. It is not that the writing matters little: it is that the writing is precisely what your subscription already gives you, and that everything which makes a campaign fit to send is still to be done.

Three hard limits, the same for every campaign. It never sees your list: it receives a segment’s description, its size and this month’s offer, never a name, never an address, never one person’s history. What never goes into the model cannot come out of it. It does not decide who receives: the sorting follows rules you wrote, each one readable on its own, and segments are not discovered by themselves. It sends nothing: the campaign stops at a summary, and a person presses send in your own sending tool.

This circuit writes to people who said yes to you, and to them only. Cold email, to bought or harvested addresses, is not what this page describes: it is another trade, with other rules and other risks.

Besides Claude and Make, the circuit uses what you already have. Your shop, Shopify or WooCommerce. Your client records or CRM, HubSpot, Odoo, Pipedrive or Zoho. A spreadsheet, or a base such as Airtable. Your sending tool, Brevo, Mailchimp or the one inside your CRM. A tool not named here connects in the same way. Two limits are worth knowing before you start. A till with no client accounts ties its receipts to nobody: there is then nothing to segment, and that is not a setting, it is a prerequisite. And authenticating your domain is done in your DNS zone, not in the sending tool: it is what breaks most often, and it assumes access you do not always have to hand.

06What cannot be bought, and what holds everything together

Four pieces that describe your business, and that no subscription supplies. What counts as consent at your end: who said yes, when, where and to what, and the record that proves it. Without it a contact cannot be emailed, and half an inherited list turns out to be unusable the day you ask for that record. Your email template and its voice: the structure, the tone, three or four campaigns that worked as examples, and what the header and footer always carry. Your sending rules: what is never used to sort your clients, how many messages per person per month, what is never said to a client even when it is true, how long campaign data is kept. And your segmentation rules: sentences you can check, not hunches.

That is where segmentation parts company with a mail merge. “Ordered in the last twelve months in this product family” is a rule: you know who it covers, and you know how to correct it. “High-potential clients” is not: nobody can say who it applies to, and nobody will be able to tell a client why they received that message.

To those add a piece the tools do not have: the segment that does not receive. Whoever bought the day before yesterday, whoever already had this offer, whoever is in dispute, whoever has reached their cap for the month. It is decided campaign by campaign, it shows up in no dashboard, and it is what protects the relationship. Segmenting starts with deciding who you do not write to.

07What can go wrong

Four failures. One comes from the list, one from the material, one from the person doing the final read, and one from time passing.

Someone who opted out receives anyway. They had come off one campaign, but their address was still sitting in another export, and the next month it turns up in their inbox again. They do not unsubscribe a second time; they mark you as spam. Hence the removal that applies everywhere and at once, to the whole list, not to the export that was used.

An expired offer goes to everyone. The price changed on Tuesday, the campaign went out on Thursday with the old one, and two thousand people read it. That is why every price, date and condition in the message has to be traced back to this month’s offer before anything leaves. The check confirms the figure exists and matches. It does not judge whether the message is any good; the final read decides that.

The final read becomes a glance. After three good campaigns, people skim the summary, then press send without reading it. That is the human failure, and no check removes it. What limits it: the summary arrives with the doubtful points at the top, a segment of three people, a segment that has doubled, a text too close to the last one. And sending goes in waves, with a stop button that was tried once before anyone needed it.

The list ages, and sending suffers for it. Addresses that no longer answer build up, along with a few spam reports, and mail providers take that into account for whatever leaves your domain next. This is maintenance, not an accident: dead addresses are removed at every campaign, reports are watched, and sending in waves leaves time to stop before it settles in.

08How long it takes, and what it costs

This page gives no timescale, and that is deliberate. The work depends less on the scenario than on the state of your list: kept in one place or spread across three, consent on record or nowhere to be found, a history tied to client accounts or to anonymous receipts. An export of your file would tell me in an hour, and I have not seen it yet. The audit sets the timescale, and the price with it: it is free because, before it, I cannot put a figure on anything.

We start with a single campaign, three or four segments, and nothing else until that one holds. The frequently asked questions cover your data, the cost and the timing, and my page says where I come from.

One last point, because it often decides what happens next. What stops a newsletter is almost never technical. You manage two issues, then you have nothing left to say, and the rhythm dies. There is a way to guard against it: building a stock of subjects out of what you already do, your projects, your deliveries, the questions that keep coming back. That is a different piece of work from the one described here. We talk about it during the audit, not after the first campaign.