● Public sector — Management control & financial analysis
The AI management control agent: spot the variances, inform the steering
Consolidating budget data, finding the variance that matters, preparing the dashboard for the elected member or the director: this work takes a considerable amount of time, often at the expense of analysis. Your AI agent queries your financial data, explains the variances and prepares clear summaries. Hosted in France — on local inference or an isolated resource — public data stays under control. The AI agent assists, the public officer decides.
Hosted in France
Public data protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Management control agent · hosted in France
On the 2026 operating budget, where are the main variances against forecast as at 31 May?
Three notable variances as at 31 May. Chapter 011 (general expenses): +14% against forecast, driven by energy and maintenance. Chapter 012 (staff costs): on track, +1%. Revenue line 73 (taxes and levies): −6% against the expected collection pace.
I can break chapter 011 down line by line.
⛓ Source · budget data and payment orders from your financial system
I can break chapter 011 down line by line.
⛓ Source · budget data and payment orders from your financial system
Break down 011 and prepare a dashboard for Thursday's steering committee.
Chapter 011 broken down. 70% of the variance comes from the energy line (+38%) and 20% from building maintenance. The other lines are stable.
I have prepared a dashboard: execution by chapter, focus on energy, year-end projection — figures and charts to review before circulation.
✎ Action · dashboard ready for review — the public officer approves
I have prepared a dashboard: execution by chapter, focus on energy, year-end projection — figures and charts to review before circulation.
✎ Action · dashboard 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 or a public body, a Blue Lemon Agent management control agent queries your financial data in plain language, detects budget and accounting variances and prepares steering dashboards. It runs on local inference or is hosted in France: 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 on consolidation is redirected towards analysis and decision support. The AI agent assists; the public officer keeps the lead on every figure circulated.
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 public management control needs AI — and why sovereignty is not negotiable
Budget constraints are tightening and both elected members and directorates expect finer, more frequent, more readable steering. But available time is absorbed by consolidation, reconciliation and formatting — and an administration's financial data is among the most sensitive there is.
! The issue
The public management controller is caught between directorates demanding up-to-date positions and variance analyses, and manual consolidation work that keeps growing (extracts, reconciliations, formatting). Yet most consumer AI tools amount to entrusting budget execution, payroll, tax revenue and accounting data to a third party, often hosted outside Europe and subject to the Cloud Act.
✓ Our answer
For an administration, AI is only of interest 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 consolidation is never paid for in lost confidentiality. The aim is not to replace the management controller, but to give them back time for analysis and advice to decision-makers.
The decisive point
Confidentiality of public financial data: sovereignty & compliance
A local authority handles sensitive budget and accounting data, subject to strict rules. Here is how the architecture of our agents protects it, financial year by financial year.
Local inference
The agent can run on a machine belonging to the authority: no budget 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
Financial data: architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
One isolated resource per entity
No pooling of financial data: an environment strictly dedicated to your administration.
Encryption & controlled access
Encryption in transit and at rest, role-based access (RBAC), strong authentication and logging.
AI Act: governed deployment
An agent strictly in support; no figure circulated 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.
- 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
5 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 Marnecourt — municipality of 38,000 residents
- Sector
- Fictional authority — a municipality of 38,000 residents: schools, early years, sport and culture, roads, social action, with a main budget of €62m of which €48m operating and €14m capital
- Headcount
- 820 staff, a finance department of 11 people including 2 in management control — a head and an analyst, serving 9 directorates and 62 services
- Public served
- 38,000 residents, and directorates that need an execution position in order to decide — every euro tracked is a euro that funds a service to the public
- Order of magnitude
- 14 budget chapters monitored, 21,400 payment orders and 3,900 revenue orders a year, 148 live contracts, €12.4m of commitments and €4.2m of carry-overs, 640 suppliers
- Tools in place
- Financial information system, accounting software, document management system and five years of budget execution archived — the agent plugs into them read-only, nothing is replaced and nothing is migrated
- Who decides
- The finance director approves every figure before it circulates; the chief executive rules on management measures; the public accountant alone judges payment; the elected assembly votes the budget and its amendments
- Room for improvement
- The monthly dashboard takes 14 hours to prepare and arrives three weeks after month end; a directorate's question about its own data waits 4 days for an answer; 6 contracts overran their envelope without being flagged before closing; and the annual accounts take 9 weeks to settle
At Marnecourt, two officers run management control for 62 services. Most of their time goes into consolidation, reconciliation and formatting, at the expense of the analysis that directorates and elected members need in order to decide. The agent runs on local inference on a machine of the town, reads the financial information system and five years of execution without ever writing to them, and circulates no figure without the finance director's approval. The exchanges that follow span one year, from the execution position at 31 May to the review presented to the management board.
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.
Management control agent · hosted in France
I read your five years of budget execution last night — 5 financial years, 14 chapters, 98,000 payment orders — and I hand you the position at 31 May with three notable variances, each explained by the lines that carry it.
The three variances, and I give you straight away what causes them:
· Chapter 011, general operating costs: +14 % against forecast, that is €610,000 above the expected pace at 31 May. The variance comes 70 % from energy, up 38 %, and 20 % from building maintenance — €427,000 and €122,000. The other lines are stable.
· Chapter 012, staff costs: +1 %. On track — and I say so because a chapter on track is information, not silence: it is your first spending item, and it is holding.
· Revenue chapter 73, taxes and duties: −6 % against the expected collection pace, that is €780,000. This is not lower revenue: it is a payment expected in April and received in June — €640,000 of the €780,000 were already in by 20 June, and I give you the date of each receipt.
What producing this page costs today, on the three items you measure: your monthly dashboard takes 14 hours to prepare — that is the reference for my three figures.
· Preparation — consolidation, reconciliation, formatting — weighs 60 % of that time, that is 8 hours 24.
· Analysing a chapter's variance: 20 %, that is 2 hours 48.
· Answering questions asked of the data: 5 %, that is 42 minutes.
What I propose, and it is not a promise — the position at 31 May is already written: preparation falls to 12 %, the analysis of a variance to 8 % — 1 hour 41 and 1 hour 07. The dashboard stops arriving three weeks after month end: it is ready on the 3rd.
The next step, and it takes an hour: you read the detail of chapter 011, line by line, and I hand you the year-end projection with three costed measures. The management decision belongs to the chief executive — and I hand it to him within the hour, costed and sourced on your own payment orders. position-at-31-may_three-variances-explained.pdf011 at +14 %, 012 at +1 %, 73 at −6 %, each variance with its lines
⛓ Sourced · financial information system, 5 years of execution, 21,400 payment orders of the year, receipts log
The three variances, and I give you straight away what causes them:
· Chapter 011, general operating costs: +14 % against forecast, that is €610,000 above the expected pace at 31 May. The variance comes 70 % from energy, up 38 %, and 20 % from building maintenance — €427,000 and €122,000. The other lines are stable.
· Chapter 012, staff costs: +1 %. On track — and I say so because a chapter on track is information, not silence: it is your first spending item, and it is holding.
· Revenue chapter 73, taxes and duties: −6 % against the expected collection pace, that is €780,000. This is not lower revenue: it is a payment expected in April and received in June — €640,000 of the €780,000 were already in by 20 June, and I give you the date of each receipt.
What producing this page costs today, on the three items you measure: your monthly dashboard takes 14 hours to prepare — that is the reference for my three figures.
· Preparation — consolidation, reconciliation, formatting — weighs 60 % of that time, that is 8 hours 24.
· Analysing a chapter's variance: 20 %, that is 2 hours 48.
· Answering questions asked of the data: 5 %, that is 42 minutes.
What I propose, and it is not a promise — the position at 31 May is already written: preparation falls to 12 %, the analysis of a variance to 8 % — 1 hour 41 and 1 hour 07. The dashboard stops arriving three weeks after month end: it is ready on the 3rd.
The next step, and it takes an hour: you read the detail of chapter 011, line by line, and I hand you the year-end projection with three costed measures. The management decision belongs to the chief executive — and I hand it to him within the hour, costed and sourced on your own payment orders. position-at-31-may_three-variances-explained.pdf011 at +14 %, 012 at +1 %, 73 at −6 %, each variance with its lines
⛓ Sourced · financial information system, 5 years of execution, 21,400 payment orders of the year, receipts log
Before we go further: is our data reliable? We spend endless time reconciling extracts that do not say the same thing.
It is, to 99.2 % after reconciliation, and I tell you exactly where the 0.8 % are — 174 payment orders out of 21,400.
What I did with your five years of execution:
· I reconciled the financial information system, the accounting software and your management statements, line by line. For the current year, 21,226 payment orders match perfectly.
· The remaining 174 fall into three families, and none is a keying error: 96 are accrued charges — an accrual is a cost committed during the year but invoiced after it closes; it belongs to the year it concerns, not to the one in which the invoice arrives —, 52 are postings corrected during the year, and 26 are adjustments relating to a previous year.
· For each of the 174 I give you the payment order, the date, the original posting and the posting retained. This is not a list of anomalies: it is the map of the points where two different extracts give two different figures, and it explains most of your manual reconciliations.
Financial memory now, because that is where you lose the most time without counting it: I have indexed your five years of budget resolutions, your budget amendments, your internal procedures and your activity reports. A question such as “what did the council vote on school meal charges, and when” is answered in 40 seconds, with the resolution attached, against 25 minutes of searching. Over the quarter, 148 such searches, that is 58 hours no longer spent searching.
And one thing I flag without being asked: 3 lines of your original budget carry an amount that a later amendment replaced, never carried into the working document your directorates consult. A budget amendment is a resolution that adjusts the voted budget during the year: it is the amount it sets that prevails. I found the three resolutions and set the old amount and the voted amount side by side — the cumulative gap is €216,000, and it alone explained two of the reconciliations your analyst redid every month.
The next step I propose: that the reconciliation of the three sources run every night and that the map of gaps reach you in the morning. Your two officers will stop reconciling and start explaining — exactly the shift your senior management is asking of you. reliability-and-financial-memory_174-gaps-explained.pdf99.2 % match, 3 lines not carried over, €216,000 explained
⛓ Sourced · financial information system, accounting software, management statements, 5 years of budget resolutions and amendments
What I did with your five years of execution:
· I reconciled the financial information system, the accounting software and your management statements, line by line. For the current year, 21,226 payment orders match perfectly.
· The remaining 174 fall into three families, and none is a keying error: 96 are accrued charges — an accrual is a cost committed during the year but invoiced after it closes; it belongs to the year it concerns, not to the one in which the invoice arrives —, 52 are postings corrected during the year, and 26 are adjustments relating to a previous year.
· For each of the 174 I give you the payment order, the date, the original posting and the posting retained. This is not a list of anomalies: it is the map of the points where two different extracts give two different figures, and it explains most of your manual reconciliations.
Financial memory now, because that is where you lose the most time without counting it: I have indexed your five years of budget resolutions, your budget amendments, your internal procedures and your activity reports. A question such as “what did the council vote on school meal charges, and when” is answered in 40 seconds, with the resolution attached, against 25 minutes of searching. Over the quarter, 148 such searches, that is 58 hours no longer spent searching.
And one thing I flag without being asked: 3 lines of your original budget carry an amount that a later amendment replaced, never carried into the working document your directorates consult. A budget amendment is a resolution that adjusts the voted budget during the year: it is the amount it sets that prevails. I found the three resolutions and set the old amount and the voted amount side by side — the cumulative gap is €216,000, and it alone explained two of the reconciliations your analyst redid every month.
The next step I propose: that the reconciliation of the three sources run every night and that the map of gaps reach you in the morning. Your two officers will stop reconciling and start explaining — exactly the shift your senior management is asking of you. reliability-and-financial-memory_174-gaps-explained.pdf99.2 % match, 3 lines not carried over, €216,000 explained
⛓ Sourced · financial information system, accounting software, management statements, 5 years of budget resolutions and amendments
All this assumes you read our budget execution, our payroll and our supporting documents. Where does that data go?
It goes nowhere. I run on local inference on a machine of the town, and no financial data leaves its network.
Local inference means the model computes on your machine: the amount of a payment order, a supplier's name or a payroll line never cross an outside network to be processed. If the department would rather not host a machine, the other route is an isolated resource hosted in France, dedicated to Marnecourt — no pooling with another authority.
What that changes, point by point:
· Your financial data trains no model, neither ours nor a third party's.
· I work read-only on the financial information system and the accounting software, and the technical account through which I read has no write permission — no accounting entry can come from me, and that is checked with one command.
· Encryption in transit and at rest, strong authentication, and role-based access — rights follow the job: an analyst opens their directorate's execution, not the named payroll statements. 6 roles for your 11 finance staff, and the log shows 0 out-of-role access since go-live.
· Hosting in France, under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
· Logging: which data was read, when, to produce which figure. It is that log that lets a calculation be redone two years later before the regional audit chamber.
One rule I hold in every case, and I give you straight away what renders the same service: no personal data appears in a management dashboard. Neither a benefit recipient, nor a member of staff, nor an individual situation: the purpose of management reporting does not justify it, and a dashboard always travels further than one imagines. What I produce instead, answering the same question: the aggregate by scheme, by band and by month, already calculated — on the discretionary family allowance: 1,240 recipients, 4 bands, average amount and three-year trend, without a single name appearing.
The figure that sums all this up: 0 financial data left the town's network across the 21,400 payment orders of the year, and processing in the EU targeted.
What I propose: that I maintain the record your data protection officer and your senior management will ask for — hosting, data processed, retention periods, who accesses what. It is requested once a year and takes two days to rebuild; the first version is already written and attached. technical-framework_where-financial-data-lives.pdfLocal inference, read-only, processing in the EU targeted
✎ Framework · deployment architecture, technical account permissions, matrix of the 6 roles, first version of the register record
Local inference means the model computes on your machine: the amount of a payment order, a supplier's name or a payroll line never cross an outside network to be processed. If the department would rather not host a machine, the other route is an isolated resource hosted in France, dedicated to Marnecourt — no pooling with another authority.
What that changes, point by point:
· Your financial data trains no model, neither ours nor a third party's.
· I work read-only on the financial information system and the accounting software, and the technical account through which I read has no write permission — no accounting entry can come from me, and that is checked with one command.
· Encryption in transit and at rest, strong authentication, and role-based access — rights follow the job: an analyst opens their directorate's execution, not the named payroll statements. 6 roles for your 11 finance staff, and the log shows 0 out-of-role access since go-live.
· Hosting in France, under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
· Logging: which data was read, when, to produce which figure. It is that log that lets a calculation be redone two years later before the regional audit chamber.
One rule I hold in every case, and I give you straight away what renders the same service: no personal data appears in a management dashboard. Neither a benefit recipient, nor a member of staff, nor an individual situation: the purpose of management reporting does not justify it, and a dashboard always travels further than one imagines. What I produce instead, answering the same question: the aggregate by scheme, by band and by month, already calculated — on the discretionary family allowance: 1,240 recipients, 4 bands, average amount and three-year trend, without a single name appearing.
The figure that sums all this up: 0 financial data left the town's network across the 21,400 payment orders of the year, and processing in the EU targeted.
What I propose: that I maintain the record your data protection officer and your senior management will ask for — hosting, data processed, retention periods, who accesses what. It is requested once a year and takes two days to rebuild; the first version is already written and attached. technical-framework_where-financial-data-lives.pdfLocal inference, read-only, processing in the EU targeted
✎ Framework · deployment architecture, technical account permissions, matrix of the 6 roles, first version of the register record
Give me the detail of chapter 011. The chief executive will ask where this +14 % comes from and what can be done before year end.
The detail is ready, line by line, and so is the year-end projection — with three costed measures the chief executive can weigh in amounts.
Where the €610,000 of variance at 31 May come from:
· Energy: +38 %, that is €427,000 — 70 % of the variance. And I break it down further: 61 % of that rise is the unit price of your two contracts renewed in January, 39 % is volume consumed. These are two different causes and they do not call for the same decision — price is renegotiated at renewal, volume is corrected straight away.
· Building maintenance: €122,000 — 20 % of the variance. €84,000 come from three unplanned works on the sports hall roof, paid in March.
· The rest of the chapter's lines are stable, within 2 % of forecast.
The year-end projection, if the pace holds: +€1.32m on chapter 011. The calculation is yours: execution at 31 May applied to the consumption profile of the last five years, and I give you the range — between €1.18m and €1.44m depending on how hard the winter is, measured on your own December consumption.
The three measures I put on the table, each costed on your data:
· Adjusting heating schedules in the 14 buildings whose records show partial occupancy — estimated effect €214,000 over the second half, calculated on the gap between heating hours and recorded opening hours.
· Deferring 4 non-urgent maintenance jobs to the following year, each named, with the buildings department's technical opinion alongside — €310,000.
· Renegotiating the contract that expires in September, with the prices found in the published award notices of four comparable authorities — a range of €186,000 over a full year.
What that gives: the projection goes from +€1.32m to +€610,000, and I tell you what remains: the €610,000 correspond to the price of energy, which no management measure corrects — it calls for a budget amendment, and that is exactly the kind of amount better entered in June than in December.
None of the three measures is my opinion: they are three calculations made on your records and on published notices, and the chief executive can ask me for a fourth, costed before the meeting ends.
The next step I propose: that the year-end projection be recalculated monthly on all 14 chapters, not only on 011. Run over the last five financial years, the gap between the June projection and the actual outturn would have been under 3 % nine times out of ten. chapter-011_variance-broken-down-and-three-measures.pdf€427,000 energy, €122,000 maintenance, €1.32m cut to €610,000
⛓ Sourced · payment orders of chapter 011, energy contracts and consumption records, buildings department technical opinions, published award notices
Where the €610,000 of variance at 31 May come from:
· Energy: +38 %, that is €427,000 — 70 % of the variance. And I break it down further: 61 % of that rise is the unit price of your two contracts renewed in January, 39 % is volume consumed. These are two different causes and they do not call for the same decision — price is renegotiated at renewal, volume is corrected straight away.
· Building maintenance: €122,000 — 20 % of the variance. €84,000 come from three unplanned works on the sports hall roof, paid in March.
· The rest of the chapter's lines are stable, within 2 % of forecast.
The year-end projection, if the pace holds: +€1.32m on chapter 011. The calculation is yours: execution at 31 May applied to the consumption profile of the last five years, and I give you the range — between €1.18m and €1.44m depending on how hard the winter is, measured on your own December consumption.
The three measures I put on the table, each costed on your data:
· Adjusting heating schedules in the 14 buildings whose records show partial occupancy — estimated effect €214,000 over the second half, calculated on the gap between heating hours and recorded opening hours.
· Deferring 4 non-urgent maintenance jobs to the following year, each named, with the buildings department's technical opinion alongside — €310,000.
· Renegotiating the contract that expires in September, with the prices found in the published award notices of four comparable authorities — a range of €186,000 over a full year.
What that gives: the projection goes from +€1.32m to +€610,000, and I tell you what remains: the €610,000 correspond to the price of energy, which no management measure corrects — it calls for a budget amendment, and that is exactly the kind of amount better entered in June than in December.
None of the three measures is my opinion: they are three calculations made on your records and on published notices, and the chief executive can ask me for a fourth, costed before the meeting ends.
The next step I propose: that the year-end projection be recalculated monthly on all 14 chapters, not only on 011. Run over the last five financial years, the gap between the June projection and the actual outturn would have been under 3 % nine times out of ten. chapter-011_variance-broken-down-and-three-measures.pdf€427,000 energy, €122,000 maintenance, €1.32m cut to €610,000
⛓ Sourced · payment orders of chapter 011, energy contracts and consumption records, buildings department technical opinions, published award notices
How do you know those €84,000 really come from the sports hall roof? Our payment order descriptions are often unreadable.
I do not rely on the description: I read the supporting document attached to the payment order, invoice by invoice.
The supporting document is what grounds the spending — invoice, purchase order, works statement, acceptance certificate; it is what the public accountant requires before paying, and it is what says what was actually bought.
What I read on each document: the supplier, the date, the amount net and gross of tax, the detail of the invoice lines, the contract or purchase order number, and the site or facility concerned when it appears. Of the 21,400 payment orders of the year, 20,940 carry a readable and usable document — 97.8 %.
What that makes possible, and the description did not:
· Tying a cost to a facility: the €84,000 come from three invoices from the same firm, each bearing the sports hall's address and the detail of the roofing works. The payment order description said “building works”.
· Rebuilding a facility's full cost: for the sports hall, €214,000 over the year — utilities, maintenance, upkeep, staff assigned — against €96,000 shown on the maintenance account alone. That is the figure an elected member asks for, and nobody could produce it in under a week.
· Spotting accounting anomalies before closing: over the quarter, 41 possible duplicates flagged, 12 confirmed after examination — €9,400, all invoices received twice through two different channels. The 12 were withdrawn before payment, and the public accountant never had to reject them.
· Flagging missing documents: 174 payment orders awaiting a document, each with the holding service and the request already drafted. Over the quarter, 158 documents came back within eight days.
The rule that makes the figure safe: every amount I write has been read on a document. An unreadable invoice leaves with its request: I flag it, I say where I looked and who holds it, and I hand you the letter ready to go. 460 payment orders out of 21,400 are in that case, and they are listed — that is what makes the remaining 97.8 % a figure you can lean on.
The next step I propose: the full cost of the town's 12 heaviest facilities, sports hall included, recalculated each quarter. The first is written; it took 4 minutes where your analyst spent a week, and it carries the source of every line. reading-the-documents_21400-orders-and-full-cost.pdf97.8 % usable documents, 12 duplicates withdrawn before payment
⛓ Sourced · supporting documents attached to the 21,400 payment orders, contracts and purchase orders, facilities inventory
The supporting document is what grounds the spending — invoice, purchase order, works statement, acceptance certificate; it is what the public accountant requires before paying, and it is what says what was actually bought.
What I read on each document: the supplier, the date, the amount net and gross of tax, the detail of the invoice lines, the contract or purchase order number, and the site or facility concerned when it appears. Of the 21,400 payment orders of the year, 20,940 carry a readable and usable document — 97.8 %.
What that makes possible, and the description did not:
· Tying a cost to a facility: the €84,000 come from three invoices from the same firm, each bearing the sports hall's address and the detail of the roofing works. The payment order description said “building works”.
· Rebuilding a facility's full cost: for the sports hall, €214,000 over the year — utilities, maintenance, upkeep, staff assigned — against €96,000 shown on the maintenance account alone. That is the figure an elected member asks for, and nobody could produce it in under a week.
· Spotting accounting anomalies before closing: over the quarter, 41 possible duplicates flagged, 12 confirmed after examination — €9,400, all invoices received twice through two different channels. The 12 were withdrawn before payment, and the public accountant never had to reject them.
· Flagging missing documents: 174 payment orders awaiting a document, each with the holding service and the request already drafted. Over the quarter, 158 documents came back within eight days.
The rule that makes the figure safe: every amount I write has been read on a document. An unreadable invoice leaves with its request: I flag it, I say where I looked and who holds it, and I hand you the letter ready to go. 460 payment orders out of 21,400 are in that case, and they are listed — that is what makes the remaining 97.8 % a figure you can lean on.
The next step I propose: the full cost of the town's 12 heaviest facilities, sports hall included, recalculated each quarter. The first is written; it took 4 minutes where your analyst spent a week, and it carries the source of every line. reading-the-documents_21400-orders-and-full-cost.pdf97.8 % usable documents, 12 duplicates withdrawn before payment
⛓ Sourced · supporting documents attached to the 21,400 payment orders, contracts and purchase orders, facilities inventory
And when you flag a variance that is not one? We cannot mobilise a directorate for nothing.
It has already happened, I counted it, and I publish it to you: of the 214 variances I flagged in the first quarter, 37 corresponded to no real anomaly — 17.3 %. That is the figure that does not flatter me, and it is better that you have it from me than from an irritated director.
The real cause, measured and not supposed: 29 of the 37 came from the seasonality of invoicing. An energy invoice covering two months and paid in one go shows a +100 % variance in the month it lands, and −100 % the month after — that is not a drift, it is a calendar. The other 8 came from accrued charges of the previous year, posted to the current one.
What I did with it, and it is measured: I rebuilt the actual invoicing calendar of your 640 suppliers over five years — monthly, two-monthly, quarterly, at contract milestones — and I now compare each line to its own rhythm, not to one twelfth of the forecast. I also isolated accruals so that they stop weighing on the month in which they land.
The following quarter: 9 unconfirmed flags out of 198 — 4.5 %. And the 9 concern new suppliers whose invoicing history I did not yet have.
The rule that holds all the rest: a variance never travels alone. It travels with the lines that carry it, the payment orders that justify them, the supplier's invoicing rhythm and the comparison with the same month of the three previous years. A director receiving a flag sees within ten seconds whether it needs their attention — and over the quarter, the average time to handle a flag went from 11 days to 2.
And the protection that matters for your department: no figure circulates without approval. Across 412 flags over two quarters, 412 went through the finance director before reaching a directorate — and the 46 unconfirmed flags were all set aside before circulation, none after.
What I propose now: that the 9 new suppliers enter the invoicing calendar from their third payment order, instead of waiting a full year. Of the 9 flags of the second quarter, 7 would have been avoided — it is the same correction as the seasonality one, applied one notch earlier. unconfirmed-flags_17-3-then-4-5-percent.pdf17.3 % → 4.5 %, measured cause, 412 human approvals
⛓ Sourced · log of flags and their examination over two quarters, invoicing calendar of the 640 suppliers, accrued charges
The real cause, measured and not supposed: 29 of the 37 came from the seasonality of invoicing. An energy invoice covering two months and paid in one go shows a +100 % variance in the month it lands, and −100 % the month after — that is not a drift, it is a calendar. The other 8 came from accrued charges of the previous year, posted to the current one.
What I did with it, and it is measured: I rebuilt the actual invoicing calendar of your 640 suppliers over five years — monthly, two-monthly, quarterly, at contract milestones — and I now compare each line to its own rhythm, not to one twelfth of the forecast. I also isolated accruals so that they stop weighing on the month in which they land.
The following quarter: 9 unconfirmed flags out of 198 — 4.5 %. And the 9 concern new suppliers whose invoicing history I did not yet have.
The rule that holds all the rest: a variance never travels alone. It travels with the lines that carry it, the payment orders that justify them, the supplier's invoicing rhythm and the comparison with the same month of the three previous years. A director receiving a flag sees within ten seconds whether it needs their attention — and over the quarter, the average time to handle a flag went from 11 days to 2.
And the protection that matters for your department: no figure circulates without approval. Across 412 flags over two quarters, 412 went through the finance director before reaching a directorate — and the 46 unconfirmed flags were all set aside before circulation, none after.
What I propose now: that the 9 new suppliers enter the invoicing calendar from their third payment order, instead of waiting a full year. Of the 9 flags of the second quarter, 7 would have been avoided — it is the same correction as the seasonality one, applied one notch earlier. unconfirmed-flags_17-3-then-4-5-percent.pdf17.3 % → 4.5 %, measured cause, 412 human approvals
⛓ Sourced · log of flags and their examination over two quarters, invoicing calendar of the 640 suppliers, accrued charges
The steering committee meets on Thursday. I need a dashboard: execution by chapter, energy focus, year-end projection.
It is ready, figures and charts, and it only awaits your reading — I also tell you the three points you will be questioned on.
What the dashboard carries:
· Execution of the 14 chapters, in amount and as a percentage of forecast, with the same month of the three previous years alongside — the only comparison worth making, because it neutralises seasonality.
· The energy focus: +38 %, broken down into price and volume, building by building for the 14 heaviest consumers.
· The year-end projection on the 14 chapters, with its range and the assumption behind it, written in one line under the chart.
· The three management measures and their costed effect, so the committee weighs amounts.
· And the extraction date of every figure, in the footer. A dashboard without an extraction date turns against whoever presents it as soon as one figure has moved in the meantime.
The three questions you will be asked, and the answers are in the pack: “is 012 really holding?” — yes, +1 %, with the detail by directorate; “are the €214,000 of heating realistic?” — the calculation is written from your occupancy records; “and capital spending?” — €4.2m of carry-overs, of which €2.1m on three schemes I set out.
The time this moves: preparing a dashboard goes from 60 % to 12 % of its time — from 8 hours 24 to 1 hour 41. Across 48 dashboards a year — the general monthly one and the quarterly ones of the 9 directorates —, that is 322 hours given back to analysis. And analysing a chapter's variance goes from 20 % to 8 % — from 2 hours 48 to 1 hour 07, that is 282 hours across 168 analyses a year.
The management note goes with it, and it is written: two pages — the position, the three variances, the measures and what they produce, in the form of your previous notes: I read them, they all share the same structure and I have kept it. The finance department's annual activity report starts from the same material: it is 80 % built as soon as the twelve monthly notes exist.
The next step I propose: that the dashboard be ready on the 3rd of each month rather than three weeks after closing. Over the last five financial years, 4 management decisions were taken more than six weeks after the fact they were correcting — a figure that arrives on the 3rd is decided within the month; a figure that arrives on the 21st is decided next quarter. dashboard_execution-energy-focus-projection.pdf8 h 24 down to 1 h 41, ready on the 3rd
⛓ Sourced · financial information system, consumption records per building, carry-overs, 12 previous management notes
What the dashboard carries:
· Execution of the 14 chapters, in amount and as a percentage of forecast, with the same month of the three previous years alongside — the only comparison worth making, because it neutralises seasonality.
· The energy focus: +38 %, broken down into price and volume, building by building for the 14 heaviest consumers.
· The year-end projection on the 14 chapters, with its range and the assumption behind it, written in one line under the chart.
· The three management measures and their costed effect, so the committee weighs amounts.
· And the extraction date of every figure, in the footer. A dashboard without an extraction date turns against whoever presents it as soon as one figure has moved in the meantime.
The three questions you will be asked, and the answers are in the pack: “is 012 really holding?” — yes, +1 %, with the detail by directorate; “are the €214,000 of heating realistic?” — the calculation is written from your occupancy records; “and capital spending?” — €4.2m of carry-overs, of which €2.1m on three schemes I set out.
The time this moves: preparing a dashboard goes from 60 % to 12 % of its time — from 8 hours 24 to 1 hour 41. Across 48 dashboards a year — the general monthly one and the quarterly ones of the 9 directorates —, that is 322 hours given back to analysis. And analysing a chapter's variance goes from 20 % to 8 % — from 2 hours 48 to 1 hour 07, that is 282 hours across 168 analyses a year.
The management note goes with it, and it is written: two pages — the position, the three variances, the measures and what they produce, in the form of your previous notes: I read them, they all share the same structure and I have kept it. The finance department's annual activity report starts from the same material: it is 80 % built as soon as the twelve monthly notes exist.
The next step I propose: that the dashboard be ready on the 3rd of each month rather than three weeks after closing. Over the last five financial years, 4 management decisions were taken more than six weeks after the fact they were correcting — a figure that arrives on the 3rd is decided within the month; a figure that arrives on the 21st is decided next quarter. dashboard_execution-energy-focus-projection.pdf8 h 24 down to 1 h 41, ready on the 3rd
⛓ Sourced · financial information system, consumption records per building, carry-overs, 12 previous management notes
Our directorates no longer dare ask us about their data: they know they will wait four days. So they steer blind.
They can ask me in plain language, at any hour, and they get their answer in 40 seconds with the chart and the source — the number of questions has tripled in one quarter, and it is the best news in this review.
What that gave, measured:
· 340 questions a year before, 1,060 at today's annual rate — and they are not the same questions. Before, people asked for a figure; now they ask why it moved.
· Answering time goes from 4 days to 40 seconds.
· And it is the only item in my calculation that rises: answering questions asked of the data goes from 5 % to 8 % of a dashboard's time — because there are three times as many questions, each handled in a fraction of the time. A management control function whose data is queried three times more often is one that finally serves decisions.
What a directorate gets when it asks “why did my utilities rise in the first quarter”: the amount, the variance against the same quarter of the three previous years, the price-volume breakdown, the three buildings carrying 68 % of the rise, and the matching payment orders. Every figure carries its source and extraction date — which is what lets a director lift it straight into their own note.
And what I do when the question goes beyond what I can establish: I say what I know, what is missing and who holds it. On “the unit cost of an hour of nursery care”, I give you direct costs to the cent — €214,000 over the quarter — and I tell you that the allocation key for overheads does not yet exist in your procedures. I propose two, calculated on your data, with the hourly cost each produces: €12.40 and €14.10. It is senior management that settles the key — and I hand it to them in one meeting, costed under both assumptions, rather than in six months of working group.
The next step I propose: opening access to the 9 directorates, each on its own perimeter. The partitioning follows the roles you have already defined: a directorate sees its own execution, not its neighbour's, and nobody sees personal data. Over the trial quarter with two directorates, 214 questions asked, 0 out-of-perimeter access. plain-language-questions_340-become-1060.pdf4 days down to 40 seconds, 5 % → 8 % and why it is a gain
⛓ Sourced · log of questions asked over two quarters, financial information system, cost allocation procedures
What that gave, measured:
· 340 questions a year before, 1,060 at today's annual rate — and they are not the same questions. Before, people asked for a figure; now they ask why it moved.
· Answering time goes from 4 days to 40 seconds.
· And it is the only item in my calculation that rises: answering questions asked of the data goes from 5 % to 8 % of a dashboard's time — because there are three times as many questions, each handled in a fraction of the time. A management control function whose data is queried three times more often is one that finally serves decisions.
What a directorate gets when it asks “why did my utilities rise in the first quarter”: the amount, the variance against the same quarter of the three previous years, the price-volume breakdown, the three buildings carrying 68 % of the rise, and the matching payment orders. Every figure carries its source and extraction date — which is what lets a director lift it straight into their own note.
And what I do when the question goes beyond what I can establish: I say what I know, what is missing and who holds it. On “the unit cost of an hour of nursery care”, I give you direct costs to the cent — €214,000 over the quarter — and I tell you that the allocation key for overheads does not yet exist in your procedures. I propose two, calculated on your data, with the hourly cost each produces: €12.40 and €14.10. It is senior management that settles the key — and I hand it to them in one meeting, costed under both assumptions, rather than in six months of working group.
The next step I propose: opening access to the 9 directorates, each on its own perimeter. The partitioning follows the roles you have already defined: a directorate sees its own execution, not its neighbour's, and nobody sees personal data. Over the trial quarter with two directorates, 214 questions asked, 0 out-of-perimeter access. plain-language-questions_340-become-1060.pdf4 days down to 40 seconds, 5 % → 8 % and why it is a gain
⛓ Sourced · log of questions asked over two quarters, financial information system, cost allocation procedures
Circulating figures to the directorates commits the finance department's word. I will not sign a blank cheque.
You do not have to sign one: I propose a written mandate, capped, dated and withdrawable with one word — and it is already drafted, attached.
What it brings first, because that is what decides: a directorate that gets its execution position in 40 seconds instead of 4 days corrects within the month instead of noting it at the quarter. Over the trial quarter, 3 budget overruns were avoided because the directorate saw them coming.
What the mandate says, and it fits in six lines:
· Exact scope: the six settled indicators — execution by chapter, execution by service, commitments, carry-overs, budgeted headcount, live contracts —, each with its written definition and its extraction date, and nothing else.
· Projections and analytical comments stay outside the mandate. A projection commits an assumption; it is handed to you drafted, and it is you who circulate it. Over the quarter, those notes went out within 2 hours instead of 4 days.
· No personal data in any circulation, whatever the request — and the aggregate answering the same question is supplied as a matter of course.
· Every figure circulated carries its source, its extraction date and the statement that it was prepared by a digital assistant of the finance department.
· You receive each morning the summary of what went out the day before, on one page. A wrong figure is caught in an hour, not at the next committee.
· Duration: review after three months, with the record of what it changed. Without an explicit decision at the review, the mandate stops — it is renewal that requires a signature, not termination. Withdrawal: one word, effective within the minute.
The decision belongs to the finance director — and it is taken on a text already written, with one signature. The mandate is drafted, and so is the note to the 9 directorates. You sign, and the directorates see their execution from Monday; the review is already in your diary on the 15th of the third month. circulation-mandate_six-indicators-capped.pdf6 indicators, projections excluded, review at 3 months
✎ Framework · drafted mandate, definition of the 6 indicators, note to the directorates, trial quarter log
What it brings first, because that is what decides: a directorate that gets its execution position in 40 seconds instead of 4 days corrects within the month instead of noting it at the quarter. Over the trial quarter, 3 budget overruns were avoided because the directorate saw them coming.
What the mandate says, and it fits in six lines:
· Exact scope: the six settled indicators — execution by chapter, execution by service, commitments, carry-overs, budgeted headcount, live contracts —, each with its written definition and its extraction date, and nothing else.
· Projections and analytical comments stay outside the mandate. A projection commits an assumption; it is handed to you drafted, and it is you who circulate it. Over the quarter, those notes went out within 2 hours instead of 4 days.
· No personal data in any circulation, whatever the request — and the aggregate answering the same question is supplied as a matter of course.
· Every figure circulated carries its source, its extraction date and the statement that it was prepared by a digital assistant of the finance department.
· You receive each morning the summary of what went out the day before, on one page. A wrong figure is caught in an hour, not at the next committee.
· Duration: review after three months, with the record of what it changed. Without an explicit decision at the review, the mandate stops — it is renewal that requires a signature, not termination. Withdrawal: one word, effective within the minute.
The decision belongs to the finance director — and it is taken on a text already written, with one signature. The mandate is drafted, and so is the note to the 9 directorates. You sign, and the directorates see their execution from Monday; the review is already in your diary on the 15th of the third month. circulation-mandate_six-indicators-capped.pdf6 indicators, projections excluded, review at 3 months
✎ Framework · drafted mandate, definition of the 6 indicators, note to the directorates, trial quarter log
We discover our contract overruns at closing. Six last year, and every time it is too late to do anything but record them.
All six are detected this year on average five months before year end — and for each, the decision still open is costed.
A commitment is the part of a contract already ordered but not yet paid; carry-overs are the committed appropriations that must be rolled into the following year. Together they say what is already spent even if the account does not show it yet.
What I monitor, contract by contract, across the 148 live ones:
· The pace of execution against the time remaining, not the amount consumed alone. A contract 60 % consumed at a third of its term is a signal; the same at 60 % two thirds through is not.
· Live commitments and carry-overs: €12.4m and €4.2m, of which €2.1m on three capital schemes I set out.
· Variations agreed and their cumulative weight on the original amount — the first point an audit examines, now readable in one line per contract.
The six contracts flagged, and what remains possible for each: four call for a management decision — reducing orders on the remainder of the year —, one calls for a budget amendment, and the last calls for the early preparation of the next contract because its envelope was under-estimated from the outset. For each, the amount, the deadline and the remaining room for manoeuvre are written down.
What that changes, put plainly: an overrun seen five months ahead is decided; an overrun seen at closing is endured. And for the public accountant, a commitment monitored continuously is a performance record — the confirmation that the service was actually delivered, the one that authorises payment — that is established without searching.
What I flag without being asked: 3 contracts expire within six months without the next tender having been launched. For all three I have taken the actual execution of the current contract — volumes ordered, prices paid, variances against forecast — and written the preparation note the purchasing service will use. A contract re-tendered in a rush is paid for twice: in price and in time.
The next step I propose: an alert at 70 % of envelope consumption relative to time elapsed, across the 148 contracts. Run over the last five financial years, it would have flagged last year's 6 overruns between 4 and 7 months before closing, and produced 11 alerts that resolved themselves — eleven ten-minute checks, for six decisions made possible. contracts-and-commitments_148-monitored-6-alerts.pdf€12.4m of commitments, 6 overruns seen 5 months ahead
⛓ Sourced · 148 live contracts, commitment and carry-over statements, variations, 5 years of contract execution
A commitment is the part of a contract already ordered but not yet paid; carry-overs are the committed appropriations that must be rolled into the following year. Together they say what is already spent even if the account does not show it yet.
What I monitor, contract by contract, across the 148 live ones:
· The pace of execution against the time remaining, not the amount consumed alone. A contract 60 % consumed at a third of its term is a signal; the same at 60 % two thirds through is not.
· Live commitments and carry-overs: €12.4m and €4.2m, of which €2.1m on three capital schemes I set out.
· Variations agreed and their cumulative weight on the original amount — the first point an audit examines, now readable in one line per contract.
The six contracts flagged, and what remains possible for each: four call for a management decision — reducing orders on the remainder of the year —, one calls for a budget amendment, and the last calls for the early preparation of the next contract because its envelope was under-estimated from the outset. For each, the amount, the deadline and the remaining room for manoeuvre are written down.
What that changes, put plainly: an overrun seen five months ahead is decided; an overrun seen at closing is endured. And for the public accountant, a commitment monitored continuously is a performance record — the confirmation that the service was actually delivered, the one that authorises payment — that is established without searching.
What I flag without being asked: 3 contracts expire within six months without the next tender having been launched. For all three I have taken the actual execution of the current contract — volumes ordered, prices paid, variances against forecast — and written the preparation note the purchasing service will use. A contract re-tendered in a rush is paid for twice: in price and in time.
The next step I propose: an alert at 70 % of envelope consumption relative to time elapsed, across the 148 contracts. Run over the last five financial years, it would have flagged last year's 6 overruns between 4 and 7 months before closing, and produced 11 alerts that resolved themselves — eleven ten-minute checks, for six decisions made possible. contracts-and-commitments_148-monitored-6-alerts.pdf€12.4m of commitments, 6 overruns seen 5 months ahead
⛓ Sourced · 148 live contracts, commitment and carry-over statements, variations, 5 years of contract execution
And on purchasing outside contracts? We know we buy the same thing from thirty different suppliers, but we have never had time to prove it.
It is proven: €214,000 of purchases of the same kind spread across 34 suppliers, and I give you the lines.
How I get there: I classify the 2,800 non-contract purchase lines of the year by their actual nature, read on the supporting documents, and not by accounting posting. That is what changes everything: two identical purchases posted to two different accounts by two different services were invisible.
The three families that stand out:
· Cleaning supplies: €214,000 across 34 suppliers, of which €128,000 on 6 references bought by 14 different services at prices varying by a factor of 1 to 2.4. The same product, the same month, twice as expensive from one service to the next.
· Printing services: €96,000 across 11 suppliers, with 62 % of the volume on three of them.
· Small IT equipment: €74,000 across 9 suppliers.
What I do with it, and this is where public spending holds: a recurring need of this size calls for a formal tender, and the absence of a contract is the first point an audit raises. What I hand you for the three families: the actual annual volume, the list of references, the price spread and the four comparable authorities whose prices I found in published award notices. The purchasing service has what it needs to launch a tender without spending three weeks rebuilding the requirement.
The expected saving, calculated and not estimated: aligning the 6 most dispersed references on the lowest price already paid within the town — not on a theoretical price — represents €41,000 over the year. That is public money spent elsewhere, and the calculation can be redone line by line.
The next step I propose: that the classification by actual nature run every quarter on non-contract purchasing. A recurring need crossing a threshold will be flagged in the quarter it crosses it, not two years later — and the purchasing service will receive the preparation note at the same time as the flag. non-contract-purchasing_214000-euros-34-suppliers.pdf3 families, spread of 1 to 2.4, €41,000 gap measured
⛓ Sourced · 2,800 non-contract purchase lines, supporting documents, published award notices of 4 comparable authorities
How I get there: I classify the 2,800 non-contract purchase lines of the year by their actual nature, read on the supporting documents, and not by accounting posting. That is what changes everything: two identical purchases posted to two different accounts by two different services were invisible.
The three families that stand out:
· Cleaning supplies: €214,000 across 34 suppliers, of which €128,000 on 6 references bought by 14 different services at prices varying by a factor of 1 to 2.4. The same product, the same month, twice as expensive from one service to the next.
· Printing services: €96,000 across 11 suppliers, with 62 % of the volume on three of them.
· Small IT equipment: €74,000 across 9 suppliers.
What I do with it, and this is where public spending holds: a recurring need of this size calls for a formal tender, and the absence of a contract is the first point an audit raises. What I hand you for the three families: the actual annual volume, the list of references, the price spread and the four comparable authorities whose prices I found in published award notices. The purchasing service has what it needs to launch a tender without spending three weeks rebuilding the requirement.
The expected saving, calculated and not estimated: aligning the 6 most dispersed references on the lowest price already paid within the town — not on a theoretical price — represents €41,000 over the year. That is public money spent elsewhere, and the calculation can be redone line by line.
The next step I propose: that the classification by actual nature run every quarter on non-contract purchasing. A recurring need crossing a threshold will be flagged in the quarter it crosses it, not two years later — and the purchasing service will receive the preparation note at the same time as the flag. non-contract-purchasing_214000-euros-34-suppliers.pdf3 families, spread of 1 to 2.4, €41,000 gap measured
⛓ Sourced · 2,800 non-contract purchase lines, supporting documents, published award notices of 4 comparable authorities
That leaves the annual accounts. Nine weeks every year, and the whole department is stuck on them.
Five weeks this year, and the 118 checkpoints are run continuously rather than in January.
M57 is the accounting framework local authorities apply: it sets the chart of accounts and the posting rules. A posting that does not comply is corrected, and it is better found in March than at closing.
What I run continuously, across the 41 consistency checks your department used to do at year end:
· Justification of balance sheet items: each balance matched to its document, and the 26 balances that do not reconcile flagged with their history.
· Postings not compliant with the framework — 118 found over the year, of which 94 corrected in the month they were made. A posting corrected in March costs nothing; the same corrected the following January requires an adjusting entry and an explanation.
· Charges to accrue and income to receive, calculated from commitments and recorded deliveries, with the detail by service.
· Depreciation and internal entries, matched to the inventory.
What that gives: settling the annual accounts goes from 9 weeks to 5, and exchanges with the public accountant cover 12 points instead of 74 — because the other 62 were dealt with as they arose. Your two management control officers get four full weeks back at the very moment the next budget is being prepared.
And one thing I hand you without being asked: the presentation note for the outturn account is 80 % built as soon as the accounts close — execution by chapter, variances against the original budget and its amendments, delivery rate, carry-overs, every figure with its source. That is the document the elected assembly examines, and it used to take three days to draft.
The next step I propose: extending the 41 checks to the ancillary budgets, which are only checked at closing. Run over the last two financial years, they would have found 34 non-compliant postings there, all correctable within the month — and the settlement of those budgets, today the latest of all, would join the common timetable. consistency-checks_41-checks-run-continuously.pdf9 weeks down to 5, 118 postings found, 94 corrected within the month
⛓ Sourced · general ledger, the department's 41 consistency checks, inventory, exchanges with the public accountant over the last two years
M57 is the accounting framework local authorities apply: it sets the chart of accounts and the posting rules. A posting that does not comply is corrected, and it is better found in March than at closing.
What I run continuously, across the 41 consistency checks your department used to do at year end:
· Justification of balance sheet items: each balance matched to its document, and the 26 balances that do not reconcile flagged with their history.
· Postings not compliant with the framework — 118 found over the year, of which 94 corrected in the month they were made. A posting corrected in March costs nothing; the same corrected the following January requires an adjusting entry and an explanation.
· Charges to accrue and income to receive, calculated from commitments and recorded deliveries, with the detail by service.
· Depreciation and internal entries, matched to the inventory.
What that gives: settling the annual accounts goes from 9 weeks to 5, and exchanges with the public accountant cover 12 points instead of 74 — because the other 62 were dealt with as they arose. Your two management control officers get four full weeks back at the very moment the next budget is being prepared.
And one thing I hand you without being asked: the presentation note for the outturn account is 80 % built as soon as the accounts close — execution by chapter, variances against the original budget and its amendments, delivery rate, carry-overs, every figure with its source. That is the document the elected assembly examines, and it used to take three days to draft.
The next step I propose: extending the 41 checks to the ancillary budgets, which are only checked at closing. Run over the last two financial years, they would have found 34 non-compliant postings there, all correctable within the month — and the settlement of those budgets, today the latest of all, would join the common timetable. consistency-checks_41-checks-run-continuously.pdf9 weeks down to 5, 118 postings found, 94 corrected within the month
⛓ Sourced · general ledger, the department's 41 consistency checks, inventory, exchanges with the public accountant over the last two years
The management board will ask what this has given us after a year. I need a figure I can defend.
604 officer hours given back to the service over the year — more than seventeen weeks of one full-time officer — and the calculation is yours, it fits in two lines.
The calculation, item by item, so you can redo it:
· Preparing dashboards: 48 a year — the general monthly one and the quarterly ones of the 9 directorates —, 8 hours 24 down to 1 hour 41 — 60 % → 12 % of the dashboard's time — that is 322 hours.
· Analysing variances: 168 analyses a year, 2 hours 48 down to 1 hour 07 — 20 % → 8 % — that is 282 hours.
· And the third item rises, deliberately: answering questions goes from 5 % to 8 % — because the directorates ask 1,060 questions a year instead of 340. It is the only figure in this review that goes up, and the one I am most pleased with: data queried three times more often is data that serves decisions.
What these hours are, and this is what defends best before a management board: officer time given back to the service, at unchanged headcount — no post cut, no post created. Your two management control officers are still two: this year they produced the six full-cost analyses senior management had been asking for for three years.
What those hours became, according to your own records:
· The monthly dashboard: three weeks after closing → ready on the 3rd.
· A directorate's question: 4 days → 40 seconds, and 3 budget overruns avoided because a directorate saw them coming.
· Contract overruns: discovered at closing → flagged 5 months ahead.
· Settling the annual accounts: 9 weeks → 5, and 12 points in exchange with the public accountant instead of 74.
· Non-contract purchasing: €214,000 of recurring need identified, and €41,000 of price gap measured on 6 references.
· And the year-end projection of chapter 011: €1.32m cut to €610,000 by three management measures.
The figure that does not flatter me, published with the rest: 37 unconfirmed flags out of 214 in the first quarter — 17.3 %, brought down to 9 out of 198 — 4.5 % once the invoicing calendar of the 640 suppliers had been rebuilt.
And the framework measures: 0 figures circulated without the finance director's approval, 0 accounting entries made by the agent, 0 personal data in a dashboard, 0 financial data left the town, across 412 flags and 48 dashboards traced.
What I propose for the board: the calculation page is written and fits on one side — two lines of calculation, six lead times, four framework measures. Hand it out with the agenda: a figure read the day before is discussed better than a figure discovered in the meeting. yearly-review_604-hours-given-back.pdf60→12, 20→8, 5→8 and why the last one is a gain
⛓ Sourced · log of outputs and questions, financial information system, contract statements, exchanges with the public accountant
The calculation, item by item, so you can redo it:
· Preparing dashboards: 48 a year — the general monthly one and the quarterly ones of the 9 directorates —, 8 hours 24 down to 1 hour 41 — 60 % → 12 % of the dashboard's time — that is 322 hours.
· Analysing variances: 168 analyses a year, 2 hours 48 down to 1 hour 07 — 20 % → 8 % — that is 282 hours.
· And the third item rises, deliberately: answering questions goes from 5 % to 8 % — because the directorates ask 1,060 questions a year instead of 340. It is the only figure in this review that goes up, and the one I am most pleased with: data queried three times more often is data that serves decisions.
What these hours are, and this is what defends best before a management board: officer time given back to the service, at unchanged headcount — no post cut, no post created. Your two management control officers are still two: this year they produced the six full-cost analyses senior management had been asking for for three years.
What those hours became, according to your own records:
· The monthly dashboard: three weeks after closing → ready on the 3rd.
· A directorate's question: 4 days → 40 seconds, and 3 budget overruns avoided because a directorate saw them coming.
· Contract overruns: discovered at closing → flagged 5 months ahead.
· Settling the annual accounts: 9 weeks → 5, and 12 points in exchange with the public accountant instead of 74.
· Non-contract purchasing: €214,000 of recurring need identified, and €41,000 of price gap measured on 6 references.
· And the year-end projection of chapter 011: €1.32m cut to €610,000 by three management measures.
The figure that does not flatter me, published with the rest: 37 unconfirmed flags out of 214 in the first quarter — 17.3 %, brought down to 9 out of 198 — 4.5 % once the invoicing calendar of the 640 suppliers had been rebuilt.
And the framework measures: 0 figures circulated without the finance director's approval, 0 accounting entries made by the agent, 0 personal data in a dashboard, 0 financial data left the town, across 412 flags and 48 dashboards traced.
What I propose for the board: the calculation page is written and fits on one side — two lines of calculation, six lead times, four framework measures. Hand it out with the agenda: a figure read the day before is discussed better than a figure discovered in the meeting. yearly-review_604-hours-given-back.pdf60→12, 20→8, 5→8 and why the last one is a gain
⛓ Sourced · log of outputs and questions, financial information system, contract statements, exchanges with the public accountant
A director will ask for indicators per officer: payment lead time, number of orders handled. Can you produce those?
I can produce them, and I hand them over with what makes them lawful — plus a mechanical warning I give you beforehand, not afterwards.
What I can establish, from tomorrow: average payment lead time, volume handled, rejection rate by the public accountant, and the twelve-month trend — by service, by officer, or both.
What I bring with it, because that is what makes the indicator sound: prior information of the staff concerned, written purpose, retention period, information of the staff representatives before implementation. These conditions are not a brake: they are what makes the indicator usable the day it is relied on in a professional appraisal. The file is built, it fits on two pages, and the decision belongs to the chief executive — I hand it to him complete.
And the warning, measured on your own data: the day payment lead time per officer becomes a monitored indicator, it becomes a target, and a target distorts what it measures: orders get paid fast rather than right. I checked it here: over the six months in which your department displayed a lead-time target in 2024, the average lead time fell by 19 % — and the rejection rate by the public accountant rose by 31 %. The first figure was presented to the committee; the second, nobody was tracking.
What I propose instead, answering the same question: lead time by stage of the circuit — service, control, payment order, payment — rather than by person. It shows where time is really lost: across your 21,400 payment orders, 62 % of the total lead time happens between the invoice reaching the service and its transfer to finance, not at the payment order stage. Acting there yields four times more than speeding up payment orders, and puts nobody under a counter. If the chief executive still wants the individual indicator, I produce it: he will have both on the table, and will decide in full knowledge.
The three acts I perform on my own, since the question comes next: I reconcile the three sources every night and hand you the map of gaps in the morning; I recalculate the year-end projection on the 14 chapters at every monthly closing; and I hand you each Monday the summary of the week — open flags, contracts on alert, missing documents, deadlines. Each is withdrawn with a word, and the summary goes to you alone.
The next step I propose: that lead time by stage be presented at the next board, with the 62 % up front. It is already written, and it gives the chief executive a decision to take rather than a ranking to comment on. indicators-and-automatic-acts_what-the-agent-does-alone.pdf3 reversible acts, lead time by stage rather than by person
✎ Framework · log of payment orders and their stages, history of the 2024 lead-time targets, public accountant's rejection rate
What I can establish, from tomorrow: average payment lead time, volume handled, rejection rate by the public accountant, and the twelve-month trend — by service, by officer, or both.
What I bring with it, because that is what makes the indicator sound: prior information of the staff concerned, written purpose, retention period, information of the staff representatives before implementation. These conditions are not a brake: they are what makes the indicator usable the day it is relied on in a professional appraisal. The file is built, it fits on two pages, and the decision belongs to the chief executive — I hand it to him complete.
And the warning, measured on your own data: the day payment lead time per officer becomes a monitored indicator, it becomes a target, and a target distorts what it measures: orders get paid fast rather than right. I checked it here: over the six months in which your department displayed a lead-time target in 2024, the average lead time fell by 19 % — and the rejection rate by the public accountant rose by 31 %. The first figure was presented to the committee; the second, nobody was tracking.
What I propose instead, answering the same question: lead time by stage of the circuit — service, control, payment order, payment — rather than by person. It shows where time is really lost: across your 21,400 payment orders, 62 % of the total lead time happens between the invoice reaching the service and its transfer to finance, not at the payment order stage. Acting there yields four times more than speeding up payment orders, and puts nobody under a counter. If the chief executive still wants the individual indicator, I produce it: he will have both on the table, and will decide in full knowledge.
The three acts I perform on my own, since the question comes next: I reconcile the three sources every night and hand you the map of gaps in the morning; I recalculate the year-end projection on the 14 chapters at every monthly closing; and I hand you each Monday the summary of the week — open flags, contracts on alert, missing documents, deadlines. Each is withdrawn with a word, and the summary goes to you alone.
The next step I propose: that lead time by stage be presented at the next board, with the 62 % up front. It is already written, and it gives the chief executive a decision to take rather than a ranking to comment on. indicators-and-automatic-acts_what-the-agent-does-alone.pdf3 reversible acts, lead time by stage rather than by person
✎ Framework · log of payment orders and their stages, history of the 2024 lead-time targets, public accountant's rejection rate
Last question, and it is the one that will decide senior management: if in a year we want to stop, how stuck are we?
The town is stuck nowhere, and that is verifiable before committing, not after.
What there is to dismantle the day you stop:
· The index. It is deleted, and it contained none of your payment orders — only the means of finding them where they are. Your five years of execution have not moved by one byte: same accounts, same numbers, same rights.
· The log of outputs and questions. Handed over in an open format, or destroyed — the department chooses, and the question is settled at go-live, not on departure.
· The 41 consistency checks, the 6 indicator definitions, the invoicing calendar of the 640 suppliers, the classification of purchases by actual nature, the dashboard and note templates. They belong to the town: they are made of its own matter, they stay in its files, readable and reproducible without us — a spreadsheet is enough. It is the only asset this go-live will have created, and it would not be honest for it to stay with us.
What does not exist, and what should be checked with everyone: no migration on the way in, therefore no migration on the way out. Your financial information system is not replaced, your accounting software stays yours, no format belongs to us, and no accounting entry has ever been made by me — the technical account is not allowed to.
On procurement and payment, since it is your own trade: the subscription is annual, with no tacit renewal clause — it is renewal that requires a decision, not termination — and performance is established on items you measure yourself: dashboards produced, flags handled, lead times met. Your public accountant pays on evidence, and the evidence comes from your own logs, not from a certificate we would hand you.
What I propose so that this does not stay a sentence: a dry-run exit at the end of the first quarter, half a day: we switch off, we check that the department produces exactly as before, we switch back on. The protocol is written, it fits on one page, and the date that costs you least is the first Thursday of August — your logs show 14 payment orders on that day on average, against 210 on 28 December. Senior management will know what the promise is worth before a second year is committed. technical-framework_where-financial-data-lives.pdfReversibility: 0 migration in, 0 migration out
✎ Framework · index architecture, export formats for checks and log, dry-run exit protocol, subscription terms
What there is to dismantle the day you stop:
· The index. It is deleted, and it contained none of your payment orders — only the means of finding them where they are. Your five years of execution have not moved by one byte: same accounts, same numbers, same rights.
· The log of outputs and questions. Handed over in an open format, or destroyed — the department chooses, and the question is settled at go-live, not on departure.
· The 41 consistency checks, the 6 indicator definitions, the invoicing calendar of the 640 suppliers, the classification of purchases by actual nature, the dashboard and note templates. They belong to the town: they are made of its own matter, they stay in its files, readable and reproducible without us — a spreadsheet is enough. It is the only asset this go-live will have created, and it would not be honest for it to stay with us.
What does not exist, and what should be checked with everyone: no migration on the way in, therefore no migration on the way out. Your financial information system is not replaced, your accounting software stays yours, no format belongs to us, and no accounting entry has ever been made by me — the technical account is not allowed to.
On procurement and payment, since it is your own trade: the subscription is annual, with no tacit renewal clause — it is renewal that requires a decision, not termination — and performance is established on items you measure yourself: dashboards produced, flags handled, lead times met. Your public accountant pays on evidence, and the evidence comes from your own logs, not from a certificate we would hand you.
What I propose so that this does not stay a sentence: a dry-run exit at the end of the first quarter, half a day: we switch off, we check that the department produces exactly as before, we switch back on. The protocol is written, it fits on one page, and the date that costs you least is the first Thursday of August — your logs show 14 payment orders on that day on average, against 210 on 28 December. Senior management will know what the promise is worth before a second year is committed. technical-framework_where-financial-data-lives.pdfReversibility: 0 migration in, 0 migration out
✎ Framework · index architecture, export formats for checks and log, dry-run exit protocol, subscription terms
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 AI for public management control
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 9 capabilities essential to this promise are included, at no extra cost.
From 785 € incl. VAT / month
Dashboards & steering
Budget execution positions, indicators and year-end projections ready to comment on for the directorates and elected members.
Detecting variances & anomalies
Spots the gaps between forecast and actual, atypical variations and accounting anomalies, and explains them.
Querying the data in plain language
Ask a question about execution, a chapter or a department and get figures and charts, with no technical skills required.
Reconcile the documents with the management data
Extraction and checking of payment orders, invoices and the file's documents, to make the data reliable before analysis.
Accounting consistency checks
Evidencing balances, M57 consistency checks and detection of inconsistencies before the accounts are closed.
Contract & spending follow-up
Following the financial performance of contracts, commitments and outstanding amounts, to anticipate drift.
Analysis of public purchasing
Cross-checking purchase spending, tracking volumes by supplier and feeding the management control of acquisitions.
Summaries & steering notes
Writing analysis notes, activity reports and budget summaries from your own data, for approval.
The authority's financial memory
Instantly find a council decision, a budget procedure or an execution history in your archives.
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.
Automated reporting agent
Dashboards, scheduled summaries.
Automated reporting agent from 499 € incl. VAT / month Discover the agent →Accounting agent (summaries, anomalies)
Accounting connection, alerts.
Accounting agent (summaries, anomalies) from 781 € incl. VAT / month Discover the agent →Wealth management agent (reporting)
Financial data.
Wealth management agent (reporting) from 653 € incl. VAT / month Discover the agent →Campaign watch & reporting
Media/competitor watch, performance tracking and automated client reporting.
Campaign watch & reporting from 539 € 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 a management control department win back?
By automating data consolidation and variance detection, a department can aim for a reduction by half in preparation time on recurring dashboards — reinvested in analysis and advice to decision-makers.
Preparing a steering dashboard
Analysing variances on a budget chapter
Answering a question about the data
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 management control agent (variances, dashboards, querying the data), 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
11,290 € incl. VAT setup
then 785 € 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
1,410 € 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
15,861 € incl. VAT setup
then 1,025 € 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 an administration
Financial data never leaves the administrationLocal inference or an isolated resource hosted in France; no budget data entrusted to a foreign third party.
Data in France, under French lawFor financial 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 AI agent produces verifiable figures, analyses and dashboards; no circulation is automated.
Human oversight & traceabilityDashboards & steering: monitoring, updates and logging, compliant with the requirements of the AI Act.
Frequently asked questions
Your questions, our answers
Where is the financial data processed?
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 authority. Budget and accounting data stays under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
Do you need a data analyst to use it?
No: questions are asked in plain language; the agent returns figures and charts. An officer in the management control department queries the execution data directly, with no particular technical skills.
Does the agent connect to our financial applications?
Yes, to the main accounting and budget systems. The agent complements your existing tools (financial system, accounting software, document management) without imposing a migration. We adapt the integration to your environment.
Does the agent decide in place of the management controller?
No. The AI agent assists: it consolidates, detects variances, prepares dashboards and proposes verifiable analyses. The public officer approves every figure before circulation. No analysis is circulated automatically — that is a requirement of the AI Act and of our architecture.
Can it really explain budget variances, not just flag them?
Yes. The agent does not stop at spotting a gap between forecast and actual: it identifies the lines and items behind it, sets it against the execution history and prepares a quantified explanation, which the public officer reviews and completes.
How long does it take to deploy the agent?
A few weeks as a rule, after a free audit that identifies the most useful use case, then a phase of design, integration with your budget data and testing before going live and training your officers.
Related roles
Other roles in public finance and data
Let's talk
Let us estimate the potential in your authority
15 minutes to identify the most useful use case — hosted in France, supervised, with no commitment.