We build Agents. You build the Future.
● B2B offer — Social mediaSocial media assistant: templates, calendar and scheduling
Filling the editorial calendar every week, adapting one post per template for each network, then scheduling it at the right time: a simple routine, but a constant one. Your assistant prepares the posts from your templates and keeps the calendar up to date, ready to schedule. Data hosted in France; the AI model is called at its publisher, except with the Sovereign model option. With the 100% Sovereign package, your content and your brand guidelines stay with you. Nothing is published without your go-ahead.
Updated on
Deployed in one to two weeks
Social media assistant · hosted in France
Fictional simulation — not a client result
Prepare next week's calendar from our templates: three posts, one per network, on the back-to-school theme.
3 posts prepared from the "announcement", "tip" and "behind the scenes" templates, adapted to each network's format.
The week's calendar filled in, with slots proposed according to your usual posting times.
Nothing is scheduled: the drafts are ready to read over.
✎ Action · drafts placed in the calendar — you approve before scheduling
The week's calendar filled in, with slots proposed according to your usual posting times.
Nothing is scheduled: the drafts are ready to read over.
✎ Action · drafts placed in the calendar — you approve before scheduling
The "tip" post works for me, schedule it for Thursday 9am. I will rework the other two.
"Tip" post scheduled for Thursday at 9am, as approved.
The other two stay as drafts in the calendar, waiting for you to read them over — nothing is published automatically for those two.
✎ Action · scheduling applied to what you approved
The other two stay as drafts in the calendar, waiting for you to read them over — nothing is published automatically for those two.
✎ Action · scheduling applied to what you approved
Data hosted in France · sovereign model as an option
Content hosted in France
Hosted in FranceAI model called at its publisher, except with the Sovereign model option
Content & templates protected
GDPR & AI Act: governed deploymentTraceability & human oversight
TurnkeyDesigned, installed and operated for you
You decideThe agent assists, never publishes alone
For an organisation established outside the European Union, the law of its own country applies; at its request, we put equivalent safeguards in place (human review before any action, a log of the criteria, information for the people concerned).
✦ In brief
A Blue Lemon Agent social media agent takes on the repetitive part of posting: it adapts your templates for each network, keeps the editorial calendar up to date and proposes a schedule at your usual slots. It does not moderate comments or reply to reviews — those are dedicated agents, available separately. Data hosted in France; the AI model is called at its publisher, except with the Sovereign model option. With that option, your content and your brand guidelines are never exposed to a foreign service, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. Nothing goes live without your approval. Live in one to two weeks, Your team writes to it from Microsoft Teams, Slack or their email, just as they would to a colleague, and steers the agent from a web dashboard. These connections are included in every plan, at no extra cost, within the number of connections set out in your agent's quote. The templates are adapted for the accounts you own: Facebook pages, Instagram professional accounts and LinkedIn pages. These are third-party platforms whose interfaces and rules change without notice: publishing in your name stays under your control, and any access fees they charge are passed on at actual cost, with no margin, outside the subscription.
100%
hosted in France in the target architecture
0
transfer outside the EU with the Sovereign model option (target architecture)
3
uses ready to deploy on this scope
0
decision with a legal or significant effect taken by an agent
These figures describe our offer, not results measured at a client: how large the gain is on your volume of posts is confirmed by a pilot.
Indicative estimate, to be confirmed by a pilot on your real files (general terms of sale, article 4).
Pricing
One package, one agent
A social media assistant (templates, calendar, scheduling), installed and operated for you.
Agility
Setup + controlled subscription
2,775 € excl. VAT setup
then 272 € 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
- Hosting in France, a dedicated and isolated resource
The simplest
Serenity
All inclusive, no setup fee
427 € 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
- Hosting in France, managed end to end
100% Sovereign
On site, you own it
4,625 € excl. VAT setup
then 376 € excl. VAT/month · + hardware from 2,491 € (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.
- Hardware installed on your premises (you own it)
- French / European AI models run locally
- Secure remote maintenance (Pro support included)
Included in the package: hosting and AI consumption up to the volume of your capacity tier, in documents, exchanges, tokens and storage; beyond that, re-invoiced at real cost with no margin, tracked in your client area as usage statements are issued. 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 for the package chosen, in France by default, 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.
Use cases
What does the assistant actually do?
One task, carried out from your templates, with links to neighbouring agents. The assistant carries out by itself what you authorise it to do; you decide what it does on its own and what it submits to you.
Included in your agent
The 3 capabilities essential to this promise are included, at no extra cost.
From 272 € excl. VAT / month
Adaptation by template
Applies your existing templates (announcement, tip, behind the scenes and so on) and adapts the format to Facebook, Instagram and LinkedIn.
Editorial calendar
Keeps the week's calendar up to date and proposes the slots according to your usual times.
Scheduling
Schedules the approved posts at the slots chosen; no unapproved draft goes live.
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
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.
Moderation & replies to comments
To watch and reply to comments, the full community management offer takes over.
Community management from 320 € excl. VAT / month Community management →Replying to customer reviews
A dedicated agent writes replies to the reviews on your Google Business Profile and on Trustpilot, in your brand's tone.
Customer review reply agent from 272 € excl. VAT / month Review replies →Visual creation
To produce the visuals that go with your templates, a dedicated agent adapts your formats and your guidelines.
Visual creation & adaptations from 227 € excl. VAT / month Visual creation →
Does your need fall outside this?
Build your agent
In 30 minutes we identify the most relevant agent — without oversizing the project.
Demonstration
See the agent at work
5 everyday 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 companyMaison Ferréol — artisan pâtisserie, 3 shops and a bakehouse
- Sector
- Pâtisserie and tearoom — shop sales and corporate orders (platters, gift boxes)
- Headcount
- 24 staff, including one part-time communications officer who also runs events
- Audience served
- Local customers across 3 shops and around a hundred corporate accounts — 3 networks kept up: Instagram, Facebook, LinkedIn
- Volume
- 12 posts a month planned, 7 actually published; 240 posts over two years; 4 commercial pushes a quarter
- Tools in place
- A scheduling tool, a drive holding the brand guide and photographs, the commercial diary and the till software — the agent plugs into them, nothing is replaced
- Who decides
- The communications officer approves every draft before scheduling; the owner settles the commercial pushes and the prices displayed
- Room for improvement
- Deriving a post for three networks eats 55 % of the time of a post; keeping the week's calendar up to date, 40 % of Monday morning; scheduling approved posts, 20 % of Thursday. The result: 5 posts out of 12 never go out, for lack of time
Maison Ferréol isn't looking to post differently, but to post what it planned to post. The agent runs on local inference on a machine at the house and plugs into the scheduling tool, the brand drive, the commercial diary and the till software: it derives the templates, keeps the calendar and proposes the slots; the communications officer approves, and nothing goes out without her. The exchanges below cover one quarter, from the reopening of the archives to the review.
This company, its figures and the exchanges that follow were invented for the demonstration. They illustrate a common situation; they describe no real client.
Social media assistant · hosted in France
Fictional simulation — not a client result
I read your two years of posts — 240 of them — and your brand guide, and drew five written templates from them. You could name three.
A template is the model for one kind of post: its structure, its tone, its length, what it shows and what it closes on. “Announcement”, “tip” and “behind the scenes” existed in your head; I have added “product of the month” and “team portrait”, which you post regularly without ever having named them.
The gap I measured, and it is the gap that decides the gain:
· “Behind the scenes” — the bakehouse, the hands, the night shift — carries 2.4 times your average reach — reach is the number of distinct accounts that saw the post — and accounts for only 9 % of what you publish.
· “Announcement” accounts for 61 % of your posts and carries 0.7 times average reach. The most published template is therefore not the one that travels furthest.
What I suggest, with the figures: move to 3 “behind the scenes”, 4 “tip” and 5 “announcement” a month — same volume, same workload. At constant reach per template, that rebalancing alone is worth around +38 % of monthly reach, without one extra photograph.
What it does not change: your commercial announcements keep going out — they sell on Saturdays. I am suggesting we move them, not drop them.
The decision is yours, and I suggest taking it once: approve the split, I fill the calendar with it, and we look at the result after four weeks. If reach doesn't move, we go back to your original split with one sentence. 5-templates_read-in-your-240-posts.pdfWhat each template carries · 2.4× for “behind the scenes”
⛓ Sourced · 240 posts over 2 years, reach and interactions per post, the house brand guide
A template is the model for one kind of post: its structure, its tone, its length, what it shows and what it closes on. “Announcement”, “tip” and “behind the scenes” existed in your head; I have added “product of the month” and “team portrait”, which you post regularly without ever having named them.
The gap I measured, and it is the gap that decides the gain:
· “Behind the scenes” — the bakehouse, the hands, the night shift — carries 2.4 times your average reach — reach is the number of distinct accounts that saw the post — and accounts for only 9 % of what you publish.
· “Announcement” accounts for 61 % of your posts and carries 0.7 times average reach. The most published template is therefore not the one that travels furthest.
What I suggest, with the figures: move to 3 “behind the scenes”, 4 “tip” and 5 “announcement” a month — same volume, same workload. At constant reach per template, that rebalancing alone is worth around +38 % of monthly reach, without one extra photograph.
What it does not change: your commercial announcements keep going out — they sell on Saturdays. I am suggesting we move them, not drop them.
The decision is yours, and I suggest taking it once: approve the split, I fill the calendar with it, and we look at the result after four weeks. If reach doesn't move, we go back to your original split with one sentence. 5-templates_read-in-your-240-posts.pdfWhat each template carries · 2.4× for “behind the scenes”
⛓ Sourced · 240 posts over 2 years, reach and interactions per post, the house brand guide
We have a brand guide, yet our posts don't look alike from one network to the next. We end up writing three different texts.
That is normal: the brand guide says what the brand must say, it doesn't say how each network displays it. I have written the second half.
What the brand guide fixes, and what moves on no network: formal address, the origin of the products named whenever it is true, never a superlative — your 240 posts contain none, a rule you held without ever writing it down — and the sign-off.
What each network imposes, and what I handle in your place:
· Instagram: the hook must fit before the text folds, the link is not clickable in a caption — so I point to your profile link, and I keep it up to date —, square or vertical image.
· Facebook: clickable link, preview to be checked, longer posts tolerated, and the first comment must not carry the link: on your account it halves reach.
· LinkedIn: professional tone, the “corporate orders” angle rather than “Sunday afternoon tea”, 3 hashtags maximum — beyond that your own history shows reach falling.
And one thing that is rarely done: I write the alt text for every image — the description a screen reader reads out, and which the networks also use to understand the photograph. None of your 240 posts carried one. It takes thirty seconds per image, and it makes your posts readable by customers you were losing without knowing it.
What it gives you: one text to review instead of three to write. Deriving is no longer writing, it is approving.
What I suggest: that you read the per-network constraints table once. Once approved it applies on its own to every post — and it takes one line to correct the day a network changes its rules. brand-guide-and-constraints_3-networks.pdfWhat never moves, what each network imposes
⛓ Sourced · house brand guide, 240 posts, reach compared by the form used
What the brand guide fixes, and what moves on no network: formal address, the origin of the products named whenever it is true, never a superlative — your 240 posts contain none, a rule you held without ever writing it down — and the sign-off.
What each network imposes, and what I handle in your place:
· Instagram: the hook must fit before the text folds, the link is not clickable in a caption — so I point to your profile link, and I keep it up to date —, square or vertical image.
· Facebook: clickable link, preview to be checked, longer posts tolerated, and the first comment must not carry the link: on your account it halves reach.
· LinkedIn: professional tone, the “corporate orders” angle rather than “Sunday afternoon tea”, 3 hashtags maximum — beyond that your own history shows reach falling.
And one thing that is rarely done: I write the alt text for every image — the description a screen reader reads out, and which the networks also use to understand the photograph. None of your 240 posts carried one. It takes thirty seconds per image, and it makes your posts readable by customers you were losing without knowing it.
What it gives you: one text to review instead of three to write. Deriving is no longer writing, it is approving.
What I suggest: that you read the per-network constraints table once. Once approved it applies on its own to every post — and it takes one line to correct the day a network changes its rules. brand-guide-and-constraints_3-networks.pdfWhat never moves, what each network imposes
⛓ Sourced · house brand guide, 240 posts, reach compared by the form used
And if we need a template we don't have? We're hiring two pastry chefs in September and we've never known how to announce it.
A template is built in twenty minutes from three approved posts — and this one I have already written. It is number six.
How I built it with no hiring history: I took your three “team portraits”, which already speak about your trades in your own words, and added what a job post must carry to be useful: the job title, the location, the type of contract, the hours, and how to apply in one action. Your last three job adverts, posted elsewhere, did not say how to apply in one action — that is the point that decides whether an application is sent, and it is now in the template.
What I corrected in the draft the owner dictated, and exactly why: it read “a young and dynamic team”. Article L1132-1 of the French Labour Code prohibits ruling out a candidate on grounds of age, and an advert that describes the team by its youth exposes you for no gain. The lawful path is simple and it says the same thing better: I wrote “a team of nine pastry chefs, from first qualification to thirty years at the bench” — it is true, it is verifiable, and it appeals to an experienced candidate and a beginner alike.
The template is derived for all three networks: LinkedIn first — that is where the candidates are —, Facebook for the local catchment, Instagram as “behind the scenes”: a photograph of the bakehouse at 5 am says more than a job advert.
What this opens up: your templates are not frozen at launch. Every time a need comes back — a shop moving, an annual closure, a shortage — we write one, it is reusable, and you stop reinventing the text under pressure. Tell me the next one and I'll prepare it today.
✎ Framework · 3 reference posts per template — the new template stays editable at any time
How I built it with no hiring history: I took your three “team portraits”, which already speak about your trades in your own words, and added what a job post must carry to be useful: the job title, the location, the type of contract, the hours, and how to apply in one action. Your last three job adverts, posted elsewhere, did not say how to apply in one action — that is the point that decides whether an application is sent, and it is now in the template.
What I corrected in the draft the owner dictated, and exactly why: it read “a young and dynamic team”. Article L1132-1 of the French Labour Code prohibits ruling out a candidate on grounds of age, and an advert that describes the team by its youth exposes you for no gain. The lawful path is simple and it says the same thing better: I wrote “a team of nine pastry chefs, from first qualification to thirty years at the bench” — it is true, it is verifiable, and it appeals to an experienced candidate and a beginner alike.
The template is derived for all three networks: LinkedIn first — that is where the candidates are —, Facebook for the local catchment, Instagram as “behind the scenes”: a photograph of the bakehouse at 5 am says more than a job advert.
What this opens up: your templates are not frozen at launch. Every time a need comes back — a shop moving, an annual closure, a shortage — we write one, it is reusable, and you stop reinventing the text under pressure. Tell me the next one and I'll prepare it today.
✎ Framework · 3 reference posts per template — the new template stays editable at any time
Fictional simulation — not a client result
Prepare next week's calendar from our templates: three posts, one per network, on the back-to-school theme.
The three posts are prepared and the week's calendar is complete. Nothing is scheduled: these are drafts, waiting for your review.
One template per post, chosen for the network where it carries:
· “Announcement” — Facebook: the return of the back-to-school pastries and the extended hours of the three shops. Clickable link to the opening hours page.
· “Tip” — LinkedIn: how to order a platter for a back-to-work meeting, lead time and minimum headcount. That is your corporate angle, and September is your strongest month on that line: 34 orders last year against 11 in August.
· “Behind the scenes” — Instagram: the bakehouse at 5 am, the first batch of the new term. The template that carries 2.4 times your average reach.
The slots I propose, with what justifies them: Facebook Tuesday 12:45 pm · LinkedIn Thursday 9 am · Instagram Saturday 10:30 am. These are your three best slots over two years, network by network — I'll set them out if you want to discuss them.
The time this moves: keeping the week's calendar took 40 % of your Monday morning; it now takes 10 %, the time to read three drafts.
What I suggest next: approve what you keep, correct what you want, and schedule with one click — I put nothing online in your place. Drafts that aren't approved stay in the calendar; they never go out on their own. week-calendar_3-drafts.pdf3 templates, 3 networks, 3 justified slots
⛓ Sourced · 5 templates, brand guide, commercial diary, 2 years of reach by slot, corporate order history
One template per post, chosen for the network where it carries:
· “Announcement” — Facebook: the return of the back-to-school pastries and the extended hours of the three shops. Clickable link to the opening hours page.
· “Tip” — LinkedIn: how to order a platter for a back-to-work meeting, lead time and minimum headcount. That is your corporate angle, and September is your strongest month on that line: 34 orders last year against 11 in August.
· “Behind the scenes” — Instagram: the bakehouse at 5 am, the first batch of the new term. The template that carries 2.4 times your average reach.
The slots I propose, with what justifies them: Facebook Tuesday 12:45 pm · LinkedIn Thursday 9 am · Instagram Saturday 10:30 am. These are your three best slots over two years, network by network — I'll set them out if you want to discuss them.
The time this moves: keeping the week's calendar took 40 % of your Monday morning; it now takes 10 %, the time to read three drafts.
What I suggest next: approve what you keep, correct what you want, and schedule with one click — I put nothing online in your place. Drafts that aren't approved stay in the calendar; they never go out on their own. week-calendar_3-drafts.pdf3 templates, 3 networks, 3 justified slots
⛓ Sourced · 5 templates, brand guide, commercial diary, 2 years of reach by slot, corporate order history
How do you choose the slots? We've been posting more or less at random for two years.
I don't choose them by feel: I read them in your 240 posts, hour by hour and network by network. And the gap is wider than you would expect.
What your history says:
· LinkedIn, Thursday 9 am: 1.9 times your average reach on that network. Your corporate customers open LinkedIn as they arrive, not in the evening.
· Instagram, Saturday 10:30 am: 2.3 times average reach, and it is also the moment when people decide on a dessert for lunch. The best slot you have, and it is used 4 times out of 10.
· Facebook, Tuesday 12:45 pm: 1.5 times.
And the slot I advise against, with the figure: Sunday 8 pm — 0.4 times average reach, and it is the slot you use most often, because it is when you have time. Your best posts have been going out at the least rewarding hours: that is exactly what a calendar prepared in advance fixes.
What I do not conclude: that these slots will still hold in six months. An audience moves, and a network changes its display rules without warning.
What I suggest so we know: a simple test over four weeks — two posts at the usual slot, two moved by an hour, same template, same subject. Four posts are enough to settle it, and I'll hand you the record. You set the slots, I hold to them; they only change if you decide so.
⛓ Sourced · 240 posts, reach recorded per network, day and hour over 24 months
What your history says:
· LinkedIn, Thursday 9 am: 1.9 times your average reach on that network. Your corporate customers open LinkedIn as they arrive, not in the evening.
· Instagram, Saturday 10:30 am: 2.3 times average reach, and it is also the moment when people decide on a dessert for lunch. The best slot you have, and it is used 4 times out of 10.
· Facebook, Tuesday 12:45 pm: 1.5 times.
And the slot I advise against, with the figure: Sunday 8 pm — 0.4 times average reach, and it is the slot you use most often, because it is when you have time. Your best posts have been going out at the least rewarding hours: that is exactly what a calendar prepared in advance fixes.
What I do not conclude: that these slots will still hold in six months. An audience moves, and a network changes its display rules without warning.
What I suggest so we know: a simple test over four weeks — two posts at the usual slot, two moved by an hour, same template, same subject. Four posts are enough to settle it, and I'll hand you the record. You set the slots, I hold to them; they only change if you decide so.
⛓ Sourced · 240 posts, reach recorded per network, day and hour over 24 months
Our real problem is that we forget the key dates. Twelfth Night, Mother's Day, the end-of-year gift boxes: we always start three days before.
The calendar is laid out over twelve weeks, with 18 dated milestones — and above all with the date you have to start, which is never the date of the event.
Where the 18 come from: the key dates of your trade, your own commercial pushes as they appear in your diary, and your sales peaks recorded at the till over two years. These are not generic dates: they are yours.
The four that need three weeks of preparation, and why:
· The end-of-year gift boxes. Your corporate orders are decided in mid-November: last year 61 % of orders came in before 25 November, and the first post went out on 2 December — after your customers had decided. The calendar moves it back to 10 November.
· Halloween, where production has to be settled ten days ahead.
· Black Friday, which implies a pricing decision — so an owner's call, not a post written the night before.
· Saint Nicholas' Day, strong in only one of your three shops: I suggest one targeted post rather than three generic ones.
What I do without being asked: I move every milestone back to its preparation date, not its event date, with the template proposed and the draft already written. You arrive on Monday to a filled calendar, not a blank page.
What that is worth, in your own figures: you aim for 12 posts a month and publish 7. The 5 missing ones are almost all key-date posts, the ones that sell. They are what the calendar wins back first. 12-week-calendar_18-milestones.pdf18 milestones, each at its preparation date
⛓ Sourced · commercial diary, sales peaks recorded at the till over 24 months, corporate order history
Where the 18 come from: the key dates of your trade, your own commercial pushes as they appear in your diary, and your sales peaks recorded at the till over two years. These are not generic dates: they are yours.
The four that need three weeks of preparation, and why:
· The end-of-year gift boxes. Your corporate orders are decided in mid-November: last year 61 % of orders came in before 25 November, and the first post went out on 2 December — after your customers had decided. The calendar moves it back to 10 November.
· Halloween, where production has to be settled ten days ahead.
· Black Friday, which implies a pricing decision — so an owner's call, not a post written the night before.
· Saint Nicholas' Day, strong in only one of your three shops: I suggest one targeted post rather than three generic ones.
What I do without being asked: I move every milestone back to its preparation date, not its event date, with the template proposed and the draft already written. You arrive on Monday to a filled calendar, not a blank page.
What that is worth, in your own figures: you aim for 12 posts a month and publish 7. The 5 missing ones are almost all key-date posts, the ones that sell. They are what the calendar wins back first. 12-week-calendar_18-milestones.pdf18 milestones, each at its preparation date
⛓ Sourced · commercial diary, sales peaks recorded at the till over 24 months, corporate order history
Fictional simulation — not a client result
The “tip” post works for me. Derive it properly for the three networks.
The three versions are ready, and they say the same thing without looking alike — which is exactly what a derivative is for.
What is identical across all three: the information, the promise, the lead time, the origin of the products and the sign-off. If you correct the lead time in one version, I correct it in the other two — that is where mistakes slip in when three texts are written by hand.
What I adapt, constraint by constraint:
· LinkedIn: 780 characters, “back-to-work meeting” angle, 3 hashtags, call to action to the order form.
· Facebook: 420 characters, clickable link with the preview checked, warmer tone, the link in the body and not in the first comment.
· Instagram: 210 characters before the fold, the whole hook visible without “see more”, pointing to the profile link I have updated, vertical image.
The nine constraints held, without you having to think about them: length, hook before the fold, number of hashtags, link clickable or not, image format and crop, alt text, origin notice, call to action, and sign-off.
The time this moves: deriving a post for three networks took 55 % of the time of a post; it now takes 16 % — and that 16 % is your review. Across 12 posts a month, that is around 6 hours back in the shop.
What I suggest next: read the Instagram version first. It is the one whose hook is most constrained, so it is where your eye adds the most. derivation_1-post-3-networks.pdfWhat doesn't move, what each network imposes
⛓ Sourced · approved post, per-network constraints table, house brand guide
What is identical across all three: the information, the promise, the lead time, the origin of the products and the sign-off. If you correct the lead time in one version, I correct it in the other two — that is where mistakes slip in when three texts are written by hand.
What I adapt, constraint by constraint:
· LinkedIn: 780 characters, “back-to-work meeting” angle, 3 hashtags, call to action to the order form.
· Facebook: 420 characters, clickable link with the preview checked, warmer tone, the link in the body and not in the first comment.
· Instagram: 210 characters before the fold, the whole hook visible without “see more”, pointing to the profile link I have updated, vertical image.
The nine constraints held, without you having to think about them: length, hook before the fold, number of hashtags, link clickable or not, image format and crop, alt text, origin notice, call to action, and sign-off.
The time this moves: deriving a post for three networks took 55 % of the time of a post; it now takes 16 % — and that 16 % is your review. Across 12 posts a month, that is around 6 hours back in the shop.
What I suggest next: read the Instagram version first. It is the one whose hook is most constrained, so it is where your eye adds the most. derivation_1-post-3-networks.pdfWhat doesn't move, what each network imposes
⛓ Sourced · approved post, per-network constraints table, house brand guide
And the visuals? All we have is a photograph of the bakehouse taken on a phone.
One photograph is enough for all three networks — provided it is cropped three times, and I have done that.
The three formats produced from your photograph: 1080 × 1350 for Instagram, 1200 × 628 for Facebook, 1200 × 627 for LinkedIn. Same image, three crops, and the subject stays in the same place in all three — that is what makes a series recognisable.
What I check, and what costs most when it does not get done: the safe area — the margin where the network's interface overlays its own buttons, and where nothing important must fall. On your 14 June post the price sat under the share button on mobile: it was unreadable on every phone.
What I did with what you have: the three crops, a light correction — your 5 am bakehouse photographs are consistently two stops underexposed — and the addition of your brand banner, in the colours and typeface of the drive.
Where my work on your photographs stops, and here is what takes over: producing a visual that doesn't exist — a staged shot, an illustration, a graphic template. That is the work of the visual creation agent, which derives your formats and your brand guide from your art direction. I have put figures on it for you: of this month's 12 posts, 9 need only a crop — which I have done — and 3 would gain from a produced visual. So you know exactly what the dedicated agent would bring you, and on what volume.
What I suggest straight away: that I flag, on every post, which of the two routes is enough. You only call in the reinforcement for the three that deserve it.
⛓ Sourced · photograph supplied, brand guide from the drive, safe areas of the 3 networks, history of cropped posts
The three formats produced from your photograph: 1080 × 1350 for Instagram, 1200 × 628 for Facebook, 1200 × 627 for LinkedIn. Same image, three crops, and the subject stays in the same place in all three — that is what makes a series recognisable.
What I check, and what costs most when it does not get done: the safe area — the margin where the network's interface overlays its own buttons, and where nothing important must fall. On your 14 June post the price sat under the share button on mobile: it was unreadable on every phone.
What I did with what you have: the three crops, a light correction — your 5 am bakehouse photographs are consistently two stops underexposed — and the addition of your brand banner, in the colours and typeface of the drive.
Where my work on your photographs stops, and here is what takes over: producing a visual that doesn't exist — a staged shot, an illustration, a graphic template. That is the work of the visual creation agent, which derives your formats and your brand guide from your art direction. I have put figures on it for you: of this month's 12 posts, 9 need only a crop — which I have done — and 3 would gain from a produced visual. So you know exactly what the dedicated agent would bring you, and on what volume.
What I suggest straight away: that I flag, on every post, which of the two routes is enough. You only call in the reinforcement for the three that deserve it.
⛓ Sourced · photograph supplied, brand guide from the drive, safe areas of the 3 networks, history of cropped posts
Next week's “announcement” post carries a 20 % discount on the gift boxes. Can we write it however we like?
You can announce it, and I have written it — but not on the price you gave me, and the right price is in your till.
What applies: article L112-1-1 of the French Consumer Code requires any announcement of a price reduction to state the prior price, and that prior price must be the lowest price applied over the thirty days preceding the reduction. This holds for a social media post exactly as for a shop window sign: what counts is the announcement, not the medium.
What your till says, and this is where it turns: the gift box is listed at €28.00, but it sold at €24.00 during the 3-9 August push. The reference price is therefore €24.00, not €28.00 — and a 20 % discount announced against €28.00 would be inaccurate.
The lawful path, and it costs you nothing: I have written two wordings, both correct.
· “Tasting box €19.20 instead of €24.00 — lowest price applied in the last 30 days.” The real discount is still 20 %, it is simply calculated on the right price.
· “Tasting box €19.20, our best price of the year”, if the owner would rather not display a percentage — and it is true: €19.20 is the lowest price of your last 24 months.
What I checked at the same time, because it sits in the same place: the dates of the push, which shops are covered — the discount applies in two of the three — and the stock on hand, which your till holds: 142 boxes. An announcement that promises more than the stock costs more than a cautious one.
What I suggest: that I run this check on every price announcement, automatically, with the reference price drawn from the till and the period used. The owner settles the price; I make sure what is written matches what was sold. discount-announcement_reference-price.pdfReference price €24.00 · 2 correct wordings
⛓ Sourced · till software (prices applied over 30 days and 24 months), commercial diary, article L112-1-1 of the French Consumer Code
What applies: article L112-1-1 of the French Consumer Code requires any announcement of a price reduction to state the prior price, and that prior price must be the lowest price applied over the thirty days preceding the reduction. This holds for a social media post exactly as for a shop window sign: what counts is the announcement, not the medium.
What your till says, and this is where it turns: the gift box is listed at €28.00, but it sold at €24.00 during the 3-9 August push. The reference price is therefore €24.00, not €28.00 — and a 20 % discount announced against €28.00 would be inaccurate.
The lawful path, and it costs you nothing: I have written two wordings, both correct.
· “Tasting box €19.20 instead of €24.00 — lowest price applied in the last 30 days.” The real discount is still 20 %, it is simply calculated on the right price.
· “Tasting box €19.20, our best price of the year”, if the owner would rather not display a percentage — and it is true: €19.20 is the lowest price of your last 24 months.
What I checked at the same time, because it sits in the same place: the dates of the push, which shops are covered — the discount applies in two of the three — and the stock on hand, which your till holds: 142 boxes. An announcement that promises more than the stock costs more than a cautious one.
What I suggest: that I run this check on every price announcement, automatically, with the reference price drawn from the till and the period used. The owner settles the price; I make sure what is written matches what was sold. discount-announcement_reference-price.pdfReference price €24.00 · 2 correct wordings
⛓ Sourced · till software (prices applied over 30 days and 24 months), commercial diary, article L112-1-1 of the French Consumer Code
Fictional simulation — not a client result
The “tip” post works for me, schedule it for Thursday 9 am. I'll rework the other two.
“Tip” post scheduled for Thursday 9 am, exactly as you approved it. The other two stay as drafts — no automatic publication on those.
What the scheduling records, and what will serve you one day: the exact version of the text that went out, your name as approver, the time of approval and the time of publication. That log is what lets you answer, six months later, “who wrote this?” — and the answer is never “the tool”.
What I checked before scheduling, in twelve seconds: the link responds, the preview shows the right image, the alt text is in place, and no other post goes out on LinkedIn within two hours — two posts close together on the same network take reach from each other.
The time this moves: scheduling approved posts took 20 % of your Thursday; it now takes 6 %, the time of the approval click.
What I suggest for the two drafts: I keep them at their slots, and I bring them back to you on Wednesday at 5 pm if you haven't reworked them. A forgotten draft is the leading cause of your 5 missing posts a month — not a lack of ideas. scheduling-log_who-approved-what.pdfVersion, approver, time — for every post
✎ Framework · scheduling applied to what you approved — named, dated approval log
What the scheduling records, and what will serve you one day: the exact version of the text that went out, your name as approver, the time of approval and the time of publication. That log is what lets you answer, six months later, “who wrote this?” — and the answer is never “the tool”.
What I checked before scheduling, in twelve seconds: the link responds, the preview shows the right image, the alt text is in place, and no other post goes out on LinkedIn within two hours — two posts close together on the same network take reach from each other.
The time this moves: scheduling approved posts took 20 % of your Thursday; it now takes 6 %, the time of the approval click.
What I suggest for the two drafts: I keep them at their slots, and I bring them back to you on Wednesday at 5 pm if you haven't reworked them. A forgotten draft is the leading cause of your 5 missing posts a month — not a lack of ideas. scheduling-log_who-approved-what.pdfVersion, approver, time — for every post
✎ Framework · scheduling applied to what you approved — named, dated approval log
And if we want to cancel after scheduling? We once published a promotion on the day of an accident in the neighbourhood.
Anything scheduled can be pulled or moved with one sentence, up to two minutes before it publishes — and anything already online comes down just as fast.
The three moves, and they take the same time: “pull Thursday's post” · “push the whole week back seven days” · “freeze the calendar”, which suspends everything scheduled across the three networks without deleting anything.
The case you describe, as it happened last month: an incident in a nearby street, on a Friday at 8:12 am. You said “freeze everything”; 14 drafts and 3 scheduled posts were suspended in 40 seconds, across the three networks. The thaw, on the Monday, put every post back at its slot, moved one notch. Nothing was lost, nothing was rewritten.
The only two moves I make without asking, and they protect the relationship, not the numbers:
· I push a post back by thirty minutes if another goes out on the same network in between. You are told, and you can tell me to stop doing it with one word.
· I suspend the queue if a network returns an access error — password changed, authorisation expired — and I tell you at once. A post that fails in silence is worse than a post that never went out: you believe you have communicated.
Everything else waits for your decision, and the reverse holds too: nothing I have suspended disappears. A freeze is not a deletion, and it lifts with one word.
What I suggest: that we set an automatic freeze rule together — for instance, suspending commercial posts when the town hall issues an alert for the district. You set it, I apply it, and you can withdraw it just as easily.
✎ Framework · scheduling queue, freeze and thaw log, 2 automatic moves, named and reversible
The three moves, and they take the same time: “pull Thursday's post” · “push the whole week back seven days” · “freeze the calendar”, which suspends everything scheduled across the three networks without deleting anything.
The case you describe, as it happened last month: an incident in a nearby street, on a Friday at 8:12 am. You said “freeze everything”; 14 drafts and 3 scheduled posts were suspended in 40 seconds, across the three networks. The thaw, on the Monday, put every post back at its slot, moved one notch. Nothing was lost, nothing was rewritten.
The only two moves I make without asking, and they protect the relationship, not the numbers:
· I push a post back by thirty minutes if another goes out on the same network in between. You are told, and you can tell me to stop doing it with one word.
· I suspend the queue if a network returns an access error — password changed, authorisation expired — and I tell you at once. A post that fails in silence is worse than a post that never went out: you believe you have communicated.
Everything else waits for your decision, and the reverse holds too: nothing I have suspended disappears. A freeze is not a deletion, and it lifts with one word.
What I suggest: that we set an automatic freeze rule together — for instance, suspending commercial posts when the town hall issues an alert for the district. You set it, I apply it, and you can withdraw it just as easily.
✎ Framework · scheduling queue, freeze and thaw log, 2 automatic moves, named and reversible
There are 43 comments under this week's posts. We never look at them before the weekend.
I counted them, sorted them, and drew an editorial decision from them that earns you more than replies would: 26 of those 43 comments ask the same question.
What the 43 say:
· 26 are about the opening hours and days of the three shops — mostly the town-centre one, whose summer hours were up to date nowhere.
· 9 are product questions: allergens, gluten-free, platter sizes.
· 4 are compliments after the “behind the scenes” post.
· 4 deserve your eye today — two orders placed in the comments, one complaint about a Saturday order, one report of wrong opening hours. I sent those to you within ten minutes, with the direct link; they did not wait for the weekend.
The decision I draw from it, and that is my job: I have written a “practical information” template — hours of the three shops, closing days, how to order — pinned at the top of both accounts and reposted on the first Monday of each month. Since then, opening-hours questions have gone from 26 to 9 a week. The best answer to a question asked forty times is a post, not forty replies.
Where my role on the thread stops, and here is exactly what takes over: replying to comments and moderating them. That is the work of the community management agent, which writes the replies in your tone and escalates the cases that belong to you. I have already laid the ground for it, and you know the volume it would work on: 43 comments a week, 35 of them in three families with nothing at stake.
What I keep doing every week: the count, the sorting, the escalation of urgent cases within ten minutes, and the template that makes the most frequent question fall away. Tell me the next question that comes back too often and I'll turn it into a post. 43-comments_the-signal-and-the-decision.pdf26 opening-hours questions → 9, one pinned template
⛓ Sourced · comment threads across the 3 networks, shop opening hours, reach of the “practical information” template before and after
What the 43 say:
· 26 are about the opening hours and days of the three shops — mostly the town-centre one, whose summer hours were up to date nowhere.
· 9 are product questions: allergens, gluten-free, platter sizes.
· 4 are compliments after the “behind the scenes” post.
· 4 deserve your eye today — two orders placed in the comments, one complaint about a Saturday order, one report of wrong opening hours. I sent those to you within ten minutes, with the direct link; they did not wait for the weekend.
The decision I draw from it, and that is my job: I have written a “practical information” template — hours of the three shops, closing days, how to order — pinned at the top of both accounts and reposted on the first Monday of each month. Since then, opening-hours questions have gone from 26 to 9 a week. The best answer to a question asked forty times is a post, not forty replies.
Where my role on the thread stops, and here is exactly what takes over: replying to comments and moderating them. That is the work of the community management agent, which writes the replies in your tone and escalates the cases that belong to you. I have already laid the ground for it, and you know the volume it would work on: 43 comments a week, 35 of them in three families with nothing at stake.
What I keep doing every week: the count, the sorting, the escalation of urgent cases within ten minutes, and the template that makes the most frequent question fall away. Tell me the next question that comes back too often and I'll turn it into a post. 43-comments_the-signal-and-the-decision.pdf26 opening-hours questions → 9, one pinned template
⛓ Sourced · comment threads across the 3 networks, shop opening hours, reach of the “practical information” template before and after
Fictional simulation — not a client result
After a quarter, what has this actually brought us?
What you planned to publish is published. You were aiming for 12 posts a month and getting 7 out: you are at 12.
On the three areas you were measuring:
· Deriving a post for 3 networks: 55 % → 16 % of the time of a post.
· Keeping the weekly calendar: 40 % → 10 % of Monday morning.
· Scheduling approved posts: 20 % → 6 % of Thursday.
What that is in hours, against a half-time post: around 14 hours a month, which is more than four weeks across the year (168 h, 35-hour basis) — and they went to events and corporate orders, according to your own record.
What the regularity delivered, measured on the networks: +41 % of cumulative monthly reach, much of it simply from rebalancing the templates and the slots — not one extra post, but the right post at the right hour.
The figure I publish about myself: of 148 drafts prepared, 27 were rewritten by more than half — 18 %. And they are not spread at random: 19 of the 27 are LinkedIn posts. The explanation is simple, and it lies in the material available: you have 140 Instagram posts behind you and only 22 on LinkedIn. My material is thin where your history is.
What I do about it: three approved LinkedIn posts are enough to rebuild the template. We already did it on Facebook: rewriting fell from 24 % to 7 % in five weeks. It is the same fix, applied to the last network.
And the measurement you didn't have: zero posts published without a named approval this quarter, across 148 logged approvals. quarterly-review_what-the-house-got-back.pdf14 h a month, 12 posts out of 12, 18 % rewritten
⛓ Sourced · the communications officer's time record, approval log, monthly reach across the 3 networks, successive versions of the drafts
On the three areas you were measuring:
· Deriving a post for 3 networks: 55 % → 16 % of the time of a post.
· Keeping the weekly calendar: 40 % → 10 % of Monday morning.
· Scheduling approved posts: 20 % → 6 % of Thursday.
What that is in hours, against a half-time post: around 14 hours a month, which is more than four weeks across the year (168 h, 35-hour basis) — and they went to events and corporate orders, according to your own record.
What the regularity delivered, measured on the networks: +41 % of cumulative monthly reach, much of it simply from rebalancing the templates and the slots — not one extra post, but the right post at the right hour.
The figure I publish about myself: of 148 drafts prepared, 27 were rewritten by more than half — 18 %. And they are not spread at random: 19 of the 27 are LinkedIn posts. The explanation is simple, and it lies in the material available: you have 140 Instagram posts behind you and only 22 on LinkedIn. My material is thin where your history is.
What I do about it: three approved LinkedIn posts are enough to rebuild the template. We already did it on Facebook: rewriting fell from 24 % to 7 % in five weeks. It is the same fix, applied to the last network.
And the measurement you didn't have: zero posts published without a named approval this quarter, across 148 logged approvals. quarterly-review_what-the-house-got-back.pdf14 h a month, 12 posts out of 12, 18 % rewritten
⛓ Sourced · the communications officer's time record, approval log, monthly reach across the 3 networks, successive versions of the drafts
Our templates, our brand guide, our bakehouse photographs: where exactly do they go? We care about our identity staying with us.
With the 100% Sovereign package, they don't leave your walls. I run on local inference on a machine at the house, and your drive is copied nowhere.
Local inference means the model computes on your machine: the text of a template crosses no outside network to be processed. If you would rather not host a machine, the other route is an isolated resource hosted in France under French law, strictly dedicated to your business — no pooling with another shop.
What that changes, item by item:
· Your templates and your brand guide train no third-party model. What I learn from Maison Ferréol serves Maison Ferréol. Nothing you entrust to me resurfaces anywhere else — and this is your brand identity, not a technical file.
· Encryption in transit and at rest, and role-based access — rights follow the job: your communications officer opens drafts and the calendar; prices and the commercial diary stay with the owner.
· A full log: who approved what, when, which version went out. That log is what lets you answer a question a year later.
· Hosting in France, under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity — which no American tool can offer, even hosted in Europe.
· Human oversight end to end, which is what the AI Act expects of an assistive system: I prepare, a named person approves, nothing goes out without her.
And you don't change tools: I plug into your scheduling tool, your drive, your commercial diary and your till. No migration, no extra subscription — and if you add a network tomorrow, it is added to the constraints table, not to your workload.
What I suggest: that I keep up to date the one-page sheet saying where your data lives and who has access to it. It is rarely asked for, and it takes two days to assemble on the day it is. technical-framework_where-your-templates-live.pdf100% Sovereign package: local inference, role-based access, processing in the EU targeted
✎ Framework · deployment architecture, access log, scope of roles set by the owner
Local inference means the model computes on your machine: the text of a template crosses no outside network to be processed. If you would rather not host a machine, the other route is an isolated resource hosted in France under French law, strictly dedicated to your business — no pooling with another shop.
What that changes, item by item:
· Your templates and your brand guide train no third-party model. What I learn from Maison Ferréol serves Maison Ferréol. Nothing you entrust to me resurfaces anywhere else — and this is your brand identity, not a technical file.
· Encryption in transit and at rest, and role-based access — rights follow the job: your communications officer opens drafts and the calendar; prices and the commercial diary stay with the owner.
· A full log: who approved what, when, which version went out. That log is what lets you answer a question a year later.
· Hosting in France, under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity — which no American tool can offer, even hosted in Europe.
· Human oversight end to end, which is what the AI Act expects of an assistive system: I prepare, a named person approves, nothing goes out without her.
And you don't change tools: I plug into your scheduling tool, your drive, your commercial diary and your till. No migration, no extra subscription — and if you add a network tomorrow, it is added to the constraints table, not to your workload.
What I suggest: that I keep up to date the one-page sheet saying where your data lives and who has access to it. It is rarely asked for, and it takes two days to assemble on the day it is. technical-framework_where-your-templates-live.pdf100% Sovereign package: local inference, role-based access, processing in the EU targeted
✎ Framework · deployment architecture, access log, scope of roles set by the owner
One last thing: we get about forty Google reviews a month and we never reply. Can you do anything?
Replying to reviews, no — a dedicated agent does that. But I have read them, and they tell you something your social accounts don't: I have drawn two posts from them, already written.
What your 40 reviews this month say, sorted: 31 are positive and nearly all cite the same two things — the quality of the pastry and the welcome in store. 9 are negative, and they come down to two subjects: the Saturday morning queue (6 reviews) and the lack of a gluten-free option (3 reviews).
What I do with that, and it is my scope:
· A “practical information” post on the Saturday queue — ordering by phone the day before already exists at your shops, and it has never been said in any of the 240 posts. Six customers wrote about it publicly; the post that answers it is written and waiting for you.
· A “tip” post on your two naturally gluten-free recipes, which your bakehouse already produces. This is not a reply to a review: it is a reply to a demand, addressed to everyone who wonders without writing in.
· And a signal for the owner: the Saturday queue appears in 6 of the 9 negative reviews and in 4 comments this week. Two channels saying the same thing is an operations subject, not a communications one.
What takes over, and on what volume: the customer review agent writes the replies to Google and review-platform posts, in your brand's tone, and submits them to you. You now know what it would handle: 40 reviews a month, 31 calling for a three-line thank-you and 9 for a carefully written reply.
What I keep doing every month: read what comes in, draw out the subjects that deserve a post, and bring them to you written. Your customers are already telling you what to publish; they only needed counting.
⛓ Sourced · this month's 40 reviews sorted by subject, comment threads, bakehouse product list
What your 40 reviews this month say, sorted: 31 are positive and nearly all cite the same two things — the quality of the pastry and the welcome in store. 9 are negative, and they come down to two subjects: the Saturday morning queue (6 reviews) and the lack of a gluten-free option (3 reviews).
What I do with that, and it is my scope:
· A “practical information” post on the Saturday queue — ordering by phone the day before already exists at your shops, and it has never been said in any of the 240 posts. Six customers wrote about it publicly; the post that answers it is written and waiting for you.
· A “tip” post on your two naturally gluten-free recipes, which your bakehouse already produces. This is not a reply to a review: it is a reply to a demand, addressed to everyone who wonders without writing in.
· And a signal for the owner: the Saturday queue appears in 6 of the 9 negative reviews and in 4 comments this week. Two channels saying the same thing is an operations subject, not a communications one.
What takes over, and on what volume: the customer review agent writes the replies to Google and review-platform posts, in your brand's tone, and submits them to you. You now know what it would handle: 40 reviews a month, 31 calling for a three-line thank-you and 9 for a carefully written reply.
What I keep doing every month: read what comes in, draw out the subjects that deserve a post, and bring them to you written. Your customers are already telling you what to publish; they only needed counting.
⛓ Sourced · this month's 40 reviews sorted by subject, comment threads, bakehouse product list
Data hosted in France · sovereign model as an option
Your case is not here? That is exactly what a 30-minute conversation is for. Discuss your project →
How it works
The stages of your AI agent project
1
Audit & scoping
30 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.
The context
What does an AI agent bring to your social presence?
Adapting a post per network and keeping the calendar up to date is simple but constant work, which leaves little time for creating. Entrusting your templates and your guidelines to a consumer tool means losing control of them.
! The issue
Regular posting tolerates no interruption, but adapting the same post by hand for several networks takes a time out of all proportion to its value. Consumer AI tools, for their part, amount to entrusting templates, brand guidelines and posting calendar to a third party, often hosted outside Europe and subject to the Cloud Act.
✓ Our answer
AI is only of interest here if it stays sovereign and faithful to your templates. Local inference or an isolated resource hosted in France, systematic human oversight, nothing published automatically: time gained on adaptation is never paid for in lost control of your image. The aim is not to replace your editorial judgement, but to hand you the calendar ready.
The decisive point
Your templates and your guidelines: sovereignty & compliance
Your post templates embody your brand identity. Here is how the architecture of our agents protects them, post by post.
Local inference
With the 100% Sovereign package, the agent runs on a machine belonging to the company: no template leaves the network, nothing passes through a public cloud.
Hosting in France
Otherwise, with the Agility and Serenity packages, a dedicated, isolated resource hosted in France, under French law: data hosted in France; the AI model is called at its publisher, except with the Sovereign model option.
Reduced extraterritorial exposure
For your templates and your calendar, 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 immunityeven when hosted in Europe by an American provider.
Isolated resource
No pooling: an environment strictly dedicated to your company and your guidelines.
Templates & creative work preserved
Your templates are not used to train a third-party model; encryption, role-based access and logging.
AI Act: governed deployment
The agent carries out what you have authorised; no automatic publishing; 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 the most sensitive accounts, 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.
The gain
How much time can a team win back?
By automating the adaptation by template and the keeping of the calendar, the time spent on repetitive tasks comes down, reinvested in creating and in strategy. How large the gain is depends on your volume and remains to be confirmed by a pilot.
Adapting a post for 3 networks
Updating the weekly calendar
Scheduling the approved posts
Qualitative, non-contractual comparison: the proportions shown illustrate the shift of the work towards review, they represent no measurement. The agent does by itself what you authorise it to do; whatever leaves the company, and any decision reserved to a person, waits for a human go-ahead.
Indicative estimate, to be confirmed by a pilot on your real files (general terms of sale, article 4).
Our commitment
Four guarantees that matter to your social media
With the Sovereign model option, your templates and guidelines never leaveData hosted in France on an isolated resource; the AI model is called at its publisher, except with the Sovereign model option. With that option, no template entrusted to a foreign third party.
Data in France, under French lawYour content: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
You keep the decisionThe agent produces drafts and a calendar that can be checked; no publication is automated.
Human oversight & traceabilityFor adaptation and scheduling: systematic tracking and logging, in line with the AI Act.
Going further
Related resources
Frequently asked questions
Your questions, our answers
Does the agent post on our networks by itself?
No. It prepares the posts from your templates and proposes a schedule, but nothing goes live without your approval. Human oversight is the rule: the agent carries out what you have authorised, you decide and publish.
Are our templates and guidelines 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 templates are not used to train a third-party model.
Does the agent also handle moderation and replies to comments?
No, this offer is limited to templates, the calendar and scheduling. For moderation and replies, the community management offer or our customer review agent take over.
Does it connect to our existing posting tools?
Yes. The agent connects to your scheduling tools and to the accounts you own on Facebook and Instagram, without forcing a migration. For LinkedIn the mode is assisted: the agent prepares the copy, a person acts in the platform. The agent adapts your templates and keeps the calendar; the salesperson or the community manager checks the content themselves, adapts it, then posts it personally in LinkedIn. The BLA agent does not connect to LinkedIn, does not schedule any post there, does not visit any profile there, does not extract any data from it and does not follow up there automatically. The connection assumes you own the account and grant us access to it. We adapt the integration to your environment at deployment.
How long does it take to deploy this assistant?
One to two weeks as a rule, after a meeting with our sales team that identifies your priority templates, then a phase of design, integration and testing before going live.
Can new templates be added later?
Yes. The initial templates are defined with you at the start, then can evolve with your guidelines and your formats, without calling the agent's operation into question.
Do we have to learn a new tool to use it?
No. Your team writes to the agent from Microsoft Teams, Slack or their email, the way they would write to a colleague: the draft post comes back in the same thread, ready to read. These connections rest on open standards, including the MCP protocol, and are included in every plan, at no extra cost, within the number of connections set out in your agent's quote; the catalogue grows without a surcharge. You supervise the calendar from a web dashboard.
Does the agent post to our Facebook, Instagram and LinkedIn pages?
Yes, whenever it makes sense for you. The agent prepares and schedules posts on the accounts you own — Facebook pages, Instagram professional accounts — and follows the comments that answer them; on the review side, it picks up those left on your Google Business Profile and on Trustpilot. It flags what has just been posted, ranks it by urgency and drafts a reply in your brand's voice. For LinkedIn the mode is assisted: the agent prepares the copy, a person acts in the platform; the salesperson or the community manager checks the content themselves, adapts it, then posts it personally in LinkedIn. The BLA agent does not publish in LinkedIn and does not schedule any post there. Prepared posts and replies to comments follow the same circuit as everything else: draft, review, green light — no unapproved content goes out. Two things need saying plainly. These are third-party platforms: they belong to others, their interfaces and their rules change without notice, and the connection follows those changes. And publishing in your name stays under your control: the agent prepares and proposes, and you decide what goes live. Some of these platforms charge for their own access: those fees are passed on to you at actual cost, with no margin, outside the subscription. The exact set of connected platforms is settled during the design phase.
Related use cases
Other agents for communication
Let's talk
Let's size up the potential in your posting
30 minutes to identify your priority templates — hosted in France, supervised, with no commitment.