● Business offer — Steering & reporting
Activity reports: your templates, pre-filled
The periodic report always lands when you have the least time, and preparing it mostly consists of gathering figures that already exist. Your agent prepares the document in your template, on the schedule you set, from your activity data — and flags the variances against the previous period. Hosted in France: your internal indicators stay internal. The manager reviews, comments and circulates.
Hosted in France
Internal indicators protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Activity reports · hosted in France
Prepare the team's monthly report.
Report pre-filled in your monthly template, from the period's data: volumes, deliverables and the indicators you track.
Notable variances against the previous month are flagged, without being interpreted.
The comment boxes stay empty: they belong to you.
✎ Action · document pre-filled, the commentary is yours
Notable variances against the previous month are flagged, without being interpreted.
The comment boxes stay empty: they belong to you.
✎ Action · document pre-filled, the commentary is yours
The variance on the third indicator — where does it come from?
I set out the components of the indicator across both periods, so the comparison is legible.
Explaining the variance draws on your knowledge of the ground: I do not invent it.
✎ Support · figures restated, human analysis
Explaining the variance draws on your knowledge of the ground: I do not invent it.
✎ Support · figures restated, human analysis
Local inference · no data outside the EU
Indicators 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 team manager decidesThe agent prepares, never rules
✦ In brief
A Blue Lemon Agent reporting agent prepares your periodic documents in your templates, on the schedule you set, from your real activity data. It flags the variances against the previous period without interpreting them, and leaves the comment boxes empty. It runs on local inference or is hosted in France: your internal indicators are entrusted to nobody, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. Live within a few weeks,.
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
6
reporting uses ready to deploy
0
decision taken without human approval
Reference points describing our offer, not results measured at a client. The scale of the gain, given the number of reports and how often they fall due, is confirmed by a pilot.
The context
What does an AI agent bring to your periodic reports?
Gathering the figures takes most of the time; commenting on them takes almost none. Automating the first part gives the report back its purpose.
! The issue
A useful report rests on its commentary, not on collecting figures that already exist in your tools. The agent takes on the gathering and the formatting, on the schedule you set, and highlights the variances against the previous period — so that your time goes to analysis rather than copy-pasting.
✓ Our answer
The document arrives pre-filled and faithful to your template, with variances flagged but not interpreted: the analysis stays yours, and it is what gives the report its value. Local inference or an isolated resource hosted in France: your activity indicators, which describe the real health of your organisation, do not leave your premises.
The decisive point
Your activity data and internal indicators: sovereignty & compliance
Your activity indicators describe the real health of your organisation. Here is how the architecture of our agents protects them.
Local inference
The agent can run on a machine belonging to your organisation: no activity data leaves the network, no indicator passes through a public cloud.
Hosting in France
Otherwise, a dedicated and isolated resource hosted in France, under French law — your indicators and your reports: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
For your activity data and internal indicators, 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 whatsoever: an environment strictly dedicated to your organisation and its indicators.
Templates and schedule respected
Your templates and your deadlines are applied as they stand; encryption, role-based access (RBAC) and logging of every document prepared.
AI Act: governed deployment
An agent strictly in support; no variance interpreted or commented on in your place; 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 indicators tied to a financial transaction or a redundancy plan, SecNumCloud and reinforced hosting options are available 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
4 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.
Activity reports · hosted in France
Four flags this morning on your activity reports.
· Twelve per cent of the time fits no line. I do not spread it out. I create an "unallocated" line, and I will say why.
· Thirty-one entries charge a client for time that is not theirs.
· Sixty-eight per cent of reports are filled in on the last day. From memory, across five days.
· My pre-filled proposals are erased as soon as they are corrected. That is deliberate, and it is the most important rule here. morning-watch_4-flags.pdf12% unallocated · 31 charges to review
⛓ Source · 1,460 reports, 24 people, entry log
· Twelve per cent of the time fits no line. I do not spread it out. I create an "unallocated" line, and I will say why.
· Thirty-one entries charge a client for time that is not theirs.
· Sixty-eight per cent of reports are filled in on the last day. From memory, across five days.
· My pre-filled proposals are erased as soon as they are corrected. That is deliberate, and it is the most important rule here. morning-watch_4-flags.pdf12% unallocated · 31 charges to review
⛓ Source · 1,460 reports, 24 people, entry log
Why erase what you pre-filled?
Because a proposal kept beside the validated report reads as an accusation.
What I do: I pre-fill from the traces you have opened to me — meetings, tickets, documents opened. The person reads, corrects, validates. The moment they validate, my proposal disappears.
What would happen if I kept it: there would be, for every report, a gap between what the machine guessed and what the person declared. That gap would be consultable, comparable from one person to another, and it would end up being looked at.
And it means nothing: in the vast majority of cases, I am the one who got it wrong. A meeting cancelled without being deleted, a document opened for two minutes, a task done leaving no trace. My error rate would become somebody else's correction rate.
What I keep instead: the overall correction rate, with no names — that is my quality measure, not theirs. It is 34%, meaning one line in three that I propose is wrong.
What that implies: a proposal is never validated by default. An unread report is not sent. 34-percent-corrected_no-version-kept.pdfMy error rate would become somebody else’s correction rate
⛓ Source · 1,460 reports, 34% of lines corrected
What I do: I pre-fill from the traces you have opened to me — meetings, tickets, documents opened. The person reads, corrects, validates. The moment they validate, my proposal disappears.
What would happen if I kept it: there would be, for every report, a gap between what the machine guessed and what the person declared. That gap would be consultable, comparable from one person to another, and it would end up being looked at.
And it means nothing: in the vast majority of cases, I am the one who got it wrong. A meeting cancelled without being deleted, a document opened for two minutes, a task done leaving no trace. My error rate would become somebody else's correction rate.
What I keep instead: the overall correction rate, with no names — that is my quality measure, not theirs. It is 34%, meaning one line in three that I propose is wrong.
What that implies: a proposal is never validated by default. An unread report is not sent. 34-percent-corrected_no-version-kept.pdfMy error rate would become somebody else’s correction rate
⛓ Source · 1,460 reports, 34% of lines corrected
How do you work, and who do you tell?
I propose, a person corrects, and nothing goes out without them.
What I look at: the traces you have opened to me — calendar, tickets, documents, declared travel — and the reference list of engagements and chargeable lines.
Routing follows who can correct: a doubtful charge goes to the person concerned, before anything is sent; unallocated time to the team lead, as a volume and with no names; a closed engagement still being charged to whoever manages engagements; an unread report is blocked, not sent.
With a monthly summary: share of unallocated time, charges corrected by nature, closed engagements still active, and my overall correction rate.
Three operating rules, and the first holds up all the rest. The report is approved and sent by the person, never by a deadline expiring — one line in three that I propose is wrong, and that is my measure, not theirs. No named time statistics: no comparison between people, no ranking, no individual alert — such a counter would make the report serve the counter. And I read only traces, which say where the person was and not what they were doing: which is why I ask the question before sending rather than correcting afterwards — of this month's 31 doubtful charges, 19 were right and were explained in one sentence.
✎ Framework · nothing sent unread · no named statistics
What I look at: the traces you have opened to me — calendar, tickets, documents, declared travel — and the reference list of engagements and chargeable lines.
Routing follows who can correct: a doubtful charge goes to the person concerned, before anything is sent; unallocated time to the team lead, as a volume and with no names; a closed engagement still being charged to whoever manages engagements; an unread report is blocked, not sent.
With a monthly summary: share of unallocated time, charges corrected by nature, closed engagements still active, and my overall correction rate.
Three operating rules, and the first holds up all the rest. The report is approved and sent by the person, never by a deadline expiring — one line in three that I propose is wrong, and that is my measure, not theirs. No named time statistics: no comparison between people, no ranking, no individual alert — such a counter would make the report serve the counter. And I read only traces, which say where the person was and not what they were doing: which is why I ask the question before sending rather than correcting afterwards — of this month's 31 doubtful charges, 19 were right and were explained in one sentence.
✎ Framework · nothing sent unread · no named statistics
Why not spread the 12% pro rata?
Because spreading produces wrong figures that look exactly like right ones.
What I record: over the month, 12% of declared time matches no engagement: helping a colleague, picking up an absent person's file, internal meetings, waiting time, cross-cutting subjects.
What a pro rata split would do: those hours would be spread across engagements in proportion to time already charged. The total would balance, every engagement would carry a plausible cost, and no line would be accurate.
Why that is worse than leaving a hole: a hole is visible and gets discussed. A split is invisible — and when somebody compares an engagement's real cost with its declared cost, the gap is inexplicable, because its cause was dissolved into every line.
What I do: one "unallocated" line, with its share, and the detail of what it contains where the trace allows.
What that produced: looking at the detail, 4.1 points of the 12 came from a single cross-cutting subject that had no line. It has one now. Unallocated fell to 7.9% — not because it was hidden, because it was named. 12-percent_4-1-point-one-subject.pdfA split is invisible, a hole gets discussed
⛓ Source · 12% unallocated, 4.1 points from one unlisted subject
What I record: over the month, 12% of declared time matches no engagement: helping a colleague, picking up an absent person's file, internal meetings, waiting time, cross-cutting subjects.
What a pro rata split would do: those hours would be spread across engagements in proportion to time already charged. The total would balance, every engagement would carry a plausible cost, and no line would be accurate.
Why that is worse than leaving a hole: a hole is visible and gets discussed. A split is invisible — and when somebody compares an engagement's real cost with its declared cost, the gap is inexplicable, because its cause was dissolved into every line.
What I do: one "unallocated" line, with its share, and the detail of what it contains where the trace allows.
What that produced: looking at the detail, 4.1 points of the 12 came from a single cross-cutting subject that had no line. It has one now. Unallocated fell to 7.9% — not because it was hidden, because it was named. 12-percent_4-1-point-one-subject.pdfA split is invisible, a hole gets discussed
⛓ Source · 12% unallocated, 4.1 points from one unlisted subject
And the thirty-one charges to a client?
Cross-cutting time charged to a client, which makes them pay for something else.
What I record: 31 entries charge a client engagement with time whose traces point to an internal subject — a team meeting, a tool upgrade, training.
Why the question rather than the correction: traces say where the person was, not what they were doing — a team meeting can be 80% about one engagement, and charging it is then right. Correcting on my own account would make the client pay for a guess.
What I do: I put the question to the person concerned, before anything is sent, with what I saw: "this slot is charged to engagement X, the trace shows an internal meeting — is it really X?"
What that gave: of the 31, 19 were right and the person explained it in one sentence. 12 were errors — 9 of them carried over from a line the previous week.
What becomes of those 12, and it is the whole point of asking beforehand: they were corrected before anything was sent, so they never went out — there is no anomaly to escalate to a manager, the invoice that left is right. What I do escalate instead, and it can actually be fixed: 9 of the 12 came from a line carried over from the previous week — it is the "copy last week" button that produces them, and it can be settled once and for all. 31-questions_19-right-12-corrected.pdfTraces say where the person was, not what they were doing
⛓ Source · 31 charges questioned, 19 right, 12 corrected before sending
What I record: 31 entries charge a client engagement with time whose traces point to an internal subject — a team meeting, a tool upgrade, training.
Why the question rather than the correction: traces say where the person was, not what they were doing — a team meeting can be 80% about one engagement, and charging it is then right. Correcting on my own account would make the client pay for a guess.
What I do: I put the question to the person concerned, before anything is sent, with what I saw: "this slot is charged to engagement X, the trace shows an internal meeting — is it really X?"
What that gave: of the 31, 19 were right and the person explained it in one sentence. 12 were errors — 9 of them carried over from a line the previous week.
What becomes of those 12, and it is the whole point of asking beforehand: they were corrected before anything was sent, so they never went out — there is no anomaly to escalate to a manager, the invoice that left is right. What I do escalate instead, and it can actually be fixed: 9 of the 12 came from a line carried over from the previous week — it is the "copy last week" button that produces them, and it can be settled once and for all. 31-questions_19-right-12-corrected.pdfTraces say where the person was, not what they were doing
⛓ Source · 31 charges questioned, 19 right, 12 corrected before sending
Where do you get those figures? We are not going to key in a report just to feed you.
Nowhere on your side: collection from your own tools reads the data where it already exists.
Where I look: your four tools — the project system for closed tasks and their dates, the ticketing tool for interventions, the shared calendar for meetings and travel, the code repository for releases. No re-keying, in either direction: I read, I do not ask again.
What the collection covers: over a month of 1,480 declared hours from your 12 consultants, 1,302 hours — 88 % — are rebuilt from your tools, timestamped and attached to their client. The remaining 178 hours are the 12 % that fit nowhere, and that is exactly why they stay on an “unallocated” line instead of being spread around.
What that gives back: filling in a timesheet from memory took 50 minutes per person per month; reading over a prepared collection takes 9. Across 12 consultants and 12 months, more than 98 hours given back over the year, rounded down.
The figure that does not flatter me: collection from the shared calendar counted 31 half-days twice — 4 % of the volume — because a client trip appears there both as a meeting and as travel. The comparison with the tickets was clear: all 31 duplicates carried the same client and the same day. I now de-duplicate by client and by day, and I show the two merged lines rather than deleting one — you see what I joined up, and you can undo it.
⛓ Sourced · 4 tools read, 1,302 h of 1,480 rebuilt, 31 duplicates corrected
Where I look: your four tools — the project system for closed tasks and their dates, the ticketing tool for interventions, the shared calendar for meetings and travel, the code repository for releases. No re-keying, in either direction: I read, I do not ask again.
What the collection covers: over a month of 1,480 declared hours from your 12 consultants, 1,302 hours — 88 % — are rebuilt from your tools, timestamped and attached to their client. The remaining 178 hours are the 12 % that fit nowhere, and that is exactly why they stay on an “unallocated” line instead of being spread around.
What that gives back: filling in a timesheet from memory took 50 minutes per person per month; reading over a prepared collection takes 9. Across 12 consultants and 12 months, more than 98 hours given back over the year, rounded down.
The figure that does not flatter me: collection from the shared calendar counted 31 half-days twice — 4 % of the volume — because a client trip appears there both as a meeting and as travel. The comparison with the tickets was clear: all 31 duplicates carried the same client and the same day. I now de-duplicate by client and by day, and I show the two merged lines rather than deleting one — you see what I joined up, and you can undo it.
⛓ Sourced · 4 tools read, 1,302 h of 1,480 rebuilt, 31 duplicates corrected
Is Friday filling a discipline problem?
No. It is a memory problem, and it is measurable.
What I record: 68% of reports are entered on the last day of the week, for the five days just past.
What that produces, measured: on lines entered the same day, my pre-fill is corrected in 21% of cases. On lines entered five days later, it is corrected in 47% of cases — and the gap does not come from me: my proposals are the same. It is the corrections that differ.
What that means: at five days, the person is no longer correcting what they know, they are judging what they believe. Half-days smooth out, interruptions vanish, small tasks dissolve into large ones.
What I do instead of a daily reminder — which becomes noise people ignore within three weeks, and adds a constraint to work that already has plenty: I pre-fill continuously, and on Monday the person finds what was proposed for Friday. Correcting is faster than remembering, and it is the only thing I can genuinely change here.
What I report to the manager: the overall figure, never who fills in late. 68-percent_21-versus-47.pdfAt five days, you write what you believe, not what you know
⛓ Source · 68% entered on the last day, correction 21% vs 47%
What I record: 68% of reports are entered on the last day of the week, for the five days just past.
What that produces, measured: on lines entered the same day, my pre-fill is corrected in 21% of cases. On lines entered five days later, it is corrected in 47% of cases — and the gap does not come from me: my proposals are the same. It is the corrections that differ.
What that means: at five days, the person is no longer correcting what they know, they are judging what they believe. Half-days smooth out, interruptions vanish, small tasks dissolve into large ones.
What I do instead of a daily reminder — which becomes noise people ignore within three weeks, and adds a constraint to work that already has plenty: I pre-fill continuously, and on Monday the person finds what was proposed for Friday. Correcting is faster than remembering, and it is the only thing I can genuinely change here.
What I report to the manager: the overall figure, never who fills in late. 68-percent_21-versus-47.pdfAt five days, you write what you believe, not what you know
⛓ Source · 68% entered on the last day, correction 21% vs 47%
Could you measure who fills in late?
I could, and that is precisely what I will not do.
What it would give: an accurate ranking of who enters late. It is technically available and it looks harmless — a matter of administrative rigour, not performance.
Why I still refuse: because the activity report is already the most sensitive document your teams fill in. It is used to bill, therefore to measure. Adding a measure of how it is filled in means measuring people on the tool that already measures them.
And the effect is predictable: somebody measured on entry punctuality enters on time. They fill in Monday morning for the five days ahead, or carry over the previous week without rereading it. The indicator becomes good and the report becomes false — the exact opposite of the aim.
What I report: the overall share of late entry and its measured effect on the correction rate. That is an argument for changing the arrangement, not for summoning anybody. what-is-not-measured.pdfMeasuring people on the tool that already measures them
✎ Framework · no ranking of entry punctuality
What it would give: an accurate ranking of who enters late. It is technically available and it looks harmless — a matter of administrative rigour, not performance.
Why I still refuse: because the activity report is already the most sensitive document your teams fill in. It is used to bill, therefore to measure. Adding a measure of how it is filled in means measuring people on the tool that already measures them.
And the effect is predictable: somebody measured on entry punctuality enters on time. They fill in Monday morning for the five days ahead, or carry over the previous week without rereading it. The indicator becomes good and the report becomes false — the exact opposite of the aim.
What I report: the overall share of late entry and its measured effect on the correction rate. That is an argument for changing the arrangement, not for summoning anybody. what-is-not-measured.pdfMeasuring people on the tool that already measures them
✎ Framework · no ranking of entry punctuality
Can you send the approved timesheets to the client?
Yes — once approved, I send them, pass them to billing and chase what is missing. It is the approval itself that stays with a person.
What I do alone as soon as you open it: sending an approved timesheet to the client, in the format and at the rhythm you set; passing it to billing; chasing missing timesheets at day 2 then day 5; and the monthly recap by engagement.
What stays with a person, and why it is not caution: the approval. An approved timesheet triggers billing and stands behind what was done. Whoever approves declares that the time described is the time spent — that is a declaration, and a declaration has an author.
What it gives you back: today an approved timesheet waits on average six days before going out. On monthly billing, six days of lag become a whole month when approval lands after the cut-off.
What I keep doing in every case: I create an "unallocated" line for the 12 % of time that fits no engagement, and I do not spread it pro rata. Spreading produces false figures that look exactly like accurate ones — and the client pays them. automatic-sending_human-approval.pdfWhat goes alone · why approval has an author
⛓ Source · six days between approval and sending, 12 % unallocated time
What I do alone as soon as you open it: sending an approved timesheet to the client, in the format and at the rhythm you set; passing it to billing; chasing missing timesheets at day 2 then day 5; and the monthly recap by engagement.
What stays with a person, and why it is not caution: the approval. An approved timesheet triggers billing and stands behind what was done. Whoever approves declares that the time described is the time spent — that is a declaration, and a declaration has an author.
What it gives you back: today an approved timesheet waits on average six days before going out. On monthly billing, six days of lag become a whole month when approval lands after the cut-off.
What I keep doing in every case: I create an "unallocated" line for the 12 % of time that fits no engagement, and I do not spread it pro rata. Spreading produces false figures that look exactly like accurate ones — and the client pays them. automatic-sending_human-approval.pdfWhat goes alone · why approval has an author
⛓ Source · six days between approval and sending, 12 % unallocated time
What do you keep?
The timesheets as the people approved them, and nothing that lets people be compared with each other.
What is kept: the timesheets as approved, the share of unallocated time and its detail, corrected allocations with their reason, the sendings and their timestamp, and the chasers issued.
What is erased, deliberately: what I had pre-filled before correction. A proposal kept beside the approved timesheet reads as an accusation — "the agent had put something else". That is not what anybody is after, and it alone would get the tool rejected.
What the trace has already enabled: spotting 31 entries allocating cross-cutting time to a client engagement — the client was paying for something else. And seeing that 68 % of timesheets are entered on the last day of the week, for five days gone by. That is not a discipline problem: it is a memory problem, and it is fixed by a one-minute daily entry, not by a reprimand.
What I do produce by default, and it is the figure that can be fixed: 68 % of timesheets entered on the last day of the week for five days gone by, and the 31 reworked allocations that follow from it. The ranking of who enters late is lawful, and I produce it if you decide so: the file is supplied. What I tell you first: it would make entry fast, not accurate — the lateness rate would fall, the 31 reworks would stay, and the accuracy of timesheets is exactly what you bill. The decision is yours; both figures are on the table. what-you-keep_timesheets.pdf5 items kept · the proposal erased, deliberately
⛓ Source · 31 allocations corrected, 68 % entered on the last day
What is kept: the timesheets as approved, the share of unallocated time and its detail, corrected allocations with their reason, the sendings and their timestamp, and the chasers issued.
What is erased, deliberately: what I had pre-filled before correction. A proposal kept beside the approved timesheet reads as an accusation — "the agent had put something else". That is not what anybody is after, and it alone would get the tool rejected.
What the trace has already enabled: spotting 31 entries allocating cross-cutting time to a client engagement — the client was paying for something else. And seeing that 68 % of timesheets are entered on the last day of the week, for five days gone by. That is not a discipline problem: it is a memory problem, and it is fixed by a one-minute daily entry, not by a reprimand.
What I do produce by default, and it is the figure that can be fixed: 68 % of timesheets entered on the last day of the week for five days gone by, and the 31 reworked allocations that follow from it. The ranking of who enters late is lawful, and I produce it if you decide so: the file is supplied. What I tell you first: it would make entry fast, not accurate — the lateness rate would fall, the 31 reworks would stay, and the accuracy of timesheets is exactly what you bill. The decision is yours; both figures are on the table. what-you-keep_timesheets.pdf5 items kept · the proposal erased, deliberately
⛓ Source · 31 allocations corrected, 68 % entered on the last day
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 reporting rhythms. 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 574 € excl. VAT / month
Periodic report
Prepares the document in your template, on the schedule you set.
Flagging variances
Highlights the changes against the previous period, without interpreting them.
Collection from your tools
Gathers the data where it already is, with no re-keying.
Controls and safeguards
These 4 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
Sources, access rights and handling of questions with no answer
Data protection, rights and human oversight
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.
Integration with existing tools without double entry
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.
Project tracking
For tracking tasks and deadlines between two reviews, a dedicated agent completes the picture.
Project tracking agent (tasks, deadlines) from 499 € excl. VAT / month Project tracking →Accounting
For financial data and its reconciliation, a dedicated accounting agent steps in.
Accounting agent (summaries, anomalies) from 781 € excl. VAT / month Accounting →
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 analysis time can a team win back?
By automating collection and formatting, effort moves from gathering figures to interpreting them. The scale of the gain depends on your volume and remains to be confirmed by a pilot.
Collecting the period's data
Formatting into the template
Spotting the variances
Illustrative, non-contractual reference points, to be confirmed by a pilot on the number of reports and how often they fall due. The commentary and the interpretation of variances belong to the manager: the agent restates the figures, it does not explain them.
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 reporting agent (templates, schedule, collection), installed and operated for you. Prices exclude VAT — annual subscription, the time it takes for the gains to settle in.
Agility
Setup + controlled subscription
6,955 € excl. VAT setup
then 574 € 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
959 € 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
10,430 € excl. VAT setup
then 774 € 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 steering
Your indicators do not leave the organisationLocal inference or an isolated resource hosted in France; no activity data entrusted to a third party, and no indicator used to train a model.
Data in France, under French lawYour activity data and internal indicators: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The team manager keeps the decisionThe agent produces pre-filled reports, verifiable and editable; no approval is automated.
Human oversight & traceabilityOn the number of reports and how often they fall due: systematic logging and monitoring, compliant with the AI Act.
Going further
Related resources
Frequently asked questions
Your questions, our answers
Does the agent interpret the variances?
No. It flags them and sets out the components of the indicator across both periods. The explanation draws on your knowledge of the ground — and that is what gives the report its value.
Does the agent circulate the report?
No. It prepares the document; reviewing, commenting and circulating remain human acts. A report binds whoever signs it.
Where does the data come from?
From your existing tools, connected during the design phase. The agent reads where the data already is, with no re-keying and no migration.
Are our indicators 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 activity data is not used to train a third-party model.
Can we have several templates?
Yes, as many as there are periodic reports to produce, each with its own schedule and data scope. The configuration is done during the design phase.
How long does it take to deploy this assistant?
A few weeks as a rule, depending on the number of data sources and templates, after a free audit and then a phase of design, integration and testing.
Related use cases
Other agents for steering
Let's talk
Let us estimate the potential across your reports
15 minutes to scope your templates and your deadlines — hosted in France, supervised, with no commitment.