● Public sector offer — Steering & resources
Forecasting & planning: projections whose assumptions are known
Anticipating footfall at a service, how benefit claims will move or what staffing will be needed calls for projections whose basis can be discussed. Your agent builds those projections from your history and shows for each one its assumptions, its reference periods and its range. Hosted in France: your activity data stays within your administration. Allocating resources rests with your decisions.
Hosted in France
Activity data protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Forecasting & planning · hosted in France
What footfall should we expect at this service next quarter?
Projection produced with its range and the reference periods used.
The assumptions are shown: seasonality observed, campaign deadlines, known changes of scope.
One exceptional period has been set aside: this is stated and explained.
🔗 Sourced · activity history, assumptions listed
The assumptions are shown: seasonality observed, campaign deadlines, known changes of scope.
One exceptional period has been set aside: this is stated and explained.
🔗 Sourced · activity history, assumptions listed
How many officers should we assign to this service?
I provide the projected workload and its range, along with the workload per officer observed over comparable periods.
Allocating staff rests with your organisational decisions and your social dialogue: it is not in the data.
✎ Support · workload projected, organisational decision
Allocating staff rests with your organisational decisions and your social dialogue: it is not in the data.
✎ Support · workload projected, organisational decision
Local inference · no data outside the EU
History hosted in France
Sovereign by designLocal inference or hosting in France
GDPR & AI Act: governed deploymentTraceability & human oversight
TurnkeyDesigned, installed and operated for you
Your steering departments decideThe agent prepares, never rules
✦ In brief
A Blue Lemon Agent forecasting agent projects footfall, benefit claims and resource needs from your history, showing for every projection its assumptions, its reference periods and its range. Allocating resources rests with your decisions. It runs on local inference or is hosted in France: your activity data stays with you, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
6
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 services tracked and the depth of your history is confirmed by a pilot.
The context
What does an AI agent bring to your steering?
A projection whose assumptions are visible can be discussed in committee and corrected; that is what makes it useful.
! The issue
Planning calls for a projection and the ability to discuss its basis. A single figure with no context invites dispute; a projection that comes with its assumptions, its reference periods and its range can be worked with. The agent produces the second, flagging the exceptional periods it has set aside.
✓ Our answer
Your steering departments have projections that can be explained, service by service, and the workload observed over comparable periods. Allocating staff and resources rests with your organisational decisions and your social dialogue — it is not in the data. Local inference or an isolated resource hosted in France: your activity and footfall data does not leave the administration.
The decisive point
Your activity and footfall data: sovereignty & compliance
Your activity and footfall data describes how your services run. Here is how it is protected.
Local inference
The agent can run on a machine belonging to your organisation: no activity data and no projection leaves the network.
Hosting in France
Otherwise, a dedicated and isolated resource hosted in France, under French law — your activity history and your resources: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
For your activity and footfall data, the architecture aims to reduce exposure to the Cloud Act and FISA 702; being located in France or in the European Union does not, on its own, guarantee immunity.
Isolated resource
No pooling: an environment strictly dedicated to your administration and its services.
Assumptions and ranges shown
Every projection states its assumptions, its reference periods and its range; encryption, role-based access and logging of the projections produced.
AI Act: governed deployment
The agent is strictly in support; no allocation of resources is decided 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 projections tied to a reorganisation in preparation, 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.
Forecasting & planning · hosted in France
Four flags this morning on your volumes.
· One department's requests are up 34% over three months, and it is not the season: the three previous years fall over the same period.
· A load peak is due in September in another department, as it has been every year for four years. This year two officers are on leave that week.
· One department's backlog has fallen 3% a month since February. At that rate it clears in January — the only good news of the four, and it deserves saying.
· A series you track has been unusable since April: the counting method changed, and the values before and after do not compare. morning-watch_4-flags.pdf4 flags · 1 piece of good news
⛓ Source · request volumes 2022-2026, leave planner, counting log
· One department's requests are up 34% over three months, and it is not the season: the three previous years fall over the same period.
· A load peak is due in September in another department, as it has been every year for four years. This year two officers are on leave that week.
· One department's backlog has fallen 3% a month since February. At that rate it clears in January — the only good news of the four, and it deserves saying.
· A series you track has been unusable since April: the counting method changed, and the values before and after do not compare. morning-watch_4-flags.pdf4 flags · 1 piece of good news
⛓ Source · request volumes 2022-2026, leave planner, counting log
The 34% rise — will it continue?
I do not know, and I am going to tell you why that is the right answer.
What I record: +34% over three months, whereas the three previous years show −8%, −11% and −6% over the same months. It is not the season.
What I looked for: a coinciding event. I find one — a scheme opened on 12 May, and the rise starts on 18 May. Six days apart, consistent with word of mouth.
Why I do not project: a rise tied to a scheme opening always does one of two things — it falls back once the pent-up demand is served, or it settles in. Three months cannot tell those apart, and extending the curve would give a September figure that would look like a forecast.
What I can do instead, and it is useful: give you both scenarios with figures, and what will separate them. If the curve softens before mid-September, it was catch-up. If not, it is the new level — and that means three more officers, not two. rise-34pc_2-scenarios.pdf2 scenarios · the signal that separates them
⛓ Source · volumes May-August 2026, same months 2023-2025, scheme opening date
What I record: +34% over three months, whereas the three previous years show −8%, −11% and −6% over the same months. It is not the season.
What I looked for: a coinciding event. I find one — a scheme opened on 12 May, and the rise starts on 18 May. Six days apart, consistent with word of mouth.
Why I do not project: a rise tied to a scheme opening always does one of two things — it falls back once the pent-up demand is served, or it settles in. Three months cannot tell those apart, and extending the curve would give a September figure that would look like a forecast.
What I can do instead, and it is useful: give you both scenarios with figures, and what will separate them. If the curve softens before mid-September, it was catch-up. If not, it is the new level — and that means three more officers, not two. rise-34pc_2-scenarios.pdf2 scenarios · the signal that separates them
⛓ Source · volumes May-August 2026, same months 2023-2025, scheme opening date
How did you spot all that? And who do you tell?
I watch, continuously, what you have opened to me: your volumes, your backlogs, your planners, and the dates of your schemes — those dates explain half the breaks in a series.
Routing follows the horizon: a foreseeable peak six weeks out goes to the head of service and the planning lead, both together, because the answer is a leave decision; a break in a series to whoever produces the data, not to whoever reads it; a new trend to the head of service, with both scenarios.
With a chase: 7 days, except on a peak less than six weeks out — 48 hours, because beyond that leave is booked and the trade-off gets expensive.
Then a monthly summary: by series and by department, never by officer.
What this morning already gives back to the service: a September peak known six weeks ahead, with two officers on leave that week — the trade-off is still open, and it will not be in a fortnight, a 34% rise traced to its origin, a scheme opened on 12 May for a rise starting on the 18th, and a backlog shrinking 3% a month and gone by January, which is worth saying too. From tomorrow: leave booked once rather than undone and redone, deadlines kept for the resident when the request comes in and not three weeks later, and a series broken since April repaired at source — the counting method changed, and whoever produces the data now knows it. The access is yours: volumes, backlogs, planners and scheme dates opened by role, logged, withdrawn with a word; your activity data stays inside your administration. Staffing and leave stay with the head of service, and I hand them back in minutes: the peak dated, both scenarios costed, the signal that will separate them — if the curve bends before mid-September it is catch-up; if not it is the new level, and that means three more officers, not two. What the net catches, and how often it tightens: any peak less than six weeks out comes up within 48 h, to the two people who decide together; any break in a series goes to whoever produces the data; the rest is chased at 7 days, with a monthly summary by series and by department, never by officer. The next step is ready: both September scenarios are costed and the leave calendar for that week is out — tell me which one we take, and I will hand back the matching roster.
✎ Proposal · watch and chases to be configured — you set the thresholds
Routing follows the horizon: a foreseeable peak six weeks out goes to the head of service and the planning lead, both together, because the answer is a leave decision; a break in a series to whoever produces the data, not to whoever reads it; a new trend to the head of service, with both scenarios.
With a chase: 7 days, except on a peak less than six weeks out — 48 hours, because beyond that leave is booked and the trade-off gets expensive.
Then a monthly summary: by series and by department, never by officer.
What this morning already gives back to the service: a September peak known six weeks ahead, with two officers on leave that week — the trade-off is still open, and it will not be in a fortnight, a 34% rise traced to its origin, a scheme opened on 12 May for a rise starting on the 18th, and a backlog shrinking 3% a month and gone by January, which is worth saying too. From tomorrow: leave booked once rather than undone and redone, deadlines kept for the resident when the request comes in and not three weeks later, and a series broken since April repaired at source — the counting method changed, and whoever produces the data now knows it. The access is yours: volumes, backlogs, planners and scheme dates opened by role, logged, withdrawn with a word; your activity data stays inside your administration. Staffing and leave stay with the head of service, and I hand them back in minutes: the peak dated, both scenarios costed, the signal that will separate them — if the curve bends before mid-September it is catch-up; if not it is the new level, and that means three more officers, not two. What the net catches, and how often it tightens: any peak less than six weeks out comes up within 48 h, to the two people who decide together; any break in a series goes to whoever produces the data; the rest is chased at 7 days, with a monthly summary by series and by department, never by officer. The next step is ready: both September scenarios are costed and the leave calendar for that week is out — tell me which one we take, and I will hand back the matching roster.
✎ Proposal · watch and chases to be configured — you set the thresholds
Give me the load forecast for the last quarter.
Four departments projected, a fifth I refuse to project.
For the four: each figure comes with the series it is drawn from, the years of history available, and the assumption carrying it. For example: "1,240 requests in November — series 2021-2025, five years, assumption: the observed seasonality repeats and the scheme opened in May does not affect this department."
Why the assumption sits beside the figure and not in a footnote: a forecast gets copied. The figure travels, the assumption stays behind. Three weeks later somebody defends a budget with a number whose conditions nobody remembers.
The fifth department: I do not project it. Its series has been broken since April — the counting method changed, and the values before and after do not compare. A projection on that series would be a wrong figure of normal appearance, which is worse than no figure at all.
What I propose for that fifth: either you give me the conversion rule between the two counting methods, or we wait until we have twelve months of the new series. Q4-forecast_4-departments.pdf4 projected · 1 refused · assumptions written
⛓ Source · 2021-2025 series for 5 departments, counting log
For the four: each figure comes with the series it is drawn from, the years of history available, and the assumption carrying it. For example: "1,240 requests in November — series 2021-2025, five years, assumption: the observed seasonality repeats and the scheme opened in May does not affect this department."
Why the assumption sits beside the figure and not in a footnote: a forecast gets copied. The figure travels, the assumption stays behind. Three weeks later somebody defends a budget with a number whose conditions nobody remembers.
The fifth department: I do not project it. Its series has been broken since April — the counting method changed, and the values before and after do not compare. A projection on that series would be a wrong figure of normal appearance, which is worse than no figure at all.
What I propose for that fifth: either you give me the conversion rule between the two counting methods, or we wait until we have twelve months of the new series. Q4-forecast_4-departments.pdf4 projected · 1 refused · assumptions written
⛓ Source · 2021-2025 series for 5 departments, counting log
The broken series — can we really get nothing from it?
We can, but not a forecast — and the distinction is not a detail.
What can be drawn from it: the values since April compare with each other. Four months is enough to see an internal trend, and it is rising 2% a month.
What cannot be drawn from it: a comparison with last year, a seasonality, or a six-month projection. Those three need the history, and the history does not speak the same language.
What would repair it: the conversion rule. If someone in the department knows what changed in April — a widened scope, a different counting unit, a duplicate removed — the whole series becomes usable again in a day.
What I do meanwhile: I flag the break on every table where that series appears, rather than let it read as a fall or a rise. A table showing a 40% drop in April tells a false story, and it will tell it to everyone who opens it.
What that has already spared you: this break appears in two documents already circulated internally. broken-series_april.pdf4 usable months · 2 circulated documents to correct
✎ Support · conversion rule to be supplied — the series becomes usable again
What can be drawn from it: the values since April compare with each other. Four months is enough to see an internal trend, and it is rising 2% a month.
What cannot be drawn from it: a comparison with last year, a seasonality, or a six-month projection. Those three need the history, and the history does not speak the same language.
What would repair it: the conversion rule. If someone in the department knows what changed in April — a widened scope, a different counting unit, a duplicate removed — the whole series becomes usable again in a day.
What I do meanwhile: I flag the break on every table where that series appears, rather than let it read as a fall or a rise. A table showing a 40% drop in April tells a false story, and it will tell it to everyone who opens it.
What that has already spared you: this break appears in two documents already circulated internally. broken-series_april.pdf4 usable months · 2 circulated documents to correct
✎ Support · conversion rule to be supplied — the series becomes usable again
And benefit claims — do they project like the rest?
No, and it is the only series where the deadline counts for more than the season. A benefit claim does not follow a rhythm: it follows a closing date.
What your three campaigns say, measured: 2023 1,186 claims, 2024 1,305, 2025 1,420. 61% arrive in October, the last month of the campaign, and 38% of the total in the final ten days. The 2022 campaign is neutralised: its closing date was moved twice, and its shares compare to nothing.
The 2026 projection, with its assumption written beside it: “1,470 to 1,630 benefit claims over the campaign — series 2023-2025, three campaigns, assumption: the scale and the means threshold stay as set by the resolution of 3 February 2026, and the closing date stays 31 October.” The mid-course check: 612 claims received by 31 August, that is 41% of the bottom of the range and 37% of the top — across the three reference campaigns, the share observed at the same date runs from 37% to 42%. The range holds.
What that gives over the final ten days, and this is where the load falls: 558 to 619 claims, 22 minutes median processing — more than 204 hours at the bottom of the range, more than 226 at the top, over ten working days. An ordinary ten-day stretch from March to September is 4 claims a working day, 40 in all, more than 14 hours: the end of the campaign weighs more than thirteen ordinary stretches. Staffing is not to be found in these series — it rests with your organisation and with staff consultation; I hand you the projected load, its range and its date, six weeks before it lands.
The figure that does not flatter me: for 2025 I projected 1,258 benefit claims; 1,420 arrived — 11.4% below. The explanation is a date, not a curve: the closing date was extended by fifteen days by the resolution of 16 October 2025, and 162 claims arrived inside that extension window. I was projecting on reference periods whose volumes I had read without having read their deadlines.
What I changed, on two points: campaign deadlines are now read from your register of resolutions, and any change of date re-runs the projection the same day, naming the text that moved it; and the projection comes out as a range, never as a single value. Re-run on 2023 and 2024 with the dates actually resolved: 3.1% and 2.4% out, inside the announced range both times. benefit-claims-projection_1470-1630_closing-31-october.pdf38% of the total in ten days · the 11.4% miss of 2025 explained by a date
⛓ Sourced · campaigns 2023-2025, register of resolutions, 612 claims by 31 August
What your three campaigns say, measured: 2023 1,186 claims, 2024 1,305, 2025 1,420. 61% arrive in October, the last month of the campaign, and 38% of the total in the final ten days. The 2022 campaign is neutralised: its closing date was moved twice, and its shares compare to nothing.
The 2026 projection, with its assumption written beside it: “1,470 to 1,630 benefit claims over the campaign — series 2023-2025, three campaigns, assumption: the scale and the means threshold stay as set by the resolution of 3 February 2026, and the closing date stays 31 October.” The mid-course check: 612 claims received by 31 August, that is 41% of the bottom of the range and 37% of the top — across the three reference campaigns, the share observed at the same date runs from 37% to 42%. The range holds.
What that gives over the final ten days, and this is where the load falls: 558 to 619 claims, 22 minutes median processing — more than 204 hours at the bottom of the range, more than 226 at the top, over ten working days. An ordinary ten-day stretch from March to September is 4 claims a working day, 40 in all, more than 14 hours: the end of the campaign weighs more than thirteen ordinary stretches. Staffing is not to be found in these series — it rests with your organisation and with staff consultation; I hand you the projected load, its range and its date, six weeks before it lands.
The figure that does not flatter me: for 2025 I projected 1,258 benefit claims; 1,420 arrived — 11.4% below. The explanation is a date, not a curve: the closing date was extended by fifteen days by the resolution of 16 October 2025, and 162 claims arrived inside that extension window. I was projecting on reference periods whose volumes I had read without having read their deadlines.
What I changed, on two points: campaign deadlines are now read from your register of resolutions, and any change of date re-runs the projection the same day, naming the text that moved it; and the projection comes out as a range, never as a single value. Re-run on 2023 and 2024 with the dates actually resolved: 3.1% and 2.4% out, inside the announced range both times. benefit-claims-projection_1470-1630_closing-31-october.pdf38% of the total in ten days · the 11.4% miss of 2025 explained by a date
⛓ Sourced · campaigns 2023-2025, register of resolutions, 612 claims by 31 August
The September peak with two officers on leave — what do we do?
Nothing urgent — and that is exactly why it had to be seen now.
What I record: week 38 concentrates +42% of requests compared with an ordinary week, and it has done so for four years with unusual regularity. Two officers out of six are on leave that week, booked in March.
What that gives if nothing moves: four officers for a load of nine. The backlog built that week takes on average five weeks to clear — I measured it across the four years.
What I rule out at once, and the mechanism is simple: moving the leave. Leave booked in March is leave taken, and a tool that suggests moving it teaches officers to stop booking early — and then predictability itself disappears, and with it these six weeks of notice.
What I propose instead: three levers that touch no leave — bringing forward by a week what can be brought forward (I have identified 180 cases), shifting the chasing campaign scheduled for the same week, and preparing standard replies for the three reasons making up 60% of the peak.
Six weeks out, those three levers are available. Two weeks out, only leave is left. week-38-peak_3-levers.pdf4 years of regularity · 3 levers touching no leave
⛓ Source · week 38 from 2022 to 2025, leave planner, reasons behind the peak
What I record: week 38 concentrates +42% of requests compared with an ordinary week, and it has done so for four years with unusual regularity. Two officers out of six are on leave that week, booked in March.
What that gives if nothing moves: four officers for a load of nine. The backlog built that week takes on average five weeks to clear — I measured it across the four years.
What I rule out at once, and the mechanism is simple: moving the leave. Leave booked in March is leave taken, and a tool that suggests moving it teaches officers to stop booking early — and then predictability itself disappears, and with it these six weeks of notice.
What I propose instead: three levers that touch no leave — bringing forward by a week what can be brought forward (I have identified 180 cases), shifting the chasing campaign scheduled for the same week, and preparing standard replies for the three reasons making up 60% of the peak.
Six weeks out, those three levers are available. Two weeks out, only leave is left. week-38-peak_3-levers.pdf4 years of regularity · 3 levers touching no leave
⛓ Source · week 38 from 2022 to 2025, leave planner, reasons behind the peak
And the good news about the backlog?
It is real, and it deserves the same rigour as the rest.
What I record: the backlog is falling 3% a month since February, steadily. At that rate it drops below the threshold you set yourselves in January 2027.
The assumption carrying that figure: that the inflow stays at the level of the last six months. It is the most fragile assumption in anything I have given you this morning — because the 34% rise in the first flag concerns a neighbouring department, and some of those requests end up here.
What that means: the two flags are connected. If the rise settles in, the January date moves out by about two months. I say so now rather than let you announce January and then explain March.
What I hand you so the date can be announced without risk: both projections, each with its assumption written out — January 2027 if the inflow stays at the level of the last six months, March 2027 if the neighbouring department's rise settles in —, and the point in the series where it will be known which of the two holds: three monthly readings are enough. The date you announce is signed, because it commits the department: good news given without its assumption turns round exactly like bad news, and it costs more because it will have been announced. backlog_clears-in-january.pdf−3%/month · 1 fragile assumption · 2 months of possible slippage
⛓ Source · backlog February-August 2026, six-month inflow, threshold set by the department
What I record: the backlog is falling 3% a month since February, steadily. At that rate it drops below the threshold you set yourselves in January 2027.
The assumption carrying that figure: that the inflow stays at the level of the last six months. It is the most fragile assumption in anything I have given you this morning — because the 34% rise in the first flag concerns a neighbouring department, and some of those requests end up here.
What that means: the two flags are connected. If the rise settles in, the January date moves out by about two months. I say so now rather than let you announce January and then explain March.
What I hand you so the date can be announced without risk: both projections, each with its assumption written out — January 2027 if the inflow stays at the level of the last six months, March 2027 if the neighbouring department's rise settles in —, and the point in the series where it will be known which of the two holds: three monthly readings are enough. The date you announce is signed, because it commits the department: good news given without its assumption turns round exactly like bad news, and it costs more because it will have been announced. backlog_clears-in-january.pdf−3%/month · 1 fragile assumption · 2 months of possible slippage
⛓ Source · backlog February-August 2026, six-month inflow, threshold set by the department
What do you refuse to forecast?
Three families — and for each, I give you what replaces it. A refused forecast with no substitute is just an evasion.
1. Anything with insufficient history. A scheme opened four months ago cannot be projected. What I give instead: the values since it opened, comparable with each other, with the month-by-month rate of change. Four months is enough to see a slope, not to project a year — and knowing that beats a one-year figure.
2. Anything whose series is broken. What I give instead: the values on either side of the break, and the exact date of the break with what happened that day. A broken series often explains more than a continuous one.
3. Anything depending on a decision not yet taken. What I give instead: one figure per scenario, with the decision that switches between them and the date by which it must be taken. That is not a forecast, it is an arbitration calendar.
What it changes concretely: on last quarter's workload, four services are projected and a fifth receives three scenarios. Nobody is left without a figure. three-families_and-their-substitute.pdfThe 3 families · what each receives instead · the arbitration calendar
⛓ Source · 4 services projected, 1 in scenarios, broken series dated
1. Anything with insufficient history. A scheme opened four months ago cannot be projected. What I give instead: the values since it opened, comparable with each other, with the month-by-month rate of change. Four months is enough to see a slope, not to project a year — and knowing that beats a one-year figure.
2. Anything whose series is broken. What I give instead: the values on either side of the break, and the exact date of the break with what happened that day. A broken series often explains more than a continuous one.
3. Anything depending on a decision not yet taken. What I give instead: one figure per scenario, with the decision that switches between them and the date by which it must be taken. That is not a forecast, it is an arbitration calendar.
What it changes concretely: on last quarter's workload, four services are projected and a fifth receives three scenarios. Nobody is left without a figure. three-families_and-their-substitute.pdfThe 3 families · what each receives instead · the arbitration calendar
⛓ Source · 4 services projected, 1 in scenarios, broken series dated
What do you keep from all this?
The forecasts and the hypothesis behind them — and above all what I forecast against what actually happened.
What is kept: the projected figure, the series and the hypothesis behind it, the years of history, the date, and — most importantly — the actual outcome beside it.
Why the actual beside it is the decisive item: a forecast without its outcome never improves. It is also the only way to know which families of volumes I am reliable on and which I am not — and that information is worth more than any individual forecast.
What it made visible: a rise of +34 % over three months in one service, when the three previous years showed a fall at the same period. It is not the season. I do not say whether it will continue — I say it has no precedent in your data, and that is what justifies going to ask why.
And the peak seen in time: week 38 concentrates +42 % of demand with two agents on leave. Flagged six weeks ahead, while arbitration still costs little. The same finding at two weeks would have left only bad options. what-you-keep_forecasting.pdf5 items kept · the actual beside it, without which a forecast never improves
⛓ Source · +34 % with no precedent, week 38 at +42 % flagged six weeks ahead
What is kept: the projected figure, the series and the hypothesis behind it, the years of history, the date, and — most importantly — the actual outcome beside it.
Why the actual beside it is the decisive item: a forecast without its outcome never improves. It is also the only way to know which families of volumes I am reliable on and which I am not — and that information is worth more than any individual forecast.
What it made visible: a rise of +34 % over three months in one service, when the three previous years showed a fall at the same period. It is not the season. I do not say whether it will continue — I say it has no precedent in your data, and that is what justifies going to ask why.
And the peak seen in time: week 38 concentrates +42 % of demand with two agents on leave. Flagged six weeks ahead, while arbitration still costs little. The same finding at two weeks would have left only bad options. what-you-keep_forecasting.pdf5 items kept · the actual beside it, without which a forecast never improves
⛓ Source · +34 % with no precedent, week 38 at +42 % flagged six weeks ahead
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 objects of planning. 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 770 € incl. VAT / month
Footfall projection
Projects a service's activity from its history and its seasonality.
Benefit claims
Projects how claims will move, including campaign deadlines.
Assumptions shown
States the reference periods, the range and the periods set aside.
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.
Data analyst
For day-to-day analysis of your data, a dedicated agent takes it on.
Natural-language querying of data and dashboards from 730 € incl. VAT / month Data analyst →Demand forecasting agent
History, models.
Demand forecasting agent from 563 € incl. VAT / month Discover the agent →Recurring appointment scheduling assistant
Diary, reminders.
Recurring appointment scheduling assistant from 474 € incl. VAT / month Discover the agent →Logistics / round planning agent
Constraints, simple optimisation.
Logistics / round planning agent from 543 € incl. VAT / month Discover the agent →
Does your need fall outside this?
Book the free audit →
Build your agent
In 15 minutes we identify the agent that will give your staff the most time back — without oversizing the project.
The gain
How many services can an administration project in detail?
By taking on the building and the documenting, the effort shifts towards deciding on resources. How large the gain is depends on your volume and remains to be confirmed by a pilot.
Building a projection
Documenting the assumptions
Setting aside exceptional periods
Indicative figures, not contractual, to be confirmed by a pilot on your number of services tracked and the depth of your history. Allocating staff and resources rests with your organisational decisions and your social dialogue: it is not in the data.
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 forecasting and planning agent (projections, assumptions, ranges), installed and operated for you.
Agility
Setup + controlled subscription
10,765 € incl. VAT setup
then 770 € incl. 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
1,365 € incl. 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
15,241 € incl. VAT setup
then 1,005 € incl. VAT/month · + hardware from 2,989 € (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 include VAT at 20%: as a public body that is not VAT-registered, you cannot reclaim it.
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 activity data stays with youLocal inference or an isolated resource hosted in France; no activity data entrusted to a third party, no data used to train a model.
Data in France, under French lawYour activity and footfall data: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
Your steering departments keep the decisionThe agent produces projections that can be explained, checked and altered; no approval is automated.
Human oversight & traceabilityOn your number of services tracked and the depth of your history: systematic logging and tracking, in line with the AI Act.
Going further
Related resources
Frequently asked questions
Your questions, our answers
Does the agent decide how officers are assigned?
No. It provides the projected workload, its range and the workload observed over comparable periods. Allocation rests with your organisational decisions and your social dialogue.
How can a projection be checked?
Every projection shows its assumptions, its reference periods and the exceptional periods set aside, which makes them open to discussion.
Why a range?
Because it shows where the uncertainty lies and lets you size things in full knowledge. The central figure is provided with it.
How much history is needed?
The deeper the history, the clearer the seasonality. What can be used is assessed during the audit, with no assumptions about what you do not have.
Is your activity data 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.
How long does it take to deploy this agent?
A few months as a rule, depending on the number of services tracked and the quality of the history, after a free audit then phases of design, integration and testing.
Related use cases
Other agents for steering
Let's talk
Let's size up the potential in your steering
15 minutes to frame your services and your history — hosted in France, supervised, with no commitment.