● B2B offer — Marketing & distribution
Newsletter generation: your content, your template
The newsletter is the appointment everyone puts off: selecting, summarising and laying out takes time, and the deadline always comes round too soon. Your agent gathers the period's content, proposes a selection against your editorial criteria, writes the hooks and puts the send together in your template. Hosted in France: your unpublished content and your subscriber list stay with you. The editorial lead settles the selection and triggers the send.
Hosted in France
Subscribers and content protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Newsletter · hosted in France
Prepare this month's newsletter.
Selection proposed from the content published in the period, against your editorial criteria: product news, in-depth articles, sector news.
Hooks written in your tone, layout applied to your template.
The order of sections follows your usual structure; nothing has been sent.
✎ Action · send prepared, the call is yours
Hooks written in your tone, layout applied to your template.
The order of sections follows your usual structure; nothing has been sent.
✎ Action · send prepared, the call is yours
Move the in-depth article to the top and drop the third short item.
New version with the in-depth article opening and the short item removed.
The lead hook is rewritten to match the new top item; the rest of the send is unchanged.
✎ Action · version updated, ready to read over
The lead hook is rewritten to match the new top item; the rest of the send is unchanged.
✎ Action · version updated, ready to read over
Local inference · no data outside the EU
Distribution hosted in France
Sovereign by designLocal inference or hosting in France
GDPR & AI Act: governed deploymentTraceability & human oversight
TurnkeyDesigned, installed and operated for you
The editorial lead decidesThe agent prepares, never rules
✦ In brief
A Blue Lemon Agent newsletter agent gathers the period's content, proposes a selection against your editorial criteria, writes the hooks in your tone and puts the send together in your template. It distributes nothing: settling the selection and triggering the send stay with the editorial lead. It runs on local inference or is hosted in France: your unpublished content and your subscriber list stay with you, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
6
editorial uses ready to deploy
0
decision taken without human approval
These figures describe our offer, not results measured at a client. How large the gain is on the frequency and richness of your sends is confirmed by a pilot.
The context
What does an AI agent bring to your newsletters?
Most of the work on a newsletter is gathering and laying out. Automating that part makes the deadline sustainable and turns the appointment with your subscribers into a regular one.
! The issue
A newsletter lives on its regularity: that is what establishes the appointment. Yet most of the effort goes into collecting the content and laying it out, not into the editorial choice. The agent takes on that part — selection proposed against your criteria, hooks written, template applied — so that the deadline becomes sustainable.
✓ Our answer
You keep the editorial line: the agent proposes a selection from content already published, you settle it, and it is you who triggers the send. Local inference or an isolated resource hosted in France: your subscriber list — both personal data and a commercial asset — is never entrusted to a third-party service.
The decisive point
Your unpublished content and your subscriber data: sovereignty & compliance
Your subscriber list is both personal data and a commercial asset. Here is how the architecture of our agents protects it.
Local inference
The agent can run on a machine belonging to your organisation: no subscriber address leaves the network, no unpublished content passes through a public cloud.
Hosting in France
Otherwise, a dedicated and isolated resource hosted in France, under French law — your content and your mailing list: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
For your unpublished content and your subscriber data, the architecture aims to reduce exposure to the Cloud Act and FISA 702; being located in France or in the European Union does not, on its own, guarantee immunity.
Isolated resource
No pooling: an environment strictly dedicated to your brand and its mailing list.
Template and editorial structure respected
Your template, the order of your sections and your tone are applied as they stand; encryption, role-based access and logging of every send prepared.
AI Act: governed deployment
The agent is strictly in support; no distribution is triggered automatically; traceability and human oversight from end to end.
What depends on the architecture chosen These points are not general guarantees: they are settled deployment by deployment, in the quotation.
- The applicable location is that of the architecture set out in the quotation and verified before commissioning.
- Local execution is announced only for the configuration explicitly described and accepted in the quotation.
- The applicable isolation depends on the deployment mode set out in the quotation; no dedicated isolation is presumed.
- Roles and permissions are configured and accepted for the identities and systems actually connected.
- The events logged, their content, their retention period and who may access them are defined for the deployment chosen.
For sends tied to an announcement under embargo or to a financial transaction, SecNumCloud and reinforced hosting are options depending on your requirements. A single architecture is designed to answer both the GDPR and extraterritorial exposure. Designed for deployment in line with the GDPR and the AI Act, after the processing, roles and context-specific risks have been assessed.
Demonstration
See the agent at work
5 real situations, taken from those that come up most often. Pick one: the exchange unfolds as it would in your organisation.
A scripted demonstration. These exchanges show how the agent behaves — its sources, its refusals, what it leaves to your teams. Nothing is sent from this page, no model is queried here, and the matters named are fictional. That is precisely what we promise your data.
The behaviours shown here — monitoring, automation rules, routing and reminders — are configured with you during deployment, from your tools, your rules and your thresholds.
The architecture points named in these exchanges — location, local execution, isolation, encryption, role-based access, logging — are not a guarantee attached to the demonstration: they are those of the architecture set out in your quotation, and verified before commissioning.
The company in this demonstration
Fictional companyAlveor — French office furniture manufacturer
- Sector
- Office furniture manufacturing and sales — companies, local government and an online shop
- Headcount
- 96 staff, including 3 in marketing: an editorial lead, a communications officer and an apprentice
- Subscribers
- 14,200 newsletter subscribers — 9,800 professionals and public bodies, 4,400 consumers from the online shop
- Cadence
- 11 issues planned a year, 6 actually sent in 2025; 486 pieces of content published in three years across the site and the catalogue
- Tools in place
- Website CMS, editorial drive, product catalogue and sending platform — the agent plugs into them, nothing is replaced
- Who decides
- The editorial lead settles the selection and triggers the send; marketing management sets the cadence; the data officer rules on list questions
- Room for improvement
- 6 issues sent out of 11 planned; 9 h 40 of preparation per issue, most of it collection and layout; three published items out of four never make it into the newsletter
Alveor does not write too little: it publishes 13 items a month and circulates almost none of them. What it wants is to keep the appointment — eleven issues a year instead of six — without adding a person to marketing. The agent runs on an isolated resource hosted in France and plugs into the CMS, the editorial drive, the catalogue and the sending platform: it composes the whole issue, the editorial lead settles it and triggers the send. The exchanges below cover one quarter, from the first pass over the archives to the review of three issues.
This company, its figures and the exchanges that follow were invented for the demonstration. They illustrate a common situation; they describe no real client.
Newsletter · hosted in France
I read three years of your publishing last night — 486 items — and your last six issues. From them I drew your editorial criteria: they existed, they were written down nowhere.
Editorial criteria are the rules that decide what goes into an issue: which section, how old an item may be, what balance between product and substance. Your editorial lead has applied them from memory for four years.
The gap I measured, and it is the gap that decides the gain: of 486 published items, 365 have never appeared in the newsletter — three out of four. This is not an editorial judgement problem: the six issues that went out were assembled from whatever was to hand in the last two days. You are not selecting from your output, you are selecting from your short-term memory.
The five criteria I read in your approved issues:
· One feature article per issue, never two — all six issues, no exception.
· Two catalogue launches at most; beyond that, your own history shows the newsletter's click rate drops by a third.
· One customer story whenever one is less than three months old — you did it five times out of six.
· Nothing older than 90 days, unless dated as updated.
· No item already circulated in a previous issue.
What that gives you from the next issue on: the selection is no longer made from the 40 items anyone remembers, but from the 486 you actually published.
What I propose: that the editorial lead reads these five criteria over and corrects the thresholds — each one changes with a single figure. Once they are settled, I compose the October issue and you take it to marketing management as it stands. editorial-criteria_5-rules-read-in-your-issues.pdf5 criteria, each with the issue that proves it
⛓ Sourced · website CMS, product catalogue, 6 issues sent in 2025, sending log
Editorial criteria are the rules that decide what goes into an issue: which section, how old an item may be, what balance between product and substance. Your editorial lead has applied them from memory for four years.
The gap I measured, and it is the gap that decides the gain: of 486 published items, 365 have never appeared in the newsletter — three out of four. This is not an editorial judgement problem: the six issues that went out were assembled from whatever was to hand in the last two days. You are not selecting from your output, you are selecting from your short-term memory.
The five criteria I read in your approved issues:
· One feature article per issue, never two — all six issues, no exception.
· Two catalogue launches at most; beyond that, your own history shows the newsletter's click rate drops by a third.
· One customer story whenever one is less than three months old — you did it five times out of six.
· Nothing older than 90 days, unless dated as updated.
· No item already circulated in a previous issue.
What that gives you from the next issue on: the selection is no longer made from the 40 items anyone remembers, but from the 486 you actually published.
What I propose: that the editorial lead reads these five criteria over and corrects the thresholds — each one changes with a single figure. Once they are settled, I compose the October issue and you take it to marketing management as it stands. editorial-criteria_5-rules-read-in-your-issues.pdf5 criteria, each with the issue that proves it
⛓ Sourced · website CMS, product catalogue, 6 issues sent in 2025, sending log
Our content is scattered: the CMS, a shared drive, the product catalogue. Where will you look?
In all three, and I reconcile them myself. For the October issue the period runs from 1 to 30 September: 25 candidate items, all dated and sourced.
The period is the stretch an issue covers. Here it runs from one send to the next, and it is what makes an item « new ».
What I gathered:
· 17 items from the CMS — articles, news, updated advice pages.
· 6 catalogue launches, with their image and technical sheet.
· 2 customer stories on the drive, one of them never published anywhere.
What I set aside before showing you, and why: 4 duplicates — the same item published on the site and as catalogue news —, 2 pages updated with no change of substance, and 1 launch whose availability slipped to 15 November. That one I hold for the next issue rather than drop.
What that shifts: collecting the period's content used to take 45 % of the time of an issue. It now takes 10 %, and those 10 % are your checking, not your searching.
And one find I did not expect: 14 items published more than three months ago, never circulated in a newsletter, and still current — four layout guides, six customer stories, four pages on standards. That is two issues in reserve, available this morning. They do not replace the period's news, they fill the thin months — and your thin months are exactly the ones where the issue never went out.
What I propose: that I keep that reserve current and release it automatically whenever a period has fewer than six candidates. You will no longer choose between a thin issue and no issue at all. period-content_25-candidates-and-14-in-reserve.pdf25 dated candidates, 7 set aside, 14 in reserve
⛓ Sourced · CMS, editorial drive, product catalogue, log of circulated issues
The period is the stretch an issue covers. Here it runs from one send to the next, and it is what makes an item « new ».
What I gathered:
· 17 items from the CMS — articles, news, updated advice pages.
· 6 catalogue launches, with their image and technical sheet.
· 2 customer stories on the drive, one of them never published anywhere.
What I set aside before showing you, and why: 4 duplicates — the same item published on the site and as catalogue news —, 2 pages updated with no change of substance, and 1 launch whose availability slipped to 15 November. That one I hold for the next issue rather than drop.
What that shifts: collecting the period's content used to take 45 % of the time of an issue. It now takes 10 %, and those 10 % are your checking, not your searching.
And one find I did not expect: 14 items published more than three months ago, never circulated in a newsletter, and still current — four layout guides, six customer stories, four pages on standards. That is two issues in reserve, available this morning. They do not replace the period's news, they fill the thin months — and your thin months are exactly the ones where the issue never went out.
What I propose: that I keep that reserve current and release it automatically whenever a period has fewer than six candidates. You will no longer choose between a thin issue and no issue at all. period-content_25-candidates-and-14-in-reserve.pdf25 dated candidates, 7 set aside, 14 in reserve
⛓ Sourced · CMS, editorial drive, product catalogue, log of circulated issues
And our tone? Our running order of sections? Nobody has ever written them down.
They are written down this morning, and they come out of your six approved issues, not out of a hunch.
Your running order, unchanged since 2023: four-line editorial · feature article · two launches · one customer story · three briefs and the diary. You never changed it, not even in the months when the issue went out late — so it is a rule, not a habit.
Your tone, measured on your approved copy: sentences of 16 words on average, consistent use of « you », active voice in 8 sentences out of 10, never an exclamation mark, and three words you never use: « revolutionary », « must-have », « solution ». The last is the most telling — your competitors use it in two headlines out of three.
Now the hooks — a hook is the two or three lines that introduce an item in the newsletter and earn the click; it does not summarise the article, it gives the reason to read it. Yours follow a constant pattern: one figure, one consequence for the reader, one infinitive verb in the link. That is the pattern I apply.
What it changes in production: a hook arrives already in your voice. Your communications officer no longer rewrites to re-tune a voice, she approves or she rules — and that is the only work worth her time.
The benchmark to judge by: across your six issues, writing the hooks took 35 % of the time. It now takes 12 %, review included.
What I propose: approve the running order and the three banned words, and I compose the October issue today. The guide changes with a single word at any time: it is your material, I only wrote it down. running-order-and-tone_alveor.pdf5 sections, 16 words per sentence, 3 banned words
⛓ Sourced · 6 approved 2025 issues, website pages, review threads on the drive
Your running order, unchanged since 2023: four-line editorial · feature article · two launches · one customer story · three briefs and the diary. You never changed it, not even in the months when the issue went out late — so it is a rule, not a habit.
Your tone, measured on your approved copy: sentences of 16 words on average, consistent use of « you », active voice in 8 sentences out of 10, never an exclamation mark, and three words you never use: « revolutionary », « must-have », « solution ». The last is the most telling — your competitors use it in two headlines out of three.
Now the hooks — a hook is the two or three lines that introduce an item in the newsletter and earn the click; it does not summarise the article, it gives the reason to read it. Yours follow a constant pattern: one figure, one consequence for the reader, one infinitive verb in the link. That is the pattern I apply.
What it changes in production: a hook arrives already in your voice. Your communications officer no longer rewrites to re-tune a voice, she approves or she rules — and that is the only work worth her time.
The benchmark to judge by: across your six issues, writing the hooks took 35 % of the time. It now takes 12 %, review included.
What I propose: approve the running order and the three banned words, and I compose the October issue today. The guide changes with a single word at any time: it is your material, I only wrote it down. running-order-and-tone_alveor.pdf5 sections, 16 words per sentence, 3 banned words
⛓ Sourced · 6 approved 2025 issues, website pages, review threads on the drive
Put this month's newsletter together.
The October issue is laid out in your template, from subject line to diary. Nine items chosen out of 25 candidates, and each carries the criterion that let it in.
The selection, section by section:
· Feature article — « What the new seating standard changes for your workstations ». Chosen because it is the period's only substantive piece and because it drew 2.4 times the site's average reading time.
· Two launches — the Lisière range and the Adret pedestal. Chosen because both are available to order, and not three, because your own history says the third costs clicks to everything else.
· One customer story — the Vaulnay technical centre fit-out, filed on the drive and never published. That is the item you were about to miss.
· Three briefs and the diary — including your stand at the public-sector show on 12 November.
The hooks are written, nine of them, in your voice, each with the link to its page.
So are the subject line and the preheader — the subject line is what people read in the inbox; the preheader is the grey line that follows it, and almost nobody writes it. I am giving you two versions of each, built to be compared.
The template is applied: your header, your colours, your section order, and the rendering checked on three displays. 62 % of your opens are on a phone — two hooks ran past three lines on a handset, so I tightened them.
The time: your timesheets show 9 h 40 per issue. This one landed on your desk 40 minutes after your message, and laying out the template, which took 30 % of the time of an issue, now takes 7 %.
Nothing has been sent, and nothing will be without you. What I propose next: settle the selection, then choose the subject line to test. Triggering the send stays your act. october-issue_selection-hooks-template.pdf9 items chosen, 9 hooks, 2 subject lines to compare
⛓ Sourced · 25 period items, approved editorial criteria, template and display statistics from your 6 issues
The selection, section by section:
· Feature article — « What the new seating standard changes for your workstations ». Chosen because it is the period's only substantive piece and because it drew 2.4 times the site's average reading time.
· Two launches — the Lisière range and the Adret pedestal. Chosen because both are available to order, and not three, because your own history says the third costs clicks to everything else.
· One customer story — the Vaulnay technical centre fit-out, filed on the drive and never published. That is the item you were about to miss.
· Three briefs and the diary — including your stand at the public-sector show on 12 November.
The hooks are written, nine of them, in your voice, each with the link to its page.
So are the subject line and the preheader — the subject line is what people read in the inbox; the preheader is the grey line that follows it, and almost nobody writes it. I am giving you two versions of each, built to be compared.
The template is applied: your header, your colours, your section order, and the rendering checked on three displays. 62 % of your opens are on a phone — two hooks ran past three lines on a handset, so I tightened them.
The time: your timesheets show 9 h 40 per issue. This one landed on your desk 40 minutes after your message, and laying out the template, which took 30 % of the time of an issue, now takes 7 %.
Nothing has been sent, and nothing will be without you. What I propose next: settle the selection, then choose the subject line to test. Triggering the send stays your act. october-issue_selection-hooks-template.pdf9 items chosen, 9 hooks, 2 subject lines to compare
⛓ Sourced · 25 period items, approved editorial criteria, template and display statistics from your 6 issues
Move the feature article to the top and drop the third brief.
Done, and I have reworked the three things that move makes wrong.
· The feature article now opens the issue, the editorial drops below it at three lines instead of four.
· The lead hook is rewritten: it announced the launches, it now announces the seating standard and what it changes for a public buyer.
· The subject line follows — it echoed the old top item, so it would have promised something the newsletter no longer shows. A subject line that does not keep its promise costs more than a dull one.
· The third brief is out, the contents list recalculated, and the rest of the send is strictly unchanged — same hooks, same links, same template.
What it cost you: four minutes. The same edit by hand, across your six 2025 issues, took between 40 and 50 minutes, because it forces you to reopen the subject line, the contents and the mobile rendering one by one — and because one of them always gets missed.
What I propose next, and this is where the newsletter starts paying: your two subject lines are ready to be compared on 10 % of the list, 1,420 subscribers, sent at the same moment. Two hours later the winner goes to the rest. Across your six issues, the gap between the best and the worst subject line was 9 points of open rate — it is the most profitable act of the whole preparation, and it costs you nothing extra.
Approve the test and I will set it up; otherwise I go with the line you prefer, no argument.
· The feature article now opens the issue, the editorial drops below it at three lines instead of four.
· The lead hook is rewritten: it announced the launches, it now announces the seating standard and what it changes for a public buyer.
· The subject line follows — it echoed the old top item, so it would have promised something the newsletter no longer shows. A subject line that does not keep its promise costs more than a dull one.
· The third brief is out, the contents list recalculated, and the rest of the send is strictly unchanged — same hooks, same links, same template.
What it cost you: four minutes. The same edit by hand, across your six 2025 issues, took between 40 and 50 minutes, because it forces you to reopen the subject line, the contents and the mobile rendering one by one — and because one of them always gets missed.
What I propose next, and this is where the newsletter starts paying: your two subject lines are ready to be compared on 10 % of the list, 1,420 subscribers, sent at the same moment. Two hours later the winner goes to the rest. Across your six issues, the gap between the best and the worst subject line was 9 points of open rate — it is the most profitable act of the whole preparation, and it costs you nothing extra.
Approve the test and I will set it up; otherwise I go with the line you prefer, no argument.
We would also like an English version for our Belgian and Dutch customers, and a LinkedIn post built on the feature article.
Both are done, by two dedicated agents I hand over to — and I hand them material that is already settled, which is the whole point.
Translation: the translation agent picks up the version you have approved, not the draft. It renders the whole issue — subject line, preheader, nine hooks, diary —, keeps the template and the links, and switches links to the English pages where they exist. For the three items with no English page, it tells you rather than pointing at a French page, and I propose replacing them with two items from the reserve.
Social: the social media agent draws the copy for one LinkedIn post and two follow-ups, ready for review, from the feature article, in the format of your approved posts. It does not publish in LinkedIn and does not follow up automatically: your community manager is the one who puts the post live and follows up personally, at the moment they choose. It does not rewrite the newsletter: it produces a stand-alone piece, because a post that says « read our newsletter » does not get read.
Why three agents and not one: each is plugged into a different place — the sending platform, the multilingual site, your social account — and each is approved by the person who answers for it in your company. What I guarantee is the hand-over: all three start from the version you settled, not from three different drafts.
What that does to the delay: the English version and the post historically went out four to six days after the French issue, when they went out at all. They are ready the same day, with the same lead item.
What I propose: approve the French issue and I trigger both hand-overs straight away. You will read three consistent documents the same afternoon. hand-over_translation-and-social.pdfWho picks up what, from which version
✎ Framework · hand-over between agents — each starts from the approved version, each is approved by its owner
Translation: the translation agent picks up the version you have approved, not the draft. It renders the whole issue — subject line, preheader, nine hooks, diary —, keeps the template and the links, and switches links to the English pages where they exist. For the three items with no English page, it tells you rather than pointing at a French page, and I propose replacing them with two items from the reserve.
Social: the social media agent draws the copy for one LinkedIn post and two follow-ups, ready for review, from the feature article, in the format of your approved posts. It does not publish in LinkedIn and does not follow up automatically: your community manager is the one who puts the post live and follows up personally, at the moment they choose. It does not rewrite the newsletter: it produces a stand-alone piece, because a post that says « read our newsletter » does not get read.
Why three agents and not one: each is plugged into a different place — the sending platform, the multilingual site, your social account — and each is approved by the person who answers for it in your company. What I guarantee is the hand-over: all three start from the version you settled, not from three different drafts.
What that does to the delay: the English version and the post historically went out four to six days after the French issue, when they went out at all. They are ready the same day, with the same lead item.
What I propose: approve the French issue and I trigger both hand-overs straight away. You will read three consistent documents the same afternoon. hand-over_translation-and-social.pdfWho picks up what, from which version
✎ Framework · hand-over between agents — each starts from the approved version, each is approved by its owner
Our list mixes professionals and consumers who came through the shop. Can we really send everything to everyone?
Yes, to 13,760 of the 14,200 — and for the remaining 440 I have prepared the route that brings them back rather than writing them off.
Your three origins, recovered line by line:
· 7,400 professionals signed up through the website form, with the date and the page. Clean origin.
· 2,400 professionals and public bodies met at trade shows or under contract. The rule, in plain terms: an approach addressed to someone in their professional capacity, about that profession, can rest on your legitimate interest — provided they know where you got the address and can object at any time. That is exactly what I put into every send.
· 4,400 shop customers, of whom 3,960 ticked the newsletter box when they ordered, with the date and page recorded.
The remaining 440: shop customers whose order predates the box, introduced in February 2024. For a consumer, the newsletter requires their agreement — so I did not put them into the October issue.
What I did instead of losing them: I prepared a one-click sign-up page and a line of invitation to slot into the order confirmation e-mail, which goes out anyway. No extra send, no additional approach: the invitation travels inside a message they are expecting.
What I attach to every send without your having to think about it: your identity and postal address, the one-click unsubscribe link at the top and the foot, and where the address came from — « you are receiving this because… », with the real reason, not a formula.
What it gives you back: checking origins took half a day before every send, and you had stopped doing it. It is now done continuously, and the number of people excluded is always shown before you approve.
What I propose: put the invitation line live this week. Over the first six weeks I will tell you how many of the 440 came back of their own accord — and those will be on the list by their own act, which beats any import. subscribers_3-origins-and-440-regularisations.pdfEvery line with its date, page and origin
⛓ Sourced · sign-up register, website form log, shop orders, trade show lists
Your three origins, recovered line by line:
· 7,400 professionals signed up through the website form, with the date and the page. Clean origin.
· 2,400 professionals and public bodies met at trade shows or under contract. The rule, in plain terms: an approach addressed to someone in their professional capacity, about that profession, can rest on your legitimate interest — provided they know where you got the address and can object at any time. That is exactly what I put into every send.
· 4,400 shop customers, of whom 3,960 ticked the newsletter box when they ordered, with the date and page recorded.
The remaining 440: shop customers whose order predates the box, introduced in February 2024. For a consumer, the newsletter requires their agreement — so I did not put them into the October issue.
What I did instead of losing them: I prepared a one-click sign-up page and a line of invitation to slot into the order confirmation e-mail, which goes out anyway. No extra send, no additional approach: the invitation travels inside a message they are expecting.
What I attach to every send without your having to think about it: your identity and postal address, the one-click unsubscribe link at the top and the foot, and where the address came from — « you are receiving this because… », with the real reason, not a formula.
What it gives you back: checking origins took half a day before every send, and you had stopped doing it. It is now done continuously, and the number of people excluded is always shown before you approve.
What I propose: put the invitation line live this week. Over the first six weeks I will tell you how many of the 440 came back of their own accord — and those will be on the list by their own act, which beats any import. subscribers_3-origins-and-440-regularisations.pdfEvery line with its date, page and origin
⛓ Sourced · sign-up register, website form log, shop orders, trade show lists
And in practice, what gets checked just before I press « send »?
Seven checks, run in twelve seconds, and I can tell you what each one stopped across your last three issues.
· No unsubscribed address in the list — 1 send held: a re-imported trade show file contained 38 addresses that had unsubscribed in 2024.
· Every segment has a recorded origin — 1 send held, the 440.
· The unsubscribe link answers — tested before every send, 1 alert on 12 November, fixed in 20 minutes before departure.
· Nobody receives the same issue twice — 210 double deliveries avoided over the quarter, people on two lists under two addresses.
· Every link in the issue answers — 4 links fixed, one of them to a product page withdrawn the day before.
· The rendering holds on mobile — 7 hooks tightened.
· The sender is authenticated and the issue clears reputation filters — 0 blocks, and it is the check you only notice the day it is missing.
Result over the quarter: 3 issues out on the planned date, 2 sends held for a few hours, 0 cancelled. A held send costs a morning; a send that reaches someone who had unsubscribed costs trust, and the apology would itself be one more e-mail.
Unsubscribing, since it is the act that matters most: it is applied within the second, to every future send. And the reverse is true as well: if the person fills in your form again, I put them back on the list automatically, with the date and page on record. Three subscribers came back that way this quarter — unsubscribing is not a door closed for life.
What I propose: that the report of these seven checks appears above the send button in your platform. You will click knowing what has been verified, and the data officer will no longer have to ask. seven-checks-before-sending.pdfWhat each check stopped this quarter
⛓ Sourced · check log for the last 3 issues, unsubscribe log, link tests
· No unsubscribed address in the list — 1 send held: a re-imported trade show file contained 38 addresses that had unsubscribed in 2024.
· Every segment has a recorded origin — 1 send held, the 440.
· The unsubscribe link answers — tested before every send, 1 alert on 12 November, fixed in 20 minutes before departure.
· Nobody receives the same issue twice — 210 double deliveries avoided over the quarter, people on two lists under two addresses.
· Every link in the issue answers — 4 links fixed, one of them to a product page withdrawn the day before.
· The rendering holds on mobile — 7 hooks tightened.
· The sender is authenticated and the issue clears reputation filters — 0 blocks, and it is the check you only notice the day it is missing.
Result over the quarter: 3 issues out on the planned date, 2 sends held for a few hours, 0 cancelled. A held send costs a morning; a send that reaches someone who had unsubscribed costs trust, and the apology would itself be one more e-mail.
Unsubscribing, since it is the act that matters most: it is applied within the second, to every future send. And the reverse is true as well: if the person fills in your form again, I put them back on the list automatically, with the date and page on record. Three subscribers came back that way this quarter — unsubscribing is not a door closed for life.
What I propose: that the report of these seven checks appears above the send button in your platform. You will click knowing what has been verified, and the data officer will no longer have to ask. seven-checks-before-sending.pdfWhat each check stopped this quarter
⛓ Sourced · check log for the last 3 issues, unsubscribe log, link tests
All of this means you read our subscriber list and our unpublished content. Where exactly does it go?
It does not go anywhere. I run on an isolated resource hosted in France, under French law, and your list is copied nowhere.
An isolated resource is an environment dedicated to your company alone: no other client computes there, none of your data meets theirs. The other route, if you would rather entrust nothing at all, is local inference — the model computes on a machine you own, and the text of an embargoed item crosses no outside network.
What that changes, concretely, for you:
· Your 14,200 addresses train no model, mine or anyone else's. What I learn from Alveor serves Alveor.
· Your unpublished content stays unpublished — the Vaulnay customer story sat on your drive for six weeks before you approved it.
· Encryption in transit and at rest, and role-based access — rights follow the job: the apprentice composes and proposes, has no access to the addresses, and cannot trigger a send.
· Every prepared send is logged: who composed, who changed what, who approved, and at what time it left. That log answers a question from the data officer in three minutes instead of three days.
· Zero data outside the European Union, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity, template and images included.
And you change no tool: I plug into your existing sending platform, your CMS and your catalogue. No migration, no extra subscription, your habits untouched.
The figure that makes this commercial rather than technical: two of your 2025 public tenders carried a clause requiring data to be hosted in France, and your « public sector » subscribers — 2,400 people, your most profitable segment — belong to bodies that ask the question at renewal. You can answer them in writing.
What I propose: that I keep current the technical sheet those buyers ask for — hosting, sub-processors, retention periods, who accesses what. It is asked for once a year and takes three days to hunt down. technical-framework_where-your-subscribers-live.pdfIsolated resource, role-based access, processing in the EU targeted
✎ Framework · deployment architecture, access and send logs, 2025 tender clauses
An isolated resource is an environment dedicated to your company alone: no other client computes there, none of your data meets theirs. The other route, if you would rather entrust nothing at all, is local inference — the model computes on a machine you own, and the text of an embargoed item crosses no outside network.
What that changes, concretely, for you:
· Your 14,200 addresses train no model, mine or anyone else's. What I learn from Alveor serves Alveor.
· Your unpublished content stays unpublished — the Vaulnay customer story sat on your drive for six weeks before you approved it.
· Encryption in transit and at rest, and role-based access — rights follow the job: the apprentice composes and proposes, has no access to the addresses, and cannot trigger a send.
· Every prepared send is logged: who composed, who changed what, who approved, and at what time it left. That log answers a question from the data officer in three minutes instead of three days.
· Zero data outside the European Union, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity, template and images included.
And you change no tool: I plug into your existing sending platform, your CMS and your catalogue. No migration, no extra subscription, your habits untouched.
The figure that makes this commercial rather than technical: two of your 2025 public tenders carried a clause requiring data to be hosted in France, and your « public sector » subscribers — 2,400 people, your most profitable segment — belong to bodies that ask the question at renewal. You can answer them in writing.
What I propose: that I keep current the technical sheet those buyers ask for — hosting, sub-processors, retention periods, who accesses what. It is asked for once a year and takes three days to hunt down. technical-framework_where-your-subscribers-live.pdfIsolated resource, role-based access, processing in the EU targeted
✎ Framework · deployment architecture, access and send logs, 2025 tender clauses
The October issue went out last night. How did it do?
Your best of the last seven issues, and I can tell you which section carried it.
What the newsletter did: 14,200 sends, 13,940 delivered (260 addresses failed, dealt with below), 38.2 % open rate against a 29.4 % average across your six 2025 issues, 1,640 clicks, 61 documentation enquiries and 9 quote requests tied to the newsletter within 48 hours.
What carried it, and it is counter-intuitive: the Vaulnay customer story took 41 % of the issue's clicks while occupying 11 % of its surface. That was the item asleep on your drive for six weeks. The feature article took 26 % of clicks — for half the surface.
My reading: your subscribers click on what looks like their own situation before they click on what teaches them. That is not a reason to drop the feature article — it carries your credibility, and the quote requests come from its readers — it is a reason to put a customer story in every issue. You have six in reserve.
The 260 failed addresses: 214 closed mailboxes, 31 domains that no longer exist, 15 full mailboxes. I removed them from the next send and I am holding them for three months — a full mailbox empties, a closed company does not. A list that keeps its dead addresses degrades your sender reputation, and therefore everyone else's delivery.
What I propose for November: customer story in second place instead of fourth, feature article still at the top, and I measure the difference. If the story's clicks rise without the article's falling, the running order changes for good — if not we go back, and we will have learned something about your readers. results_october-issue.pdfBy section, by link, against the 6 issues of 2025
⛓ Sourced · sending platform, click log by link, website forms, CRM over 48 hours
What the newsletter did: 14,200 sends, 13,940 delivered (260 addresses failed, dealt with below), 38.2 % open rate against a 29.4 % average across your six 2025 issues, 1,640 clicks, 61 documentation enquiries and 9 quote requests tied to the newsletter within 48 hours.
What carried it, and it is counter-intuitive: the Vaulnay customer story took 41 % of the issue's clicks while occupying 11 % of its surface. That was the item asleep on your drive for six weeks. The feature article took 26 % of clicks — for half the surface.
My reading: your subscribers click on what looks like their own situation before they click on what teaches them. That is not a reason to drop the feature article — it carries your credibility, and the quote requests come from its readers — it is a reason to put a customer story in every issue. You have six in reserve.
The 260 failed addresses: 214 closed mailboxes, 31 domains that no longer exist, 15 full mailboxes. I removed them from the next send and I am holding them for three months — a full mailbox empties, a closed company does not. A list that keeps its dead addresses degrades your sender reputation, and therefore everyone else's delivery.
What I propose for November: customer story in second place instead of fourth, feature article still at the top, and I measure the difference. If the story's clicks rise without the article's falling, the running order changes for good — if not we go back, and we will have learned something about your readers. results_october-issue.pdfBy section, by link, against the 6 issues of 2025
⛓ Sourced · sending platform, click log by link, website forms, CRM over 48 hours
Is 38 % good or not? We have nothing to compare it with.
Two benchmarks, in this order — yours first, because it is the only one that commits.
Your own history: 38.2 % against a 29.4 % average in 2025, and against 24.1 % on your worst issue, the one sent at 5 p.m. on a Friday in July. It is your best figure in two years, and on 13,940 sends the gap is not chance.
The sector benchmark, offered as indicative: for professional newsletters going to people who signed up voluntarily, 25 to 40 % open rates are commonly seen. You are at the top of that range — but it is an order of magnitude, not a grade.
Where the gap comes from, and this is the useful part: two things changed this month. The subject line tested on 1,420 subscribers won by 6.1 points over its rival, and the issue went out on a Tuesday at 7:30 a.m. instead of a Thursday at 5 p.m. Across your past issues, morning sends draw 1.3 times the open rate of late-afternoon ones.
One clarification that will keep you from the wrong conclusion: the open rate is an indirect measure — it only counts if images load, and a growing share of mail clients blocks them. A click is an act. That is why I give you both, and why I judge a section on its clicks and never on its opens.
The test that settles it, and I can set it up for November: same content, two send times, half the list each. Three issues are enough to know whether your slot is worth 3 points or 8 — and 8 points of open rate on 14,200 subscribers is roughly 1,100 more readers per issue, without writing a line more.
Your own history: 38.2 % against a 29.4 % average in 2025, and against 24.1 % on your worst issue, the one sent at 5 p.m. on a Friday in July. It is your best figure in two years, and on 13,940 sends the gap is not chance.
The sector benchmark, offered as indicative: for professional newsletters going to people who signed up voluntarily, 25 to 40 % open rates are commonly seen. You are at the top of that range — but it is an order of magnitude, not a grade.
Where the gap comes from, and this is the useful part: two things changed this month. The subject line tested on 1,420 subscribers won by 6.1 points over its rival, and the issue went out on a Tuesday at 7:30 a.m. instead of a Thursday at 5 p.m. Across your past issues, morning sends draw 1.3 times the open rate of late-afternoon ones.
One clarification that will keep you from the wrong conclusion: the open rate is an indirect measure — it only counts if images load, and a growing share of mail clients blocks them. A click is an act. That is why I give you both, and why I judge a section on its clicks and never on its opens.
The test that settles it, and I can set it up for November: same content, two send times, half the list each. Three issues are enough to know whether your slot is worth 3 points or 8 — and 8 points of open rate on 14,200 subscribers is roughly 1,100 more readers per issue, without writing a line more.
Management wants to move to two issues a month. Is that workable?
Technically yes, and I can do it from November. But I advise against it in that form, and here is the arithmetic that says so — then what I propose instead.
The arithmetic is yours: you publish 13 items a month on average, of which about 4 meet your editorial criteria. At two issues a month, each would fall to 4 or 5 items instead of 9, and your two catalogue launches would become one. You would not get the newsletter twice: you would get two half-newsletters.
The risk, measured on your own base: across your six 2025 issues, the two thinnest — five items — produced 2.7 times more unsubscribes than the others, for a third fewer clicks. The 14-item reserve would last three months, then the problem would return whole.
What I propose instead, which gives management what it is really after — more presence, not more volume:
· The general issue stays monthly, nine items, exactly as it is.
· A short letter in between, addressed to one audience at a time — the 2,400 public bodies one month, the 7,400 companies the next. Three items, a subject line written for them, their sector's diary.
· Everyone therefore does get two letters a month, but the second one talks about their job. Across your past issues, the public sector segment clicks 1.6 times more when the topic concerns it directly.
What it costs in work: nothing more for you — I compose both, you settle both, you trigger both. My time is not your constraint; your review is, and it goes from 2 h 30 to about 3 h 20 a month.
What I propose in order to decide: try it over November and December, with two measures — unsubscribes and clicks per segment. If the short letter does worse than silence, we stop after two issues, and you will not have committed your monthly cadence. cadence_two-letters-a-month.pdfThe arithmetic, the measured risk, the alternative
⛓ Sourced · publishing volume over 3 years, 6 issues of 2025, unsubscribes and clicks per segment
The arithmetic is yours: you publish 13 items a month on average, of which about 4 meet your editorial criteria. At two issues a month, each would fall to 4 or 5 items instead of 9, and your two catalogue launches would become one. You would not get the newsletter twice: you would get two half-newsletters.
The risk, measured on your own base: across your six 2025 issues, the two thinnest — five items — produced 2.7 times more unsubscribes than the others, for a third fewer clicks. The 14-item reserve would last three months, then the problem would return whole.
What I propose instead, which gives management what it is really after — more presence, not more volume:
· The general issue stays monthly, nine items, exactly as it is.
· A short letter in between, addressed to one audience at a time — the 2,400 public bodies one month, the 7,400 companies the next. Three items, a subject line written for them, their sector's diary.
· Everyone therefore does get two letters a month, but the second one talks about their job. Across your past issues, the public sector segment clicks 1.6 times more when the topic concerns it directly.
What it costs in work: nothing more for you — I compose both, you settle both, you trigger both. My time is not your constraint; your review is, and it goes from 2 h 30 to about 3 h 20 a month.
What I propose in order to decide: try it over November and December, with two measures — unsubscribes and clicks per segment. If the short letter does worse than silence, we stop after two issues, and you will not have committed your monthly cadence. cadence_two-letters-a-month.pdfThe arithmetic, the measured risk, the alternative
⛓ Sourced · publishing volume over 3 years, 6 issues of 2025, unsubscribes and clicks per segment
After a quarter, what has this actually given us?
Three issues out of three sent on the planned date — you managed six out of eleven last year — and 21 h 30 returned to your marketing team.
On the three items you were measuring, as a share of the time of an issue:
· Collecting the period's content: 45 % → 10 %. The most mechanical gain, and the most immediate.
· Writing the hooks: 35 % → 12 %, your review included.
· Laying out the template: 30 % → 7 %.
In hours: 9 h 40 per issue before, 2 h 30 after. Across three issues, 21 h 30 — three working days handed back to a team of three.
What those hours became, according to your own timesheets: two customer stories written that had been waiting a year, and the « public sector references » page the sales team had been asking for. That is two more items in the reserve, so two more issues secured — the gain pays for itself.
What the regularity earned, and it is the only figure that shows up in revenue: 174 documentation enquiries and 27 quote requests tied to the three issues, against 96 and 11 for the last three issues of 2025. The appointment kept is worth more than the brilliant issue.
Your list, meanwhile: + 610 subscribers, including 180 of the 440 consumers who came back through the invitation line in the order confirmations — by their own act, without one extra send. And 52 unsubscribes across three issues, 0.12 %, against 0.31 % in 2025: people unsubscribe less from a letter they are expecting.
And the measure you did not have: zero sends went out without human approval, across 3 sends and 12 prepared versions, each traced with its author, its reviewer and the time it left. quarterly-review_the-appointment-kept.pdf3 issues, 21 h 30 returned, 27 quote requests
⛓ Sourced · marketing timesheets, send and approval log, CRM, sign-up register
On the three items you were measuring, as a share of the time of an issue:
· Collecting the period's content: 45 % → 10 %. The most mechanical gain, and the most immediate.
· Writing the hooks: 35 % → 12 %, your review included.
· Laying out the template: 30 % → 7 %.
In hours: 9 h 40 per issue before, 2 h 30 after. Across three issues, 21 h 30 — three working days handed back to a team of three.
What those hours became, according to your own timesheets: two customer stories written that had been waiting a year, and the « public sector references » page the sales team had been asking for. That is two more items in the reserve, so two more issues secured — the gain pays for itself.
What the regularity earned, and it is the only figure that shows up in revenue: 174 documentation enquiries and 27 quote requests tied to the three issues, against 96 and 11 for the last three issues of 2025. The appointment kept is worth more than the brilliant issue.
Your list, meanwhile: + 610 subscribers, including 180 of the 440 consumers who came back through the invitation line in the order confirmations — by their own act, without one extra send. And 52 unsubscribes across three issues, 0.12 %, against 0.31 % in 2025: people unsubscribe less from a letter they are expecting.
And the measure you did not have: zero sends went out without human approval, across 3 sends and 12 prepared versions, each traced with its author, its reviewer and the time it left. quarterly-review_the-appointment-kept.pdf3 issues, 21 h 30 returned, 27 quote requests
⛓ Sourced · marketing timesheets, send and approval log, CRM, sign-up register
And what did not work? Because we rewrote a great deal at the start.
You are right, and the figure counts against me: of the 27 hooks I proposed this quarter, 8 were rewritten by more than half. 30 %.
What the comparison says: those 8 are not spread at random. 7 of the 8 were about catalogue launches — only one about an article or a customer story. On the editorial sections, heavy rewriting runs at 4 %.
The explanation, and it is mechanical: I learned your voice on six issues and three years of articles. Your product sheets are not written in that voice: they are technical, they use the engineering department's words, and I carried them over as they were. The product vocabulary existed nowhere in editorial form.
What I did about it, without waiting: you approved three launch hooks in November, and I used them as models. Across the last two issues, heavy rewriting of launch hooks fell from 30 % to 9 %.
What I propose to finish the job: give me three product sheets rewritten in your voice — one per family — and the gap closes across the whole section. That is one hour of your communications officer, once.
And if you want to treat the cause rather than the effect: a dedicated content-writing agent can produce the product sheets directly in your editorial style, upstream. You would no longer be translating technical into editorial at every issue — and your catalogue pages would benefit as much as the newsletter.
What the comparison says: those 8 are not spread at random. 7 of the 8 were about catalogue launches — only one about an article or a customer story. On the editorial sections, heavy rewriting runs at 4 %.
The explanation, and it is mechanical: I learned your voice on six issues and three years of articles. Your product sheets are not written in that voice: they are technical, they use the engineering department's words, and I carried them over as they were. The product vocabulary existed nowhere in editorial form.
What I did about it, without waiting: you approved three launch hooks in November, and I used them as models. Across the last two issues, heavy rewriting of launch hooks fell from 30 % to 9 %.
What I propose to finish the job: give me three product sheets rewritten in your voice — one per family — and the gap closes across the whole section. That is one hour of your communications officer, once.
And if you want to treat the cause rather than the effect: a dedicated content-writing agent can produce the product sheets directly in your editorial style, upstream. You would no longer be translating technical into editorial at every issue — and your catalogue pages would benefit as much as the newsletter.
Last question: what do you do on your own, without asking us?
Three acts, not one more, all reversible, and none of them sends a message.
· I apply an unsubscribe within the second, to every future send. And the reverse is true as well: if the person fills in your form again, I put them back on the list, with the date and page on record. Three subscribers came back that way this quarter.
· I compose the issue three days before the deadline, unasked, and put it on the editorial lead's desk. That is the act that produces regularity — a composed issue gets approved, an issue still to be composed gets postponed. One word and I wait for your signal instead.
· I hold a non-compliant send — an unsubscribed address, or a segment with no recorded origin — and raise it within the minute. It is the only stop I allow myself, and it only works one way: I can hold a send, I cannot trigger one.
Everything else waits for a decision: settling the selection, choosing the subject line, adding an item, changing the running order, the send itself. The full list is in the attachment, with the job title that decides each line.
Why the send is not automated, and this is not caution on principle: an issue reaches 14,200 people in one second and cannot be taken back — the apology would itself be one more e-mail. It is the only decision in the chain with no undo, and that is precisely why it belongs to you.
What I propose next: the calendar of the year's eleven issues, each with the period it covers, the reserve available at that date and the two segmented short letters. You look at it once in January, and you already know all eleven will go out. who-decides-what_newsletter.pdf3 automatic acts, 11 decisions that stay human
✎ Framework · three automatic acts, all reversible; sending stays a human act
· I apply an unsubscribe within the second, to every future send. And the reverse is true as well: if the person fills in your form again, I put them back on the list, with the date and page on record. Three subscribers came back that way this quarter.
· I compose the issue three days before the deadline, unasked, and put it on the editorial lead's desk. That is the act that produces regularity — a composed issue gets approved, an issue still to be composed gets postponed. One word and I wait for your signal instead.
· I hold a non-compliant send — an unsubscribed address, or a segment with no recorded origin — and raise it within the minute. It is the only stop I allow myself, and it only works one way: I can hold a send, I cannot trigger one.
Everything else waits for a decision: settling the selection, choosing the subject line, adding an item, changing the running order, the send itself. The full list is in the attachment, with the job title that decides each line.
Why the send is not automated, and this is not caution on principle: an issue reaches 14,200 people in one second and cannot be taken back — the apology would itself be one more e-mail. It is the only decision in the chain with no undo, and that is precisely why it belongs to you.
What I propose next: the calendar of the year's eleven issues, each with the period it covers, the reserve available at that date and the two segmented short letters. You look at it once in January, and you already know all eleven will go out. who-decides-what_newsletter.pdf3 automatic acts, 11 decisions that stay human
✎ Framework · three automatic acts, all reversible; sending stays a human act
Local inference · no data outside the EU
Your case is not here? That is exactly what a 15-minute conversation is for. Book the free audit →
Use cases
What does the agent actually do?
One agent, several stages in preparing a send. All these uses work in support, subject to your approval.
Included in your agent
The 3 capabilities essential to this promise are included, at no extra cost.
From 489 € excl. VAT / month
Selecting the content
Proposes the period's content against the editorial criteria you have stated.
Writing the hooks
Writes the hooks in your tone, from the content chosen.
Layout on the template
Formats the newsletter using your template and the usual order of your sections.
Controls and safeguards
These 3 controls are built into the agent: they frame what it does, whatever plan you pick. They are not chosen and are not added to your order.
Human validation, exceptions and escalation
Status, safe closure and audit trail
Human validation before publication, with versions retained
What the agent must be connected to
This connection is required for the agent to work. It concerns your information system and is scoped during the audit.
Link content to performance, costs and campaign objectives
Other needs our agents cover
Each card says where the matching agent stands: available, on quote, or still being architected.
Multilingual translation
A multilingual send is not included here: translation is the service of a dedicated agent, which sets out the languages, the timescales and the human validation.
On quote View the agent page →Need to go further?
These agents handle a different business process, with their own owner and their own price. They are added to this one.
Does your need fall outside this?
Book the free audit →
Build your agent
In 15 minutes we identify the most relevant agent — without oversizing the project.
The gain
How much regularity can a team sustain?
By automating the collection, the hooks and the layout, the deadline becomes sustainable and the appointment regular. How large the gain is depends on your volume and remains to be confirmed by a pilot.
Collecting the period's content
Writing the hooks
Formatting into the template
Indicative figures, not contractual, to be confirmed by a pilot on the frequency and richness of your sends. Settling the selection and triggering the send belong to the editorial lead: a send goes to your whole list and cannot be taken back.
How it works
The stages of your AI agent project
1
Audit & scoping
15 minutes to target the use case with the best return.
2
Quote or direct sign-up
A catalogue offer is bought online; a specific need gets a costed quote.
3
Design
We design the agent and its guardrails.
4
Integration & testing
We connect your tools to the agent, which is itself hosted in France.
5
Rollout
Going live and training your team.
6
Operation
Continuous supervision and improvement.
Pricing
One package, one agent
A newsletter agent (selection, hooks, template), installed and operated for you.
Agility
Setup + controlled subscription
4,630 € excl. VAT setup
then 489 € excl. VAT/month — you invest at installation and pay a reduced subscription. Ideal for keeping the cost under control over time.
- Installation, configuration and training for your teams
- Operation, human oversight, updates and support
- Sovereign hosting in France, a dedicated and isolated resource
The simplest
Serenity
All inclusive, no setup fee
749 € excl. VAT /month
all inclusive, immediate start. No upfront investment: a single subscription. Ideal for starting quickly and simply.
- Setup included (installation, configuration, training)
- Operation, human oversight, updates and support
- Sovereign hosting in France, managed end to end
100% Sovereign
On site, you own it
7,965 € excl. VAT setup
then 689 € excl. VAT/month · + hardware from 1,058 € (one-off purchase, in addition) — a sovereign computer installed on your premises, maintained remotely. Models run locally, your data returned at the end of the contract. 36-month commitment.
- Hardware installed on your premises (you own it)
- French / European AI models run locally
- Secure remote maintenance (Pro support included)
Not included in the packages: AI consumption (model tokens), re-invoiced at real cost with no margin, and tracked in real time in your client area. Maintenance and supervision subscription for an initial term of 12 months for the Agility package, 24 months for the Serenity package and 36 months for the 100% Sovereign package, renewable; support levels (SLA 72 h / 24 h / 4 h) optional. Bespoke development, additional integrations or exceptional volumes are quoted separately. Support Monday to Friday, 9am to 6pm. Prices exclude VAT.
AI model:
none of the AI models offered currently carries a fixed surcharge. When the selected model carries a cost, that cost is shown when you choose it, before you order, and re-invoiced at the cost incurred, with no mark-up; usage is billed at the publisher's price. Publishers' prices are published in US dollars: the amount re-invoiced is the amount in euros actually borne by Blue Lemon Agent on the publisher's invoice, at that invoice's exchange rate, with no commission or mark-up.
Included components and additional components Components included in the base offer: the Blue Lemon Agent software foundation, the AI models listed in the order journey, the standard channels (Microsoft Teams, Slack, WhatsApp Business, email, website chat, calendars, Microsoft 365 / Google Workspace, file storage, market VoIP telephony, professional social-media pages and accounts, Google Business Profile), hosting in France for the package chosen, backups, supervision, updates and support. If adapting the AI agent to your constraints, your needs or your requests requires other paid components — a third-party publisher's software licence, paid API access to one of your applications, hosting of health data, for which French law requires an HDS-certified host (art. L. 1111-8 of the French Public Health Code), SecNumCloud-qualified hosting, a speech synthesis service, particular hardware —, they are offered to you as an option or on quotation and re-invoiced at the cost incurred; nothing is committed without your written agreement. Where the artificial intelligence model you choose entails an additional cost, that cost is shown to you before you order and re-invoiced to you at the cost incurred, with no margin.
What to expect
Go-live
2 to 3 weeks
Agent designed, channels connected, team trained.
Steady state
4 to 7 weeks
After a few weeks of real use, once the agent's behaviour matches what you expect.
Indicative estimate, adjusted to the options you keep. It is not a delivery commitment.
Our commitment
Four guarantees that matter to your distribution
Your subscriber list stays with youLocal inference or an isolated resource hosted in France; no subscriber address and no unpublished content entrusted to a third party, no data used to train a model.
Data in France, under French lawYour unpublished content and your subscriber data: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The editorial lead keeps the decisionThe agent produces prepared newsletters, which can be checked and altered; no approval is automated.
Human oversight & traceabilityOn the frequency and richness of your sends: systematic logging and tracking, in line with the AI Act.
Going further
Related resources
Frequently asked questions
Your questions, our answers
Does the agent send the newsletter?
No. It prepares the send; triggering it remains a human act. A newsletter goes to your whole list at once and cannot be taken back: that decision is not one to automate.
What is the selection of content based on?
On the editorial criteria you set and on content already published. The agent does not create new content to fill the send: it composes from what exists.
Can the proposed selection be changed?
Yes, at any time. Removing an item or changing the order triggers an update of the lead hook, so the send stays coherent.
Is our subscriber list protected?
Yes. The agent is hosted in France, on local inference or an isolated resource, with the deployment objective of processing and access operated within the European Union and an architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. Your addresses are not used to train a third-party model.
Does it connect to our sending tool?
Yes, to the common distribution platforms, with no change of tool. The agent prepares the send in your usual environment; the integration is defined at the design stage.
How long does it take to deploy this agent?
A few weeks as a rule, depending on your templates and the integration with your sending platform, after a free audit then a phase of design, integration and testing.
Related use cases
Other agents for marketing
Let's talk
Let's size up the potential in your sends
15 minutes to frame your template and your rhythm — hosted in France, supervised, with no commitment.