● B2B offer — Steering & data
Automated reporting: your sources consolidated, on time
A report that arrives late is no longer any use for deciding. Most of the delay comes from the consolidation: fetching the data from each system, aligning it, laying it out. Your agent does that to the deadlines you set, matches the sources, flags the inconsistencies between them and delivers the dashboard ready. Hosted in France: your activity figures stay with you. The analysis and the judgement calls stay with the management accountant.
Hosted in France
Activity figures protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Reporting · hosted in France
Produce the weekly dashboard.
Dashboard produced from your sources: the metrics you track, in your usual layout.
Every value states which source it comes from and the date it was read.
One inconsistency is flagged between two systems on the same metric.
🔗 Sourced · every value carries its origin
Every value states which source it comes from and the date it was read.
One inconsistency is flagged between two systems on the same metric.
🔗 Sourced · every value carries its origin
That gap between the two systems — which one do I take?
I set out both values, their source, the date read and the scope each covers.
Choosing which source prevails rests with management accounting: it bears on how your results are read.
✎ Support · gap set out, human decision
Choosing which source prevails rests with management accounting: it bears on how your results are read.
✎ Support · gap set out, human decision
Local inference · no data outside the EU
Data consolidated in France
Sovereign by designLocal inference or hosting in France
GDPR & AI Act: governed deploymentTraceability & human oversight
TurnkeyDesigned, installed and operated for you
The management accountant decidesThe agent prepares, never rules
✦ In brief
A Blue Lemon Agent reporting agent consolidates your sources, produces your dashboards to the deadlines you set and ties every value to its origin and the date it was read. It flags inconsistencies between systems rather than smoothing them over. It runs on local inference or is hosted in France: your activity figures are entrusted to no one, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
6
steering uses ready to deploy
0
decision taken without human approval
These figures describe our offer, not results measured at a client. How large the gain is on your number of sources and the frequency of your editions is confirmed by a pilot.
The context
What does an AI agent bring to your steering?
A dashboard is worth what its freshness and its traceability are worth. Consolidating automatically, while keeping the origin of every figure, brings steering back to its real rhythm.
! The issue
A useful report is one that is available at the moment of deciding and whose every figure can be traced back to its source. The agent consolidates your systems at the deadline set, ties every value to its origin and its date, and brings out the gaps between sources — so the discussion is about decisions, not about whether the figures can be trusted.
✓ Our answer
You receive a traceable dashboard: every value carries its source and its date, and inconsistencies between systems are set out rather than averaged. That is what makes it possible to base a decision on it. Local inference or an isolated resource hosted in France: your activity figures, which describe your actual performance, do not leave the company.
The decisive point
Your activity figures and your steering metrics: sovereignty & compliance
Your steering metrics describe the company's actual performance. 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 consolidated data and your dashboards: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
For your activity figures and your steering metrics, the architecture aims to reduce exposure to the Cloud Act and FISA 702; being located in France or in the European Union does not, on its own, guarantee immunity.
Isolated resource
No pooling: an environment strictly dedicated to your company and its metrics.
Every value carries its origin
The source and the date read are attached to every figure; encryption, role-based access and logging of every edition produced.
AI Act: governed deployment
The agent is strictly in support; no gap between sources is smoothed over or settled automatically; traceability and human oversight from end to end.
What depends on the architecture chosen These points are not general guarantees: they are settled deployment by deployment, in the quotation.
- The applicable location is that of the architecture set out in the quotation and verified before commissioning.
- Local execution is announced only for the configuration explicitly described and accepted in the quotation.
- The applicable isolation depends on the deployment mode set out in the quotation; no dedicated isolation is presumed.
- Roles and permissions are configured and accepted for the identities and systems actually connected.
- The events logged, their content, their retention period and who may access them are defined for the deployment chosen.
For metrics tied to a financial transaction or a regulated announcement, SecNumCloud and reinforced hosting are options depending on your requirements. A single architecture is designed to answer both the GDPR and extraterritorial exposure. Designed for deployment in line with the GDPR and the AI Act, after the processing, roles and context-specific risks have been assessed.
Demonstration
See the agent at work
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.
Reporting · hosted in France
Four flags this morning, and they are all about my own figures.
· The same indicator reads 847 in one report and 912 in another. Both are right, and I will say why.
· A report rerun over the same period gives a different result. Three per cent apart, with no error involved.
· Three branches reported nothing this month. They counted as zero in the average.
· No recipient can redo my calculations. That is the defect I am fixing first. morning-watch_4-flags.pdf847 and 912, both right · 3 branches counted as zero
⛓ Source · 41 reports, 6 sources, production log
· The same indicator reads 847 in one report and 912 in another. Both are right, and I will say why.
· A report rerun over the same period gives a different result. Three per cent apart, with no error involved.
· Three branches reported nothing this month. They counted as zero in the average.
· No recipient can redo my calculations. That is the defect I am fixing first. morning-watch_4-flags.pdf847 and 912, both right · 3 branches counted as zero
⛓ Source · 41 reports, 6 sources, production log
How can the same indicator be two different right figures?
Because they do not count the same thing, and the indicator's name does not say so.
What I record: "orders this month" reads 847 in the management report and 912 in the sales report. Tracing back: the first counts invoiced orders, the second signed orders. The 65 difference are signed and not yet invoiced.
Why nobody had seen it: both reports display "Orders". The choice is in the query, not in the title — and a query does not get read in a meeting.
What that produced: two departments defending two figures, each convinced the other was wrong. The discussion was about the reliability of the tools, when neither was at fault.
What I now do: every figure carries its definition in plain words, next to it — "orders invoiced between the 1st and the 31st", "orders signed, excluding cancelled". Not in an appendix: next to the figure.
What that produced: both figures now sit in the same report, with the 65 named. Nobody had to choose: the gap was the information. 847-and-912_65-gap.pdfThe choice is in the query, not in the title
⛓ Source · 847 invoiced, 912 signed, 65 named as the gap
What I record: "orders this month" reads 847 in the management report and 912 in the sales report. Tracing back: the first counts invoiced orders, the second signed orders. The 65 difference are signed and not yet invoiced.
Why nobody had seen it: both reports display "Orders". The choice is in the query, not in the title — and a query does not get read in a meeting.
What that produced: two departments defending two figures, each convinced the other was wrong. The discussion was about the reliability of the tools, when neither was at fault.
What I now do: every figure carries its definition in plain words, next to it — "orders invoiced between the 1st and the 31st", "orders signed, excluding cancelled". Not in an appendix: next to the figure.
What that produced: both figures now sit in the same report, with the 65 named. Nobody had to choose: the gap was the information. 847-and-912_65-gap.pdfThe choice is in the query, not in the title
⛓ Source · 847 invoiced, 912 signed, 65 named as the gap
How do you work, and where do you hand over?
I produce figures with what is needed to contest them.
What each figure carries: its plain-words definition, its cut-off date, its denominator, the sources queried, and what was missing at the time of calculation.
Routing follows who can act: a silent source goes to whoever owns it, before the report is circulated; two divergent definitions of the same name to whoever commissions both reports; a gap between two runs is explained inside the report itself; a report nobody opens to whoever asked for it.
With a monthly summary: indicators sharing a name with different definitions, silent sources, gaps between runs, and report open rates.
What this gives you back from the next issue: 847 and 912 stop being a dispute between two departments and become information again — the 65 signed but not yet invoiced orders, named next to the figure; the 3 silent branches are chased before circulation instead of counting as zero in the average; the 3% gap between two runs is explained inside the report itself. Across 41 reports and 6 sources, consolidation no longer costs days: it lands at the deadline you set, ready to be contested.
That is the gain: the meeting no longer opens on the reliability of the tools but on the decision — every figure arrives with its plain-words definition, its cut-off date, its denominator and what was missing at the time of calculation. Your figures do not leave the company: access by role, logged, withdrawn on a word, local inference or an isolated resource hosted in France.
The judgement call stays with the management controller, and I hand it over in minutes rather than at month end: I fill no gap — not with zero, not with the average, not with last month's — because a flagged gap gets repaired, and a filled one spreads. Tell me the deadline and the recipients, and the first issue goes out tomorrow.
✎ Framework · nothing filled in · no comment · no indicator per person
What each figure carries: its plain-words definition, its cut-off date, its denominator, the sources queried, and what was missing at the time of calculation.
Routing follows who can act: a silent source goes to whoever owns it, before the report is circulated; two divergent definitions of the same name to whoever commissions both reports; a gap between two runs is explained inside the report itself; a report nobody opens to whoever asked for it.
With a monthly summary: indicators sharing a name with different definitions, silent sources, gaps between runs, and report open rates.
What this gives you back from the next issue: 847 and 912 stop being a dispute between two departments and become information again — the 65 signed but not yet invoiced orders, named next to the figure; the 3 silent branches are chased before circulation instead of counting as zero in the average; the 3% gap between two runs is explained inside the report itself. Across 41 reports and 6 sources, consolidation no longer costs days: it lands at the deadline you set, ready to be contested.
That is the gain: the meeting no longer opens on the reliability of the tools but on the decision — every figure arrives with its plain-words definition, its cut-off date, its denominator and what was missing at the time of calculation. Your figures do not leave the company: access by role, logged, withdrawn on a word, local inference or an isolated resource hosted in France.
The judgement call stays with the management controller, and I hand it over in minutes rather than at month end: I fill no gap — not with zero, not with the average, not with last month's — because a flagged gap gets repaired, and a filled one spreads. Tell me the deadline and the recipients, and the first issue goes out tomorrow.
✎ Framework · nothing filled in · no comment · no indicator per person
Three branches counted as zero — does that skew everything?
It skews the average in one specific direction: downwards, and nobody sees it.
What I record: 3 branches out of 34 reported nothing this month. The report was averaging over 34, counting their values as zero.
What that gave: an average 8.8% below the real value, and an apparent fall nobody could explain — because the cause was not in the activity, it was in the collection.
The two ways of filling the gap, and what they cost: the other 31's average, or their own previous month. Both produce a complete and wrong figure — and above all they erase the fact that something is missing, which was the one piece of information that allowed a fix.
What I do: I calculate over 31, and write it — "average over 31 reporting branches, 3 with no data" — and I name the three to whoever manages collection, before the report goes out.
What that produced: of the 3, one had an export failure going back six weeks and nobody had noticed: its zeros went unremarked inside an average. A visible gap gets fixed; a zero reads as a result. 3-branches_8-8-percent.pdfA zero reads as a result, a gap gets fixed
⛓ Source · 3 branches of 34, average skewed by 8.8%, an export dead for 6 weeks
What I record: 3 branches out of 34 reported nothing this month. The report was averaging over 34, counting their values as zero.
What that gave: an average 8.8% below the real value, and an apparent fall nobody could explain — because the cause was not in the activity, it was in the collection.
The two ways of filling the gap, and what they cost: the other 31's average, or their own previous month. Both produce a complete and wrong figure — and above all they erase the fact that something is missing, which was the one piece of information that allowed a fix.
What I do: I calculate over 31, and write it — "average over 31 reporting branches, 3 with no data" — and I name the three to whoever manages collection, before the report goes out.
What that produced: of the 3, one had an export failure going back six weeks and nobody had noticed: its zeros went unremarked inside an average. A visible gap gets fixed; a zero reads as a result. 3-branches_8-8-percent.pdfA zero reads as a result, a gap gets fixed
⛓ Source · 3 branches of 34, average skewed by 8.8%, an export dead for 6 weeks
And the report that changes when you rerun it?
It changes because data arrives afterwards, and that is neither an error nor an anomaly.
What I record: the report for the period closed on the 31st gives 3% more when rerun three weeks later. These are not corrections: they are records that arrived late — deferred entries, field returns, adjustments.
Why it is a trust problem rather than a calculation one: somebody who reruns a report and finds something else concludes the tool is unstable. They are right to worry, and wrong about the cause.
What I do: every report carries its production date, not just its period — "data as at the 3rd of the following month" — and the rerun report carries its own. Both are right as at their date.
What I add: the stabilisation curve. On your data, a month settles in nineteen days: beyond that, late arrivals are negligible. That figure, not an intuition, is what should set the date of your committees.
What I deliver rather than wait: the report at the stated hour, dated on its first line. A dated, incomplete figure beats an accurate figure that comes too late — on the single condition that the date can be read, and it can. 3-percent_19-days.pdfBoth are right as at their date
⛓ Source · 3% gap between runs, stabilisation in 19 days
What I record: the report for the period closed on the 31st gives 3% more when rerun three weeks later. These are not corrections: they are records that arrived late — deferred entries, field returns, adjustments.
Why it is a trust problem rather than a calculation one: somebody who reruns a report and finds something else concludes the tool is unstable. They are right to worry, and wrong about the cause.
What I do: every report carries its production date, not just its period — "data as at the 3rd of the following month" — and the rerun report carries its own. Both are right as at their date.
What I add: the stabilisation curve. On your data, a month settles in nineteen days: beyond that, late arrivals are negligible. That figure, not an intuition, is what should set the date of your committees.
What I deliver rather than wait: the report at the stated hour, dated on its first line. A dated, incomplete figure beats an accurate figure that comes too late — on the single condition that the date can be read, and it can. 3-percent_19-days.pdfBoth are right as at their date
⛓ Source · 3% gap between runs, stabilisation in 19 days
Why supply the query with the report?
Because a figure nobody can redo is not verifiable, it is merely believed.
What I record: of 41 reports circulated, none let its recipient recover the figure other than by trusting me. Nobody complained — which is exactly what worries me.
What I now supply with every report: the plain-words definition, the list of sources and their dates, the number of rows kept and set aside with the reason for setting aside, and an extract of the calculation's rows for the indicators that allow it.
What it produced immediately: two errors found by recipients in three months. A filter had been excluding a customer category since a classification change; a join was losing orders with no product reference. Neither was detectable in the result — the figures were plausible.
What it costs: longer reports, and questions. The questions are the benefit, not the cost.
What I put on the slide, and it fits on one line: the figure, its denominator and its period. An indicator stripped of those two is not a summary: it is a different figure — and it is the one that gets quoted six months later with nobody able to tie it back to anything. 41-reports_2-errors-found.pdfA figure nobody can redo is merely believed
⛓ Source · 41 reports, 2 errors found by recipients in 3 months
What I record: of 41 reports circulated, none let its recipient recover the figure other than by trusting me. Nobody complained — which is exactly what worries me.
What I now supply with every report: the plain-words definition, the list of sources and their dates, the number of rows kept and set aside with the reason for setting aside, and an extract of the calculation's rows for the indicators that allow it.
What it produced immediately: two errors found by recipients in three months. A filter had been excluding a customer category since a classification change; a join was losing orders with no product reference. Neither was detectable in the result — the figures were plausible.
What it costs: longer reports, and questions. The questions are the benefit, not the cost.
What I put on the slide, and it fits on one line: the figure, its denominator and its period. An indicator stripped of those two is not a summary: it is a different figure — and it is the one that gets quoted six months later with nobody able to tie it back to anything. 41-reports_2-errors-found.pdfA figure nobody can redo is merely believed
⛓ Source · 41 reports, 2 errors found by recipients in 3 months
Can you comment on the results, say what is going well?
No. I can describe a variation; I cannot know whether it is good.
What I can write: "down 12% on the previous month", "lowest value in the last twelve months", "gap larger than the series' usual variation". Those are facts about numbers.
What I never write: "disappointing result", "strong performance", "worrying trend". A fall in orders is bad if you were seeking growth, neutral if you discontinued a range, and good if you deliberately narrowed your offer. I do not know what you decided — and a wrong comment in a report steers a whole meeting.
What I do instead: I flag what is unusual in the series, without labelling it, and I supply what allows somebody to go and look: the breakdown carrying most of the gap.
And the rule that holds every chart: the scale starts at zero. A 2% variation then looks like a 2% variation — and two successive editions compare at a glance, without having to read the axis. what-is-never-written.pdfA fall is bad, neutral or good depending on what you decided
✎ Framework · no judgement on a result · chart scales from zero
What I can write: "down 12% on the previous month", "lowest value in the last twelve months", "gap larger than the series' usual variation". Those are facts about numbers.
What I never write: "disappointing result", "strong performance", "worrying trend". A fall in orders is bad if you were seeking growth, neutral if you discontinued a range, and good if you deliberately narrowed your offer. I do not know what you decided — and a wrong comment in a report steers a whole meeting.
What I do instead: I flag what is unusual in the series, without labelling it, and I supply what allows somebody to go and look: the breakdown carrying most of the gap.
And the rule that holds every chart: the scale starts at zero. A 2% variation then looks like a 2% variation — and two successive editions compare at a glance, without having to read the axis. what-is-never-written.pdfA fall is bad, neutral or good depending on what you decided
✎ Framework · no judgement on a result · chart scales from zero
And the recurring editions, the ones we run every Monday?
Periodic dashboards go out on the deadlines you set, in your usual layout — not in mine.
What the calendar you wrote produces:
· 4 weekly dashboards, Monday 06:30: activity by branch, receivables, complaints, lead times;
· 7 monthly dashboards, on the 3rd at 07:00, including the board's;
· 3 quarterly dashboards, two of them for a third party — your bank and your insurer.
The layout is yours: your templates, your colours, your column order, your totals in the same place as before. A dashboard that changes appearance gets read from scratch, and the time you thought you had saved goes straight back out.
What every edition carries, without exception: the plain-language definition of each indicator, the cut-off date, the denominator, and the list of sources with the time each was last refreshed. Across the 14 editions, all 14 carry those four.
The gain, on your figures: consolidation took two and a half days per monthly edition, or 17 h 30 a month for the seven. It now takes 40 minutes of review. 16 h 50 returned a month, 202 hours over the year — more than five working weeks, a week counted as 35 hours and the rounding done downwards.
What periodicity makes possible and nobody asked for: an edition unchanged over four runs is flagged. Three of your weekly dashboards have not been opened by anyone in eleven weeks — I tell you, I keep producing them, and you decide whether to stop them.
A figure against me: 6 editions out of 168 went out late this quarter, all on the 3rd of the month when it falls on a Monday — two sources refresh on Monday at 08:00, after my production time. Those editions now go out at 09:00 on that day, with the shift noted, rather than at 07:00 on two day-old sources. No edition late since, and none produced on a stale source without saying so.
✎ Framework · periodic dashboards produced on the deadlines you set
What the calendar you wrote produces:
· 4 weekly dashboards, Monday 06:30: activity by branch, receivables, complaints, lead times;
· 7 monthly dashboards, on the 3rd at 07:00, including the board's;
· 3 quarterly dashboards, two of them for a third party — your bank and your insurer.
The layout is yours: your templates, your colours, your column order, your totals in the same place as before. A dashboard that changes appearance gets read from scratch, and the time you thought you had saved goes straight back out.
What every edition carries, without exception: the plain-language definition of each indicator, the cut-off date, the denominator, and the list of sources with the time each was last refreshed. Across the 14 editions, all 14 carry those four.
The gain, on your figures: consolidation took two and a half days per monthly edition, or 17 h 30 a month for the seven. It now takes 40 minutes of review. 16 h 50 returned a month, 202 hours over the year — more than five working weeks, a week counted as 35 hours and the rounding done downwards.
What periodicity makes possible and nobody asked for: an edition unchanged over four runs is flagged. Three of your weekly dashboards have not been opened by anyone in eleven weeks — I tell you, I keep producing them, and you decide whether to stop them.
A figure against me: 6 editions out of 168 went out late this quarter, all on the 3rd of the month when it falls on a Monday — two sources refresh on Monday at 08:00, after my production time. Those editions now go out at 09:00 on that day, with the shift noted, rather than at 07:00 on two day-old sources. No edition late since, and none produced on a stale source without saying so.
✎ Framework · periodic dashboards produced on the deadlines you set
We would like the detail per salesperson. Is that possible?
Yes, and I build it with the file that makes it safe. In this job, it is the indicator's DEFINITION that decides whether it will hold.
What I deliver to install it: the prior information notice (art. L1222-4 of the Labour Code), the works council consultation file, a proportionate scope. The French DPA restates that an employer has the power to frame and monitor staff activity.
The requirement I add, specific to this job: a named indicator carries the same items as any other — plain-language definition, cut-off date, denominator, sources. Without them you get this morning's figure: "orders this month" worth 847 in one report and 912 in another, both correct.
Why it matters more when the figure carries a name: a definition gap on an aggregate makes people argue about a trend. The same gap on a named indicator makes people argue about a person in their annual review, with a figure they can neither reproduce nor contest.
So what I supply with it: the query that produces the figure, so the person concerned can reproduce it. Of 41 reports circulated, none allowed it. A figure nobody can reproduce is not verifiable: it is merely believed. named-indicator_same-requirement.pdfWhat opens · the 4 required items · the query supplied
⛓ Source · French DPA, art. L1222-4 · 847 against 912 on one indicator, 41 non-reproducible reports
What I deliver to install it: the prior information notice (art. L1222-4 of the Labour Code), the works council consultation file, a proportionate scope. The French DPA restates that an employer has the power to frame and monitor staff activity.
The requirement I add, specific to this job: a named indicator carries the same items as any other — plain-language definition, cut-off date, denominator, sources. Without them you get this morning's figure: "orders this month" worth 847 in one report and 912 in another, both correct.
Why it matters more when the figure carries a name: a definition gap on an aggregate makes people argue about a trend. The same gap on a named indicator makes people argue about a person in their annual review, with a figure they can neither reproduce nor contest.
So what I supply with it: the query that produces the figure, so the person concerned can reproduce it. Of 41 reports circulated, none allowed it. A figure nobody can reproduce is not verifiable: it is merely believed. named-indicator_same-requirement.pdfWhat opens · the 4 required items · the query supplied
⛓ Source · French DPA, art. L1222-4 · 847 against 912 on one indicator, 41 non-reproducible reports
What do you keep?
Enough to reproduce every report identically — including last year's, with that period's data.
What is kept: each indicator's definition and its modification history, the cut-off date, the sources and the query, the missing data and how it was counted, and the successive versions of a report replayed.
Why missing data is kept as missing: 3 agencies out of 34 reported nothing this month. Counted as zero, they pull the average downwards, and nobody sees it. I count them as absent and write it beneath the figure — that is the difference between an average over 31 agencies and a false average over 34.
Why the successive versions: the report for the period closed on the 31st gives 3 % more when replayed a week later. That is neither an error nor an anomaly: the data arrives afterwards. Without both versions nobody can explain the gap, and one of the two figures is taken for wrong.
What I carry, even with the named indicator open: the variation and the breakdown that explains it — "down 12 %", and the line carrying most of the gap. The judgement stays with the management accountant: a fall is bad if you were seeking growth and good if you narrowed your offer, and they alone know which. A report that judges saves people from thinking — and that is exactly what it would end up being asked to do. what-you-keep_reporting.pdf5 items kept · a year-old report reproduced identically
⛓ Source · 3 agencies absent of 34, 3 % gap on replay
What is kept: each indicator's definition and its modification history, the cut-off date, the sources and the query, the missing data and how it was counted, and the successive versions of a report replayed.
Why missing data is kept as missing: 3 agencies out of 34 reported nothing this month. Counted as zero, they pull the average downwards, and nobody sees it. I count them as absent and write it beneath the figure — that is the difference between an average over 31 agencies and a false average over 34.
Why the successive versions: the report for the period closed on the 31st gives 3 % more when replayed a week later. That is neither an error nor an anomaly: the data arrives afterwards. Without both versions nobody can explain the gap, and one of the two figures is taken for wrong.
What I carry, even with the named indicator open: the variation and the breakdown that explains it — "down 12 %", and the line carrying most of the gap. The judgement stays with the management accountant: a fall is bad if you were seeking growth and good if you narrowed your offer, and they alone know which. A report that judges saves people from thinking — and that is exactly what it would end up being asked to do. what-you-keep_reporting.pdf5 items kept · a year-old report reproduced identically
⛓ Source · 3 agencies absent of 34, 3 % gap on replay
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 rhythms and several scopes of reporting. 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 499 € excl. VAT / month
Consolidation across sources
Gathers the data where it already sits, with no migration and no re-keying.
Periodic dashboards
Produces your editions to the deadlines set, in your usual layout.
Flagging inconsistencies
Sets out the gaps between systems on the same metric, without smoothing them.
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
Explain assumptions, uncertainties and the limits of the analyses
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.
Reconcile controlled sources with data quality and lineage
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.
Activity reports
For periodic documents with commentary, a dedicated agent takes over.
Activity report assistant from 574 € excl. VAT / month Activity reports →Multiple integrations
To synchronise the systems upstream of the reporting, a dedicated agent exists.
Multi-integration agent (CRM + ERP) from 583 € excl. VAT / month Multiple integrations →Accounting
For accounting data and its reconciliation, a dedicated agent completes the picture.
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 freshness can steering gain?
By consolidating automatically at the deadline, the production of the dashboard moves closer to the moment it is needed. How large the gain is depends on your volume and remains to be confirmed by a pilot.
Consolidating the sources
Laying out the dashboard
Checking consistency between systems
Indicative figures, not contractual, to be confirmed by a pilot on your number of sources and the frequency of your editions. Choosing which source prevails when there is a gap, and analysing the results, belong to management accounting.
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 (consolidation across sources, scheduled editions), installed and operated for you.
Agility
Setup + controlled subscription
6,740 € excl. VAT setup
then 499 € 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
874 € 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
10,215 € excl. VAT setup
then 699 € 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 figures do not leave the companyLocal 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 figures and your steering metrics: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The management accountant keeps the decisionThe agent produces dashboards and summaries, which can be checked and altered; no approval is automated.
Human oversight & traceabilityOn your number of sources and the frequency of your editions: systematic logging and tracking, in line with the AI Act.
Going further
Related resources
Frequently asked questions
Your questions, our answers
Can a figure be traced back to its origin?
Yes, systematically: every value carries its source and the date it was read. A dashboard whose figures cannot be justified does not let you decide.
What does the agent do when two systems disagree?
It sets out both values with their source, their date and their scope — it does not average them and does not choose. Choosing which source prevails bears on how your results are read: it falls to management accounting.
Does the agent interpret the results?
No. It produces and records; the analysis and the judgement calls stay human. Automatic commentary on results would give a false impression of a conclusion.
Do we have to migrate our data?
No. The agent reads where the data already sits, in your existing systems. The integration is defined at the design stage.
Are our indicators protected?
Yes. Hosting 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 figures are not used to train a third-party model.
How long does it take to deploy this agent?
A few weeks as a rule, depending on the number of sources and how complex the metrics are, after a free audit then a phase of design, integration and testing.
Related use cases
Other agents for steering
Let's talk
Let's size up the potential in your reporting
15 minutes to map your sources and your deadlines — hosted in France, supervised, with no commitment.