● Public sector — Decision support & data analyst
AI decision-support agent: talking to your data in the public service
Preparing a dashboard, cross-checking files, producing a reliable figure for a council decision or a report takes a considerable amount of time — and often assumes command of a spreadsheet or a query language. Your AI agent turns a question asked in plain language into a figure and a chart, with its source. Hosted in France — on local inference or an isolated resource — public data stays under control. The public officer decides.
Hosted in France
Public data protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Decision-support assistant · hosted in France
For next week's finance committee, give me the trend in operating expenditure by department over 3 years, with a chart.
Data consolidated from the annual accounts and the general ledger. Operating expenditure moves from year N-2 to N with a moderate overall rise; two departments account for most of the gap (energy and payroll).
I have prepared a bar chart by department and a summary table, ready for the committee — to be checked.
⛓ Source · general ledger + annual accounts (isolated resource)
I have prepared a bar chart by department and a summary table, ready for the committee — to be checked.
⛓ Source · general ledger + annual accounts (isolated resource)
The energy item — how much more, and on which buildings?
The energy item rises markedly over the period, concentrated on three facilities (sports hall, town-centre school, town hall). Note: these figures mix volumes and tariff increases — the gap is not purely a drift in consumption.
I am preparing a reading note separating the volume effect from the price effect, for your approval.
✎ Action · note ready for review — the public officer approves
I am preparing a reading note separating the volume effect from the price effect, for your approval.
✎ Action · note ready for review — the public officer approves
Local inference · no data outside the EU
Data 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 public officer decidesThe AI agent assists, never decides alone
✦ In brief
In a local authority, a central government department or a public body, a Blue Lemon Agent agent lets you query your data in plain language: you ask the question in ordinary words, the agent produces the figure, the dashboard and the chart, with the source — no spreadsheet and no technical query. It cross-checks the internal sources through a single point of entry. It runs on local inference or is hosted in France on an isolated resource: public data is never exposed to a foreign service, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. The time saved goes back into analysis and decision-making. Illustrative reference points, to be validated by a pilot.
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
9
uses ready to deploy on this scope
0
decision taken without human approval
Reference points describing our offer, not results measured at a client. The scale of the gain is confirmed by a pilot on your own scope.
The context
Why decision support matters to the public sector — and why it hesitates
Elected members, directorates and audit bodies expect reliable figures, quickly, to steer public service. But producing an indicator often means a detour through a spreadsheet, and the data involved — finances, HR, the public — is among an administration's most sensitive.
! The issue
The administration is caught between a growing demand for data-driven steering (transparency, performance, accountability) and data scattered across tools that do not talk to each other. Yet most consumer AI solutions amount to entrusting financial, HR and citizen data to a third party, often hosted outside Europe and subject to the Cloud Act — which is unacceptable for public data.
✓ Our answer
Decision-support AI is only of interest to an administration if it is sovereign and confidential by design. Local inference or an isolated resource hosted in France, systematic human oversight, decisions reserved to the public officer: the time saved on preparing the figures is never paid for in lost confidentiality. The aim is not to replace human analysis, but to give back thinking time to interpret and decide, while respecting equal treatment.
The decisive point
Confidentiality of public data: sovereignty & compliance
An administration handles sensitive data — finances, staff, the public. Here is how the architecture of our agents protects it, source by source.
Local inference
The agent can run on a machine belonging to the authority: no data leaves the network, nothing passes through a cloud.
Hosting in France
Otherwise, a dedicated and isolated resource, hosted in France under French law — your data: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
Exposure of public data to the Cloud Act and FISA 702 is reduced by design; location alone does not guarantee immunity.
One isolated resource per entity
No pooling of public data: an environment strictly dedicated to your administration.
Encryption & controlled access
Encryption in transit and at rest, role-based access (RBAC) for elected members and directorates, strong authentication and logging.
AI Act: governed deployment
An agent strictly in support; no automated decision; 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.
- The encryption mechanisms in transit and at rest, their components and key management are those documented for the architecture chosen.
- Roles and permissions are configured and accepted for the identities and systems actually connected.
- The events logged, their content, their retention period and who may access them are defined for the deployment chosen.
For the most sensitive data, SecNumCloud and HDS 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.
The public body in this demonstration
Fictional public bodyTown of Rocheveyre — municipality of 28,600 inhabitants, finance and performance department (fictional municipality)
- Sector
- Municipality of 28,600 inhabitants — one main budget and 3 ancillary budgets (drinking water, car park, municipal cinema), 14 operational departments, 44 municipal buildings and a social action centre with its own budget
- Headcount
- 612 staff at 31 December; the finance and performance department has 9 people — a director, a management controller, 5 accounting officers and 2 performance analysts. The agent serves this department, the chief executive's office and the 14 operational departments; it touches neither payroll nor payment authorisation
- Public served
- 28,600 inhabitants, 33 elected members of the council, 4 standing committees — finance, works, education, social affairs — each meeting 8 times, i.e. 32 dashboards a year, and 9 full council sittings
- Order of magnitude
- €46.8m of real operating expenditure, €18.2m of capital expenditure, €24.1m of payroll; 74,200 payment orders and 21,600 revenue orders issued in the year; 6 statutory reports; 1,460 figure requests put to the finance department by elected members and departments, and 240 requests to cross-reference files
- Tools already in place
- Finance software (accounts, commitments, procurement), HR and payroll system, time and absence management, document management for council resolutions, facility booking software, school catering software, and 340 spreadsheets kept by the departments — the agent reads them, nothing is replaced or migrated
- Who decides what
- The director of finance and performance signs off every figure before release; the chief executive validates the papers put to committee; the finance councillor presents them to the council; opening a dashboard to a department or an elected member is decided by the chief executive's office
- Room for improvement
- Preparing a committee dashboard takes 26 hours, 15 h 36 of which is consolidation alone; an elected member's figure request comes back in 6 days; the 340 departmental spreadsheets are reconciled only once a year, at closing; and 9 indicators presented to committee exist in two versions depending on the source queried — 5 of them have already been presented in both values over the last three financial years
At Rocheveyre, nine people run the performance function of a municipality with 612 staff: 32 committee dashboards, 6 statutory reports, 12 monthly budget positions and 1,460 figure requests a year, served from five applications that do not talk to each other and 340 departmental spreadsheets. The time is not lost in analysis — it is lost in the consolidation that precedes it, and in the back-and-forth to establish which scope is being discussed. The exchanges below cover a year, from reconciling the sources to the twelve-month review put to the council. The agent is hosted in France on an isolated resource, reads the finance software, the HR system and the operational applications without ever writing to them, and no figure leaves the finance department without its director's sign-off.
This public body, its figures and the exchanges that follow were invented for the demonstration. They illustrate a common situation; they describe no real service.
Decision-support assistant · hosted in France
I have reviewed your 1,460 figure requests for the year and reconciled, indicator by indicator, your five sources — finance software, HR system, time management, facility booking, catering software — together with the 340 departmental spreadsheets. This is the reconciliation that year-end closing never leaves time for.
What the count says: 1,118 of the 1,460 requests bear on just 11 indicators — more than three in four, 76.6%. Real operating expenditure, budget execution by department, payroll, headcount, absence rate, supplier payment time, capital carry-forwards, grants to associations, facility attendance, meals served, contracts awarded. The 11 indicator sheets are written, each with its scope, its cut-off date and the query that produces it, set out in plain language — they are waiting for you.
What the three items you can check against your own diaries are worth:
· Preparing a committee dashboard is 26 hours end to end, 60% of which goes on extraction, consolidation and formatting — 15 h 36. After: 18%, i.e. 4 h 40. The analysis and the commentary do not move: that is the work you want to keep.
· A request to cross-reference files is 5 hours, 23% of which goes on reconciling identifiers, scopes and vintages — 1 h 09. After: 10%, i.e. 30 minutes.
· An elected member's figure request travels through 6 h 40 of cumulative time today, 8% of which is actual keying — 32 minutes. After: 6%, i.e. 24 minutes. On this item the fiche is cautious, and rightly so: what shifts is not the time, it is the lead time — 6 days of waiting brought down to the minute.
And here is the gap the reconciliation surfaced, which no one had ever measured: 9 indicators presented to committee carry two different values depending on the source queried. Five of them have already been presented in both values over the last three financial years — headcount at 31 December at 612 and at 574, payroll at €24.10m and at €23.42m, the absence rate at 7.4% and at 5.9%, operating expenditure at €46.8m and at €51.3m, grants to associations at €1.42m and at €1.68m.
The total time returned, item by item: 940 hours over the year — at 151.67 hours a month, more than six months of work returned to a nine-person department.
The next step, and it takes thirty minutes: you read the 9 lines and tell me, for each one, which scope is the reference. Your director of finance settles the version, and the 9 go out reconciled in next month's committee dashboard. data-landscape_1460-requests-11-indicators.pdf1,118 requests on 11 indicators, the three items costed, 940 hours identified
⛓ Sourced · 1,460 figure requests over 12 months, finance software, HR system, time management, 340 departmental spreadsheets, final accounts N-2 to N
What the count says: 1,118 of the 1,460 requests bear on just 11 indicators — more than three in four, 76.6%. Real operating expenditure, budget execution by department, payroll, headcount, absence rate, supplier payment time, capital carry-forwards, grants to associations, facility attendance, meals served, contracts awarded. The 11 indicator sheets are written, each with its scope, its cut-off date and the query that produces it, set out in plain language — they are waiting for you.
What the three items you can check against your own diaries are worth:
· Preparing a committee dashboard is 26 hours end to end, 60% of which goes on extraction, consolidation and formatting — 15 h 36. After: 18%, i.e. 4 h 40. The analysis and the commentary do not move: that is the work you want to keep.
· A request to cross-reference files is 5 hours, 23% of which goes on reconciling identifiers, scopes and vintages — 1 h 09. After: 10%, i.e. 30 minutes.
· An elected member's figure request travels through 6 h 40 of cumulative time today, 8% of which is actual keying — 32 minutes. After: 6%, i.e. 24 minutes. On this item the fiche is cautious, and rightly so: what shifts is not the time, it is the lead time — 6 days of waiting brought down to the minute.
And here is the gap the reconciliation surfaced, which no one had ever measured: 9 indicators presented to committee carry two different values depending on the source queried. Five of them have already been presented in both values over the last three financial years — headcount at 31 December at 612 and at 574, payroll at €24.10m and at €23.42m, the absence rate at 7.4% and at 5.9%, operating expenditure at €46.8m and at €51.3m, grants to associations at €1.42m and at €1.68m.
The total time returned, item by item: 940 hours over the year — at 151.67 hours a month, more than six months of work returned to a nine-person department.
The next step, and it takes thirty minutes: you read the 9 lines and tell me, for each one, which scope is the reference. Your director of finance settles the version, and the 9 go out reconciled in next month's committee dashboard. data-landscape_1460-requests-11-indicators.pdf1,118 requests on 11 indicators, the three items costed, 940 hours identified
⛓ Sourced · 1,460 figure requests over 12 months, finance software, HR system, time management, 340 departmental spreadsheets, final accounts N-2 to N
Nine indicators with two values is exactly what puts us in difficulty in the chamber. Where do these gaps come from, and how do we avoid creating more next year?
All nine come from the same cause, and it is not an arithmetic error: across the 9, zero addition mistakes. All nine come from an implicit scope. Two sources answer two different questions, each gives the right figure for its own question, and the word that separates them is written down nowhere.
The three costliest, to fix ideas:
· Headcount at 31 December — 612 against 574. The HR system counts every member of staff whose post exists; payroll counts those paid in December. The 38 in between are on career break, parental leave or outward secondment. Both figures are right; they do not answer the same question.
· Payroll — €24.10m against €23.42m. The accounting chapter also carries elected members' allowances, occupational health and statutory insurance, which payroll does not know about. €680,000 of gap, explained line by line.
· Absence rate — 7.4% against 5.9%. One denominator in calendar days, the other in working days. 1.5 points of gap for two words missing from a column heading.
What I have done on top, and what closing never leaves time for: I have written four scope-control rules, in the language of your own indicator sheets, and run them across 36 months of figures actually released — 96 dashboards and 18 reports. For each one I give you the number of alerts it would have raised, the share confirmed on examination, and what it would have let through — you choose on figures, not on a hunch:
· Rule A — a total of expenditure or income released without naming the budget concerned: 214 alerts over 36 months, 203 confirmed — 94.9%. The other 11 carried the mention in the document title, not in the table.
· Rule B — two values of the same indicator released less than 30 days apart: 88 alerts, 81 confirmed — 92.0%.
· Rule C — a headcount or a payroll figure released without the employment status retained: 76 alerts, 76 confirmed — 100%.
And here is what those three let through, because you must know it before you sign: 61 rates whose denominator had changed since the previous release fell into none of the three — absence moving from calendar days to working days is the textbook case. So I wrote rule D and ran it across the same 36 months: 61 alerts, 54 confirmed — 88.5%.
All four together: 439 alerts, 414 confirmed — 94.3%, and they cover 414 of the 447 divergences found over the period, i.e. 92.6%.
I advise against switching all four on the same day, and I cost that advice: all four together would produce 168 alerts in the first quarter, i.e. roughly 28 hours of examination for your management controller — in the middle of budget preparation. One rule per quarter is 7 hours each time, and you keep control of what you look at.
Bringing a rule into force stays with the department: a rule only goes live once approved by your director of finance — and that is precisely what makes the figure defensible in the chamber. I save you the writing and the measurement; the decision takes thirty minutes instead of a working group. scope-divergences_9-indicators-and-4-rules.pdfThe 9 two-valued indicators in detail, 4 rules tested over 36 months, 414 divergences of 447 covered
⛓ Sourced · 36 months of released figures (96 dashboards, 18 reports), indicator sheets, finance software, HR system
The three costliest, to fix ideas:
· Headcount at 31 December — 612 against 574. The HR system counts every member of staff whose post exists; payroll counts those paid in December. The 38 in between are on career break, parental leave or outward secondment. Both figures are right; they do not answer the same question.
· Payroll — €24.10m against €23.42m. The accounting chapter also carries elected members' allowances, occupational health and statutory insurance, which payroll does not know about. €680,000 of gap, explained line by line.
· Absence rate — 7.4% against 5.9%. One denominator in calendar days, the other in working days. 1.5 points of gap for two words missing from a column heading.
What I have done on top, and what closing never leaves time for: I have written four scope-control rules, in the language of your own indicator sheets, and run them across 36 months of figures actually released — 96 dashboards and 18 reports. For each one I give you the number of alerts it would have raised, the share confirmed on examination, and what it would have let through — you choose on figures, not on a hunch:
· Rule A — a total of expenditure or income released without naming the budget concerned: 214 alerts over 36 months, 203 confirmed — 94.9%. The other 11 carried the mention in the document title, not in the table.
· Rule B — two values of the same indicator released less than 30 days apart: 88 alerts, 81 confirmed — 92.0%.
· Rule C — a headcount or a payroll figure released without the employment status retained: 76 alerts, 76 confirmed — 100%.
And here is what those three let through, because you must know it before you sign: 61 rates whose denominator had changed since the previous release fell into none of the three — absence moving from calendar days to working days is the textbook case. So I wrote rule D and ran it across the same 36 months: 61 alerts, 54 confirmed — 88.5%.
All four together: 439 alerts, 414 confirmed — 94.3%, and they cover 414 of the 447 divergences found over the period, i.e. 92.6%.
I advise against switching all four on the same day, and I cost that advice: all four together would produce 168 alerts in the first quarter, i.e. roughly 28 hours of examination for your management controller — in the middle of budget preparation. One rule per quarter is 7 hours each time, and you keep control of what you look at.
Bringing a rule into force stays with the department: a rule only goes live once approved by your director of finance — and that is precisely what makes the figure defensible in the chamber. I save you the writing and the measurement; the decision takes thirty minutes instead of a working group. scope-divergences_9-indicators-and-4-rules.pdfThe 9 two-valued indicators in detail, 4 rules tested over 36 months, 414 divergences of 447 covered
⛓ Sourced · 36 months of released figures (96 dashboards, 18 reports), indicator sheets, finance software, HR system
Finance, payroll, absences, facility attendance: these are our most sensitive data. Where do they go, and what do I tell the council when the question comes?
Nowhere outside your walls, and that is verified rather than promised.
Where I run: on an isolated resource hosted in France, under French law, dedicated to Rocheveyre and to no one else — or on a machine in the town hall if you prefer local inference, in which case nothing even leaves the network. No data is passed to a foreign service: exposure to the Cloud Act and FISA 702 is reduced accordingly, without location alone ruling it out — it is documented for the configuration chosen. No pooling between authorities: a strictly dedicated environment.
What I read, and what is not open to me: I have read-only access to the finance software, to aggregated HR data, to time management, to the operational applications and to the 340 spreadsheets. I have no write access at all: I produce tables, charts and draft papers, and a member of staff takes them on. Individual payslips, bank details, medical reasons for sick leave and individual staff files are not open to me at all — they play no part in computing a payroll total or an absence rate, and article 5.1.c GDPR requires precisely that only what serves the purpose be exposed. Every access is withdrawn with a word, and the withdrawal takes effect within the second.
I plug into what you already have: finance software, HR system, document management, operational applications and spreadsheets. No migration, no change of tool, no data re-entry — I sit on top, and the day you change finance software it is the connector that changes, not your history.
What you tell the council, and can show: every read is logged with its date, its reason and the object consulted; every figure produced carries the records it is drawn from; access is opened by role — an elected member of the finance committee, an operational department and the chief executive's office do not see the same thing; and the record of processing activities is already written, legal basis by legal basis — public interest task, article 6.1.e GDPR, for the authority's performance management.
The next step I propose: a quarterly thirty-minute review where I show you the access rights actually used and those that served no purpose — a useless access right is one to close, and I bring you the list rather than asking you for it. sovereign-framework_hosting-access-traceability.pdfWhere the data lives, who sees what, what is withdrawn with a word
✎ Framework · hosting architecture, role access matrix, read log, the authority's record of processing activities
Where I run: on an isolated resource hosted in France, under French law, dedicated to Rocheveyre and to no one else — or on a machine in the town hall if you prefer local inference, in which case nothing even leaves the network. No data is passed to a foreign service: exposure to the Cloud Act and FISA 702 is reduced accordingly, without location alone ruling it out — it is documented for the configuration chosen. No pooling between authorities: a strictly dedicated environment.
What I read, and what is not open to me: I have read-only access to the finance software, to aggregated HR data, to time management, to the operational applications and to the 340 spreadsheets. I have no write access at all: I produce tables, charts and draft papers, and a member of staff takes them on. Individual payslips, bank details, medical reasons for sick leave and individual staff files are not open to me at all — they play no part in computing a payroll total or an absence rate, and article 5.1.c GDPR requires precisely that only what serves the purpose be exposed. Every access is withdrawn with a word, and the withdrawal takes effect within the second.
I plug into what you already have: finance software, HR system, document management, operational applications and spreadsheets. No migration, no change of tool, no data re-entry — I sit on top, and the day you change finance software it is the connector that changes, not your history.
What you tell the council, and can show: every read is logged with its date, its reason and the object consulted; every figure produced carries the records it is drawn from; access is opened by role — an elected member of the finance committee, an operational department and the chief executive's office do not see the same thing; and the record of processing activities is already written, legal basis by legal basis — public interest task, article 6.1.e GDPR, for the authority's performance management.
The next step I propose: a quarterly thirty-minute review where I show you the access rights actually used and those that served no purpose — a useless access right is one to close, and I bring you the list rather than asking you for it. sovereign-framework_hosting-access-traceability.pdfWhere the data lives, who sees what, what is withdrawn with a word
✎ Framework · hosting architecture, role access matrix, read log, the authority's record of processing activities
For next week's finance committee, give me the three-year trend of operating expenditure by department, with a chart.
Data consolidated from the general ledger and the final accounts, and the scope is written at the top of the table before the first figure: main budget only, excluding the 3 ancillary budgets and excluding the social action centre; real operating expenditure, accounting chapters 011, 012, 65, 66 and 67; cut off at 31 December of each financial year; N-2 and N-1 on approved final accounts, N on provisional final accounts as at 12 February. That is the line that spares you the trap question in the chamber, and it costs three rows of table.
What the three years say: €43.9m in N-2, €45.2m in N-1, €46.8m in N — €2.9m more over three years, an overall rise of 6.6%, moderate for the period.
Two departments carry most of the movement: energy, attached to technical services, for €1.24m, and payroll for €0.82m — €2.06m between them, i.e. 71.0% of the €2.9m. The other 12 departments share the remaining €0.84m, including +€0.19m on insurance and −€0.06m on grants to associations, the only line down across the three years.
What I have prepared, ready for committee — subject to your check: a bar chart by department across three years, with the two lines carrying the movement highlighted, a 14-row summary table, and the query that produces each row, written in plain language beneath the table. You have not opened a spreadsheet, and you can rebuild every figure in front of an elected member.
The time this returns: 15 h 36 of consolidation brought down to 4 h 40 — 60% of preparation time brought down to 18%. Across your 32 committee dashboards, 349 hours returned over the year, and they go back where they belong: analysis and commentary.
The figure that does not flatter me, and I publish it: of the 32 dashboards in the first year, 3 came back from your director of finance for a wrong scope — 9.4%. Two included the ancillary water budget in the main budget's operating expenditure; the third counted the social action centre twice. That is fixed, and not by a promise: the scope is now declared at the top and compared automatically against the previous dashboard's, and any change is flagged before the commentary. Across the 16 dashboards of the following half-year: 0 scope returns, 1 formatting return. I propose to keep publishing that rate every half-year — an agent whose errors are not measured is an agent that cannot be corrected. dashboard_finance-committee-3-years.pdfThe declared scope, the 14 departments across 3 years, the chart and the plain-language query
⛓ Sourced · general ledger and final accounts N-2 to N (isolated resource), finance department's log of returned dashboards
What the three years say: €43.9m in N-2, €45.2m in N-1, €46.8m in N — €2.9m more over three years, an overall rise of 6.6%, moderate for the period.
Two departments carry most of the movement: energy, attached to technical services, for €1.24m, and payroll for €0.82m — €2.06m between them, i.e. 71.0% of the €2.9m. The other 12 departments share the remaining €0.84m, including +€0.19m on insurance and −€0.06m on grants to associations, the only line down across the three years.
What I have prepared, ready for committee — subject to your check: a bar chart by department across three years, with the two lines carrying the movement highlighted, a 14-row summary table, and the query that produces each row, written in plain language beneath the table. You have not opened a spreadsheet, and you can rebuild every figure in front of an elected member.
The time this returns: 15 h 36 of consolidation brought down to 4 h 40 — 60% of preparation time brought down to 18%. Across your 32 committee dashboards, 349 hours returned over the year, and they go back where they belong: analysis and commentary.
The figure that does not flatter me, and I publish it: of the 32 dashboards in the first year, 3 came back from your director of finance for a wrong scope — 9.4%. Two included the ancillary water budget in the main budget's operating expenditure; the third counted the social action centre twice. That is fixed, and not by a promise: the scope is now declared at the top and compared automatically against the previous dashboard's, and any change is flagged before the commentary. Across the 16 dashboards of the following half-year: 0 scope returns, 1 formatting return. I propose to keep publishing that rate every half-year — an agent whose errors are not measured is an agent that cannot be corrected. dashboard_finance-committee-3-years.pdfThe declared scope, the 14 departments across 3 years, the chart and the plain-language query
⛓ Sourced · general ledger and final accounts N-2 to N (isolated resource), finance department's log of returned dashboards
The energy line — how much more is it, and on which buildings?
The energy line rises by €1.24m over the three years — €1.86m in N-2, €3.10m in N, i.e. +66.7%. It is concentrated on three facilities: the Peupliers sports hall, +€412,000; the town-centre school, +€286,000; the town hall, +€198,000. €896,000 between them out of €1,240,000, i.e. 72.3% of the rise — the other 41 buildings share the remaining €344,000.
And here is the warning that matters, together with what it takes to lift it: these figures mix volumes and tariffs, and presented as they stand they point the finger at teams that had nothing to do with it. I have separated the two, from the 1,184 energy invoices of the three years, each carrying kilowatt-hours and unit price: the price effect is €918,000, i.e. 74.0% of the rise; the volume effect is €322,000, i.e. 26.0%.
The fact that turns the reading around, and that will spare you an awkward moment in committee: the town-centre school has CUT its consumption by 4.1% in volume — €26,000 less — and still sees its bill rise by €286,000, because the price effect there is €312,000. Presenting its rise without that line means calling into question a team that did exactly what was asked of it.
The sports hall is the only one where volume explains a real share: +19.4% of kilowatt-hours, i.e. €144,000 of volume effect within its €412,000 — the covered hall came into service in September N-2, so 4 months in N-2 against 12 months thereafter. That is not overspend, it is one more facility, and the table says so with its commissioning date.
I have prepared a reading note separating volume effect from price effect, building by building, with the 1,184 invoices attached — it is ready to review, and you are the one who approves it before it enters the committee pack.
The next step I propose: run the same split across the other 41 buildings, for the June committee. Within their €344,000, the volume effect is €180,000 against €164,000 of price effect — that is the only part of the estate where volume wins, so the only place where an energy-saving measure has a measurable return. reading-note_energy-volume-effect-price-effect.pdf€1,240,000 broken down, €918,000 of price effect, the town-centre school down in volume
⛓ Sourced · 1,184 energy invoices N-2 to N (kWh and unit price), inventory of the 44 buildings, commissioning date of the covered hall
And here is the warning that matters, together with what it takes to lift it: these figures mix volumes and tariffs, and presented as they stand they point the finger at teams that had nothing to do with it. I have separated the two, from the 1,184 energy invoices of the three years, each carrying kilowatt-hours and unit price: the price effect is €918,000, i.e. 74.0% of the rise; the volume effect is €322,000, i.e. 26.0%.
The fact that turns the reading around, and that will spare you an awkward moment in committee: the town-centre school has CUT its consumption by 4.1% in volume — €26,000 less — and still sees its bill rise by €286,000, because the price effect there is €312,000. Presenting its rise without that line means calling into question a team that did exactly what was asked of it.
The sports hall is the only one where volume explains a real share: +19.4% of kilowatt-hours, i.e. €144,000 of volume effect within its €412,000 — the covered hall came into service in September N-2, so 4 months in N-2 against 12 months thereafter. That is not overspend, it is one more facility, and the table says so with its commissioning date.
I have prepared a reading note separating volume effect from price effect, building by building, with the 1,184 invoices attached — it is ready to review, and you are the one who approves it before it enters the committee pack.
The next step I propose: run the same split across the other 41 buildings, for the June committee. Within their €344,000, the volume effect is €180,000 against €164,000 of price effect — that is the only part of the estate where volume wins, so the only place where an energy-saving measure has a measurable return. reading-note_energy-volume-effect-price-effect.pdf€1,240,000 broken down, €918,000 of price effect, the town-centre school down in volume
⛓ Sourced · 1,184 energy invoices N-2 to N (kWh and unit price), inventory of the 44 buildings, commissioning date of the covered hall
And if an elected member asks me in the chamber where the sports hall figure comes from, what do I do? I am not going to reopen a spreadsheet in front of the council.
You reopen nothing: you click the sports hall bar, and the figure comes apart in front of you.
What the bar gives you, in the chamber, in three seconds: the 148 invoices that make it up, with their date, supplier, billing period, kilowatt-hours, unit price and the number of the payment order that settled them. An elected member who asks where a figure comes from gets the record, not an explanation.
What the table carries at the top, before the first figure: the scope — main budget, excluding ancillary budgets, excluding the social action centre —, the accounting chapters retained, and the data cut-off date. That is the rule I propose we hold on everything leaving the department: no figure goes out without its scope declared at the top, and two different scopes are never compared in the same chart. A change of scope is announced before the commentary — never after an elected member's question.
What the table carries at the foot: the query, written in plain language — « sum of payment orders on accounts 60612 and 60613 attached to the Peupliers sports hall, financial years N-2 to N, main budget ». It is there so the same question asked again in June returns the same figure, and so a member of staff who was not there can rerun it.
And the dashboard freezes: with a word, you stop its data at the date the summons went out. The figure no longer moves between the summons and the sitting — which is what prevents an elected member arriving with one value and you with another, both of them right.
What this changes over the year: your 1,460 figure requests come back in 6 days today; 1,118 of them fall under the 11 indicators and go back within the minute, with their scope and cut-off date. The other 342 go to your management controller with the file already assembled: the question as asked, the sources concerned, the possible scopes and the figures already extracted. She answers knowing, instead of answering while searching. dashboard_finance-committee-3-years.pdfThe 148 sports hall invoices in detail, the plain-language query, the freeze at summons
✎ Framework · indicator sheets, plain-language queries attached to dashboards, log of dashboard freezes
What the bar gives you, in the chamber, in three seconds: the 148 invoices that make it up, with their date, supplier, billing period, kilowatt-hours, unit price and the number of the payment order that settled them. An elected member who asks where a figure comes from gets the record, not an explanation.
What the table carries at the top, before the first figure: the scope — main budget, excluding ancillary budgets, excluding the social action centre —, the accounting chapters retained, and the data cut-off date. That is the rule I propose we hold on everything leaving the department: no figure goes out without its scope declared at the top, and two different scopes are never compared in the same chart. A change of scope is announced before the commentary — never after an elected member's question.
What the table carries at the foot: the query, written in plain language — « sum of payment orders on accounts 60612 and 60613 attached to the Peupliers sports hall, financial years N-2 to N, main budget ». It is there so the same question asked again in June returns the same figure, and so a member of staff who was not there can rerun it.
And the dashboard freezes: with a word, you stop its data at the date the summons went out. The figure no longer moves between the summons and the sitting — which is what prevents an elected member arriving with one value and you with another, both of them right.
What this changes over the year: your 1,460 figure requests come back in 6 days today; 1,118 of them fall under the 11 indicators and go back within the minute, with their scope and cut-off date. The other 342 go to your management controller with the file already assembled: the question as asked, the sources concerned, the possible scopes and the figures already extracted. She answers knowing, instead of answering while searching. dashboard_finance-committee-3-years.pdfThe 148 sports hall invoices in detail, the plain-language query, the freeze at summons
✎ Framework · indicator sheets, plain-language queries attached to dashboards, log of dashboard freezes
The education committee wants the full cost of school catering: what it costs, what families pay, how many staff work on it. Those figures sit in three different systems and nobody has ever put them side by side.
Cross-referenced and produced — and this is exactly the kind of question that has no business costing five hours.
The scope, at the top: main budget, education department, financial year N, 12 full months, 9 school restaurants. Three sources cross-referenced into a single answer: the finance software for expenditure and income, the HR system and time management for staff assigned, the catering software for meals served.
What it gives:
· Direct expenditure: €1,986,000 — food €742,000, staff €1,084,000, utilities and upkeep €160,000.
· Income from families: €1,042,000.
· Net cost to the municipality: €944,000.
· Meals served: 214,800, of which 208,400 billed to families.
· Full cost per meal: €9.24. Average price paid by families: €5.00. Net cost per meal: €4.39.
· 34 staff assigned, i.e. 28.4 full-time equivalents.
And one divergence I would rather name than smooth over: you will hear 214,800 meals on one side and 208,400 on the other. This is indicator no. 8 of the nine we reconciled: the 6,400 gap is the meals taken by staff and teachers, served but not billed to families. Both figures are right, they do not answer the same question — the table carries both, each with its definition, and the cost per meal is computed on meals served, not on meals billed.
The time this returns: a cross-referencing request is 5 hours today, 1 h 09 of which goes on reconciling identifiers, scopes and vintages; it now takes 30 minutes. Across your 240 cross-references in the year, 156 hours returned — and above all a single point of entry: you asked a question in plain language and opened none of the three systems.
The next step I propose: the same cross-reference on the 9 restaurants taken one by one. Cost per meal runs from €8.32 to €11.62 depending on the site, and the gap comes from the number of meals produced, not from food spend — that is a figure worth a council resolution, and it is already computed. cross-referenced-sources_full-cost-of-school-catering.pdfThree sources cross-referenced, €1,986,000 broken down, cost per meal across the 9 restaurants
⛓ Sourced · finance software (year N), HR system and time management, catering software (214,800 records), revenue orders issued to families
The scope, at the top: main budget, education department, financial year N, 12 full months, 9 school restaurants. Three sources cross-referenced into a single answer: the finance software for expenditure and income, the HR system and time management for staff assigned, the catering software for meals served.
What it gives:
· Direct expenditure: €1,986,000 — food €742,000, staff €1,084,000, utilities and upkeep €160,000.
· Income from families: €1,042,000.
· Net cost to the municipality: €944,000.
· Meals served: 214,800, of which 208,400 billed to families.
· Full cost per meal: €9.24. Average price paid by families: €5.00. Net cost per meal: €4.39.
· 34 staff assigned, i.e. 28.4 full-time equivalents.
And one divergence I would rather name than smooth over: you will hear 214,800 meals on one side and 208,400 on the other. This is indicator no. 8 of the nine we reconciled: the 6,400 gap is the meals taken by staff and teachers, served but not billed to families. Both figures are right, they do not answer the same question — the table carries both, each with its definition, and the cost per meal is computed on meals served, not on meals billed.
The time this returns: a cross-referencing request is 5 hours today, 1 h 09 of which goes on reconciling identifiers, scopes and vintages; it now takes 30 minutes. Across your 240 cross-references in the year, 156 hours returned — and above all a single point of entry: you asked a question in plain language and opened none of the three systems.
The next step I propose: the same cross-reference on the 9 restaurants taken one by one. Cost per meal runs from €8.32 to €11.62 depending on the site, and the gap comes from the number of meals produced, not from food spend — that is a figure worth a council resolution, and it is already computed. cross-referenced-sources_full-cost-of-school-catering.pdfThree sources cross-referenced, €1,986,000 broken down, cost per meal across the 9 restaurants
⛓ Sourced · finance software (year N), HR system and time management, catering software (214,800 records), revenue orders issued to families
I also need the payroll and the absence rate for that service. But I do not want a figure per member of staff: we have 34 people, and some sites have three.
Served in aggregate, and the threshold is written into the indicator sheet rather than left to the judgement of the moment.
What I give you: service payroll €1,084,000, 34 staff, 28.4 full-time equivalents. Absence rate 8.9% in working days, against 5.9% for the municipality as a whole — a gap of 3.0 points.
And I do not leave you with a bare figure — I compare it on two levels. Against your own history first, because that is the only benchmark that binds: this service was at 7.1% in N-2, 8.2% in N-1, 8.9% in N. The rise is steady; it is not accidental. Against sector orders of magnitude next, and I state them as indicative: catering and cleaning are commonly the most exposed services in a municipality, above the authority's average — you are not an exception, you are in a known situation.
Then I explain the gap, because a service-level rate means nothing until it is opened up: 62% of absence days come from 3 restaurants out of 9. The other 6 are at 5.4%, below the municipal average. This is not a service in trouble, it is three sites — and that is an entirely different conversation to have in committee.
The aggregation threshold, and this is a choice I propose rather than a constraint: I do not go below 5 staff in any grouping. Below five, an absence rate points at a person. You can move that threshold in either direction. Until your data controller settles it, it stays proposed, not acquired: a grouping that would fall below it is not produced — it is blocked and comes back to you for human validation, together with the grouping that would replace it.
Now the individual indicator, because the question will come: I know how to produce it, and I will produce it if you ask. An employer's power to monitor is recognised; it is the conditions that are regulated — proportionality of the indicator to the aim pursued, prior information of staff, and consultation of the competent body, the local social committee in the local civil service, as the French data protection authority sets them out in its note on monitoring employees' activity. The EU AI Act classes evaluation of workers' performance under its Annex III, point 4: high risk, with obligations applying from 2 December 2027 — regulated, not prohibited.
So my reservation is not legal, it is mechanical, and that is what counts: an absence rate per person becomes a team target, and the first thing that distorts is the declaration. Short sick days turn into booked annual leave, the rate falls, and you lose precisely the signal that would have told you which restaurant is struggling. You would lose the instrument along with the measurement.
What I propose instead, and it is already prepared: measurement by SITE — 9 restaurants, rate, days lost, unfilled posts — which tells you where it hurts without telling you who. One exception, and it is not one: whoever signs off a figure is named and dated. A signature is not a counter. cross-referenced-sources_full-cost-of-school-catering.pdfAggregated HR data, the 5-staff threshold, absence across the 9 sites
⛓ Sourced · HR system (aggregated data), time and absence management N-2 to N, HR indicator sheets, French data protection authority note on monitoring employees' activity
What I give you: service payroll €1,084,000, 34 staff, 28.4 full-time equivalents. Absence rate 8.9% in working days, against 5.9% for the municipality as a whole — a gap of 3.0 points.
And I do not leave you with a bare figure — I compare it on two levels. Against your own history first, because that is the only benchmark that binds: this service was at 7.1% in N-2, 8.2% in N-1, 8.9% in N. The rise is steady; it is not accidental. Against sector orders of magnitude next, and I state them as indicative: catering and cleaning are commonly the most exposed services in a municipality, above the authority's average — you are not an exception, you are in a known situation.
Then I explain the gap, because a service-level rate means nothing until it is opened up: 62% of absence days come from 3 restaurants out of 9. The other 6 are at 5.4%, below the municipal average. This is not a service in trouble, it is three sites — and that is an entirely different conversation to have in committee.
The aggregation threshold, and this is a choice I propose rather than a constraint: I do not go below 5 staff in any grouping. Below five, an absence rate points at a person. You can move that threshold in either direction. Until your data controller settles it, it stays proposed, not acquired: a grouping that would fall below it is not produced — it is blocked and comes back to you for human validation, together with the grouping that would replace it.
Now the individual indicator, because the question will come: I know how to produce it, and I will produce it if you ask. An employer's power to monitor is recognised; it is the conditions that are regulated — proportionality of the indicator to the aim pursued, prior information of staff, and consultation of the competent body, the local social committee in the local civil service, as the French data protection authority sets them out in its note on monitoring employees' activity. The EU AI Act classes evaluation of workers' performance under its Annex III, point 4: high risk, with obligations applying from 2 December 2027 — regulated, not prohibited.
So my reservation is not legal, it is mechanical, and that is what counts: an absence rate per person becomes a team target, and the first thing that distorts is the declaration. Short sick days turn into booked annual leave, the rate falls, and you lose precisely the signal that would have told you which restaurant is struggling. You would lose the instrument along with the measurement.
What I propose instead, and it is already prepared: measurement by SITE — 9 restaurants, rate, days lost, unfilled posts — which tells you where it hurts without telling you who. One exception, and it is not one: whoever signs off a figure is named and dated. A signature is not a counter. cross-referenced-sources_full-cost-of-school-catering.pdfAggregated HR data, the 5-staff threshold, absence across the 9 sites
⛓ Sourced · HR system (aggregated data), time and absence management N-2 to N, HR indicator sheets, French data protection authority note on monitoring employees' activity
The council votes in June on catering charges. I need a committee paper. Can you prepare it, or do I redo everything by hand?
It is written, it runs to four pages, and every figure in it carries the record it comes from and its date. You read it, correct it, sign it — it is an act of the authority, not the output of a machine.
What the paper carries:
· The full cost per meal — €9.24 — and its breakdown into three items, with the payment orders that produce it.
· The 8 income bands in force and what each pays, taken from your charging resolution of 24 June N-3, article by article.
· The trend in net cost across three financial years: €812,000, €878,000, €944,000 — +€66,000 per year, twice in a row.
· The split of the 208,400 billed meals across the 8 bands, and the 3,860 families concerned.
· The history of your own decisions: the 4 charging resolutions taken since 2019, the uprating chosen each time, and the net cost of the following year. That is the intelligence most often missing — the intelligence about your own acts.
And three costed options, because a paper that sets out without proposing leaves the work whole:
· Option 1 — uniform 3% uprating: +€31,000 of income, net cost down to €913,000, all 8 bands affected.
· Option 2 — 5% uprating on the 3 top bands only: +€27,000, net cost €917,000, 1,240 families affected out of 3,860. Almost the same yield as option 1, for a third of the families touched.
· Option 3 — grid unchanged, band 3 brought down to the social rate of €2.80: −€19,000 of income, net cost €963,000, 610 families better off.
Each option carries its assumption in plain sight, and that is what makes it arguable: meal numbers constant at 208,400. If attendance moves by 5%, the income swing is €52,000 either way — more than the gap between the three options. You must know that before voting, and it is written at the head of the options page, not in a footnote.
What the paper does not do for you, and that is as it should be: the choice. I bring you the three costed routes, their winners and losers named band by band, and the sensitivity of each; the decision is taken in committee, and it takes thirty minutes instead of a preparatory meeting. committee-paper_school-catering-charges.pdf4 sourced pages, 3 costed options, the attendance assumption and its sensitivity
⛓ Sourced · charging resolution of 24 June N-3, final accounts N-2 to N, 214,800 catering records, 4 charging resolutions since 2019
What the paper carries:
· The full cost per meal — €9.24 — and its breakdown into three items, with the payment orders that produce it.
· The 8 income bands in force and what each pays, taken from your charging resolution of 24 June N-3, article by article.
· The trend in net cost across three financial years: €812,000, €878,000, €944,000 — +€66,000 per year, twice in a row.
· The split of the 208,400 billed meals across the 8 bands, and the 3,860 families concerned.
· The history of your own decisions: the 4 charging resolutions taken since 2019, the uprating chosen each time, and the net cost of the following year. That is the intelligence most often missing — the intelligence about your own acts.
And three costed options, because a paper that sets out without proposing leaves the work whole:
· Option 1 — uniform 3% uprating: +€31,000 of income, net cost down to €913,000, all 8 bands affected.
· Option 2 — 5% uprating on the 3 top bands only: +€27,000, net cost €917,000, 1,240 families affected out of 3,860. Almost the same yield as option 1, for a third of the families touched.
· Option 3 — grid unchanged, band 3 brought down to the social rate of €2.80: −€19,000 of income, net cost €963,000, 610 families better off.
Each option carries its assumption in plain sight, and that is what makes it arguable: meal numbers constant at 208,400. If attendance moves by 5%, the income swing is €52,000 either way — more than the gap between the three options. You must know that before voting, and it is written at the head of the options page, not in a footnote.
What the paper does not do for you, and that is as it should be: the choice. I bring you the three costed routes, their winners and losers named band by band, and the sensitivity of each; the decision is taken in committee, and it takes thirty minutes instead of a preparatory meeting. committee-paper_school-catering-charges.pdf4 sourced pages, 3 costed options, the attendance assumption and its sensitivity
⛓ Sourced · charging resolution of 24 June N-3, final accounts N-2 to N, 214,800 catering records, 4 charging resolutions since 2019
The council is asking me for a twelve-month review. What do I put in front of them?
Four figures, and each one is checkable against your own logs — that is the only way to hold in front of a council.
· 940 hours returned to the department over the year, item by item: 349 h on the 32 committee dashboards, 156 h on the 240 file cross-references, 194 h on the 1,460 figure requests, 162 h on the 6 statutory reports, 78 h on the 12 monthly budget positions. At 151.67 hours a month, that is more than six months of work returned to nine people; at 35 hours a week, more than twenty-six weeks. No post cut, no post created: it is time returned to analysis, budget preparation and management control, which until now existed only as an intention.
· Lead time on a figure request: 6 days brought down to the minute for 1,118 of the 1,460 requests, with the other 342 going to the management controller with the file already assembled.
· 146 scope alerts over the year, 138 confirmed — 94.5%, where the reconciliation used to happen only at closing. And the 9 two-valued indicators are reconciled: across the year's 32 dashboards and 6 reports, none went out in two versions.
· 0 figures released without their scope, 0 dashboards published without your finance director's named sign-off, 0 automated decisions.
What I bring on top, and what waits for you every year: your budget orientation report — the one under article L. 2312-1 of the French general local authorities code — draws in part on data I already keep up to date. In a municipality of more than 10,000 inhabitants, that report covers the expected trend in the structure of headcount and staff costs, in the content set by article D. 2312-3 of the same code. The 11 corresponding indicators are extracted, dated, with the query that produces them, so that an elected member asking where a figure comes from gets the answer in the sitting. The report is still written and presented by the mayor: it is an act of the authority.
The next step I propose for the year ahead: the costliest item in your performance function is still the 340 departmental spreadsheets, reconciled once a year at closing. I propose reconciling them monthly for the 3 departments that keep the most — 118 of the 340 — and measuring, at the next closing, the difference in corrections between those 3 departments and the other 11. That is a protocol, not a promise: if the difference is nil, you will hear it from me. year-in-review_940-hours-item-by-item.pdfThe 940 hours in detail, the lead times held, the 11 budget orientation report indicators
⛓ Sourced · finance department logs over 12 months, finance software, HR system, sign-off log, 340 departmental spreadsheets
· 940 hours returned to the department over the year, item by item: 349 h on the 32 committee dashboards, 156 h on the 240 file cross-references, 194 h on the 1,460 figure requests, 162 h on the 6 statutory reports, 78 h on the 12 monthly budget positions. At 151.67 hours a month, that is more than six months of work returned to nine people; at 35 hours a week, more than twenty-six weeks. No post cut, no post created: it is time returned to analysis, budget preparation and management control, which until now existed only as an intention.
· Lead time on a figure request: 6 days brought down to the minute for 1,118 of the 1,460 requests, with the other 342 going to the management controller with the file already assembled.
· 146 scope alerts over the year, 138 confirmed — 94.5%, where the reconciliation used to happen only at closing. And the 9 two-valued indicators are reconciled: across the year's 32 dashboards and 6 reports, none went out in two versions.
· 0 figures released without their scope, 0 dashboards published without your finance director's named sign-off, 0 automated decisions.
What I bring on top, and what waits for you every year: your budget orientation report — the one under article L. 2312-1 of the French general local authorities code — draws in part on data I already keep up to date. In a municipality of more than 10,000 inhabitants, that report covers the expected trend in the structure of headcount and staff costs, in the content set by article D. 2312-3 of the same code. The 11 corresponding indicators are extracted, dated, with the query that produces them, so that an elected member asking where a figure comes from gets the answer in the sitting. The report is still written and presented by the mayor: it is an act of the authority.
The next step I propose for the year ahead: the costliest item in your performance function is still the 340 departmental spreadsheets, reconciled once a year at closing. I propose reconciling them monthly for the 3 departments that keep the most — 118 of the 340 — and measuring, at the next closing, the difference in corrections between those 3 departments and the other 11. That is a protocol, not a promise: if the difference is nil, you will hear it from me. year-in-review_940-hours-item-by-item.pdfThe 940 hours in detail, the lead times held, the 11 budget orientation report indicators
⛓ Sourced · finance department logs over 12 months, finance software, HR system, sign-off log, 340 departmental spreadsheets
In concrete terms, what does the agent do without asking us? And how far do we open it to elected members and departments? That is the question the council will put.
Three actions, not one more, and all three come undone.
· I acknowledge every figure request within the minute, with the scope I intend to use and the data cut-off date. And the reverse holds too: if the requester replies that this is not their scope, the request goes back out with theirs, and the exchange stays in the log — it is that log that will tell you, at year end, which scopes are missing from your indicator catalogue. Over the year, 74 requests were reworked this way, and 6 indicator sheets were born of them.
· I put into checking any figure that trips one of the scope rules in service. And the reverse holds too: an officer sets it aside with a word and the figure goes back out for release with the reason and the date of the set-aside — which gives you, the following quarter, the list of the most frequent reasons, and therefore the rules to tighten.
· I refresh the 11 indicators every night, on data cut off at the previous evening, and I display the cut-off date at the top of every table. And the reverse holds too: a dashboard freezes with a word on the date of your choosing — the date the summons goes out — and it thaws with the same word. A freeze is not a closed door.
Everything else waits for a named decision: no figure is released outside the finance department, no dashboard is opened to a department or an elected member, no report is transmitted, no data is exported, no indicator is created or altered.
Now the opening up, which is the council's real subject: it is done by role, and it is the chief executive's office that opens it, never me.
· An elected member of the finance committee sees the aggregates of the main budget and the 3 ancillary budgets, execution by department and the 11 indicators — no named line, no individual file.
· An operational department sees its own scope in detail, and the authority's aggregates to place itself.
· The chief executive's office and the finance department see all aggregates and the read log.
· Every consultation is logged, and every opening carries a review date: access opened for one committee closes when that committee has passed, unless decided otherwise. An access right without a review date is an access right that gets forgotten. automatic-actions_three-acts-and-the-role-matrix.pdfThe 3 actions, how each comes undone, who sees what and until when
✎ Framework · configuration of automatic actions, role matrix, log of scope reworks and dashboard freezes
· I acknowledge every figure request within the minute, with the scope I intend to use and the data cut-off date. And the reverse holds too: if the requester replies that this is not their scope, the request goes back out with theirs, and the exchange stays in the log — it is that log that will tell you, at year end, which scopes are missing from your indicator catalogue. Over the year, 74 requests were reworked this way, and 6 indicator sheets were born of them.
· I put into checking any figure that trips one of the scope rules in service. And the reverse holds too: an officer sets it aside with a word and the figure goes back out for release with the reason and the date of the set-aside — which gives you, the following quarter, the list of the most frequent reasons, and therefore the rules to tighten.
· I refresh the 11 indicators every night, on data cut off at the previous evening, and I display the cut-off date at the top of every table. And the reverse holds too: a dashboard freezes with a word on the date of your choosing — the date the summons goes out — and it thaws with the same word. A freeze is not a closed door.
Everything else waits for a named decision: no figure is released outside the finance department, no dashboard is opened to a department or an elected member, no report is transmitted, no data is exported, no indicator is created or altered.
Now the opening up, which is the council's real subject: it is done by role, and it is the chief executive's office that opens it, never me.
· An elected member of the finance committee sees the aggregates of the main budget and the 3 ancillary budgets, execution by department and the 11 indicators — no named line, no individual file.
· An operational department sees its own scope in detail, and the authority's aggregates to place itself.
· The chief executive's office and the finance department see all aggregates and the read log.
· Every consultation is logged, and every opening carries a review date: access opened for one committee closes when that committee has passed, unless decided otherwise. An access right without a review date is an access right that gets forgotten. automatic-actions_three-acts-and-the-role-matrix.pdfThe 3 actions, how each comes undone, who sees what and until when
✎ Framework · configuration of automatic actions, role matrix, log of scope reworks and dashboard freezes
And the day an elected member disputes a figure mid-sitting, saying it was the machine that produced it?
That day, you produce four records and the dispute moves onto facts.
One — the figure itself: its declared scope, its cut-off date, the plain-language query that produces it, and the lines that make it up — payment orders, revenue orders, invoices — with their numbers. No figure goes out without its scope declared at the top: that is the rule we have held since the first dashboard, and it is what makes the figure arguable in the right place — on what it measures, not on where it came from.
Two — who settled the figure: the name of the public official who approved the dashboard, the date and the time. Over the year: 32 dashboards, 32 named sign-offs; 6 reports, 6 sign-offs; 0 releases without sign-off. No figure has been published by the machine: I prepare, a public official settles.
Three — what each person knew, and when: I announced myself as an artificial intelligence in the first sentence of every exchange, in accordance with article 50(1) of the EU AI Act, applicable since 2 August 2026, and anyone could ask for a public official at any moment — 96 did over the year, and all 96 got one.
Four — reproducibility: the same question asked again in the sitting returns the same figure, because the query and the cut-off date are frozen with the dashboard. Across the year's 32 dashboards, 32 can be rerun identically — and that is what separates a figure from an impression.
And if a member of staff wants to know what the system holds about them: the access file is produced in one minute — the aggregated data concerning them, the reads made of their file and their stated reason. That is the right of access under article 15 GDPR, and the usual difficulty is assembling the records: they are already assembled.
The next step I propose: a thirty-minute annual review with your data protection officer, where I bring you the access rights that served no purpose over the year and the retention periods that have run out. You decide what to close and what to erase; I bring you the list rather than waiting to be asked. sovereign-framework_hosting-access-traceability.pdfWhat you produce on a disputed figure: the scope, the approver, the disclosure, the rerunnable query
✎ Framework · sign-off log, AI disclosure log, record of processing activities, indicator sheets frozen with each dashboard
One — the figure itself: its declared scope, its cut-off date, the plain-language query that produces it, and the lines that make it up — payment orders, revenue orders, invoices — with their numbers. No figure goes out without its scope declared at the top: that is the rule we have held since the first dashboard, and it is what makes the figure arguable in the right place — on what it measures, not on where it came from.
Two — who settled the figure: the name of the public official who approved the dashboard, the date and the time. Over the year: 32 dashboards, 32 named sign-offs; 6 reports, 6 sign-offs; 0 releases without sign-off. No figure has been published by the machine: I prepare, a public official settles.
Three — what each person knew, and when: I announced myself as an artificial intelligence in the first sentence of every exchange, in accordance with article 50(1) of the EU AI Act, applicable since 2 August 2026, and anyone could ask for a public official at any moment — 96 did over the year, and all 96 got one.
Four — reproducibility: the same question asked again in the sitting returns the same figure, because the query and the cut-off date are frozen with the dashboard. Across the year's 32 dashboards, 32 can be rerun identically — and that is what separates a figure from an impression.
And if a member of staff wants to know what the system holds about them: the access file is produced in one minute — the aggregated data concerning them, the reads made of their file and their stated reason. That is the right of access under article 15 GDPR, and the usual difficulty is assembling the records: they are already assembled.
The next step I propose: a thirty-minute annual review with your data protection officer, where I bring you the access rights that served no purpose over the year and the retention periods that have run out. You decide what to close and what to erase; I bring you the list rather than waiting to be asked. sovereign-framework_hosting-access-traceability.pdfWhat you produce on a disputed figure: the scope, the approver, the disclosure, the rerunnable query
✎ Framework · sign-off log, AI disclosure log, record of processing activities, indicator sheets frozen with each dashboard
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
The uses of decision-support AI in the public service
Each use corresponds to an agent we deploy. All of them work in support, subject to approval by the public officer.
Included in your agent
The 5 capabilities essential to this promise are included, at no extra cost.
From 730 € incl. VAT / month
Steering dashboards
Indicators, interim positions and charts ready to comment on for committees and directorates.
Querying your data in plain language
Ask the question in ordinary words, get the figure, the table and the chart — with the source, and no technical query.
Cross-checking internal sources
A single point of entry over files and applications that do not talk to each other: finances, HR, departmental activity.
Budget & financial steering
Tracking execution, variances against budget, analysis of operating and investment expenditure.
HR data & payroll
Tracking headcount, absence and payroll, aggregated and anonymised for steering.
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.
Sourced summaries & quantified monitoring
Briefing notes grounded in your data and your documentary sources, ready for a council decision.
Automated regulatory and legal watch from 825 € incl. VAT / month Information & monitoring officer →Automated reporting agent
Dashboards, scheduled summaries.
Automated reporting agent from 499 € incl. VAT / month Discover the agent →Wealth management agent (reporting)
Financial data.
Wealth management agent (reporting) from 653 € incl. VAT / month Discover the agent →Forecasting and planning
Footfall, demand for benefits, resource allocation.
Forecasting and planning from 770 € 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 much time can an administration win back?
By automating the consolidation of files and the formatting of indicators, a department can aim for a sharp reduction in the time spent preparing figures — reinvested in analysis and decision-making. Illustrative reference points, to be validated by a pilot.
Preparing a dashboard for a committee
Cross-checking several source files
Answering an elected member's question on figures
Qualitative, non-contractual comparison: the proportions shown illustrate the shift of the work towards review, they represent no measurement. Every output of the agent is reviewed and approved by a competent person.
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
Three options, one agent
A decision-support agent (querying data in plain language, dashboards, summaries), installed and operated for you. Choose according to how you work. Prices exclude VAT — annual subscription, the time it takes for the gains to settle in.
Agility
Setup + controlled subscription
11,215 € incl. VAT setup
then 730 € 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
1,355 € 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
15,706 € incl. VAT setup
then 970 € incl. VAT/month · + hardware from 1,269 € (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 an administration
Public data never leaves the perimeterLocal inference or an isolated resource hosted in France; no data entrusted to a foreign third party.
Data in France, under French lawFor public data: hosting under French law, native minimisation, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The public officer keeps the decisionThe agent produces verifiable figures, tables and summaries, with the source; no decision is automated.
Human oversight & traceabilitySources and calculations behind the dashboards recorded, updates and logging: compliant with the requirements of the AI Act.
Going further
Related resources
Frequently asked questions
Your questions, our answers
Do you need to code or master a spreadsheet?
No. You ask the question in ordinary words, the agent produces the figure and the visual, with the source. No technical query and no spreadsheet handling is required: the public officer concentrates on interpretation and decision.
Where does our data go?
It stays in France, on an isolated resource, with the deployment objective of processing and access operated within the European Union — or on local inference on a machine belonging to the administration. Your public data is covered by an architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
Can it be opened to elected members and directorates?
Yes, with role-based access (RBAC) suited to each profile. Every user only sees the data and indicators within their own remit, with logging of what is consulted.
Does the agent connect to our existing applications?
Yes. It sits on top of your tools (financial software, HR system, document management, business files) without imposing a migration. It cross-checks those internal sources through a single point of entry, and we adapt the integration to your environment.
Are the figures produced reliable enough for a council decision?
The agent produces sourced and traceable figures, but they must always be checked by a public officer before any decision or publication. The AI agent assists, the public officer decides: that is a requirement of human oversight and of equal treatment.
How long does it take to deploy the agent?
A few weeks as a rule, after a free audit that identifies the most useful decision-support use case, then a phase of design, connection to the sources, testing and training of officers before going live.
Related roles
Other roles in public steering and data
Let's talk
Let us estimate the decision-support potential in your administration
15 minutes to identify the most useful use case — hosted in France, supervised, with no commitment.