● Public sector — Tax case handler
AI agent for handlers of tax files
An assistant that carries out the preliminary assessment of claims, gathers the documents and prepares the replies — the handler keeps the analysis and the decision. Hosted in France — on local inference or an isolated resource — taxpayers' tax data never leaves the network. The AI agent assists, the handler decides and signs.
Hosted in France
Tax confidentiality preserved
GDPR & AI Act: governed deployment
A human decision on every file
Updated on
Deployed in a few weeks
Tax files assistant · hosted in France
Do the preliminary assessment of this claim about property tax: the taxpayer is arguing an error in the floor area.
File summarised. The claim is admissible as to form (filed in time, claimant identified). The taxpayer encloses a floor-area certificate and a plan; the difference from the area on record is 18 m². Missing document: the title deed, mentioned but not enclosed.
A draft request for the document is prepared and the elements of analysis gathered — with no proposed decision.
⛓ Sourced · the file as submitted + the enclosures
A draft request for the document is prepared and the elements of analysis gathered — with no proposed decision.
⛓ Sourced · the file as submitted + the enclosures
Prepare the draft reply for the request for time to pay received yesterday.
Draft letter ready from your template: a reminder of the request, the documents produced (evidence of a drop in income), and the two possible outcomes left to your judgement — a payment plan or a reasoned refusal. The decision fields are empty, for you to complete.
To be read over, completed and signed.
✎ Action · draft ready to read over — the officer approves
To be read over, completed and signed.
✎ Action · draft ready to read over — the 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 officer decidesThe agent prepares the ground, it never decides
✦ In brief
A Blue Lemon Agent agent assists the handler of tax files with the repetitive tasks — preliminary assessment of claims and requests (relief, time to pay, updates to a taxpayer's situation), checking of supporting documents, reasoned draft replies — and flags the inconsistencies to be checked. It runs on local inference or is hosted in France on a dedicated and isolated resource: tax 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 legal analysis and the decision remain taken and signed by the handler. Live within a few weeks.
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
7
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 AI matters to tax offices — and why they hesitate
Claims, requests for relief, time to pay, changes of circumstances: each file calls for gathering documents, checking time limits and giving reasons for a reply. Time is short, but the data handled is covered by tax confidentiality.
! The issue
The tax handler is caught between taxpayers who expect a prompt, reasoned reply and files whose assessment — documents, time limits, account positions — eats up time. Yet most consumer AI tools amount to entrusting taxpayers' income, assets and personal circumstances to a third party, often hosted outside Europe and subject to the Cloud Act.
✓ Our answer
For data covered by tax confidentiality, 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, the decision reserved for the handler: complex or ambiguous cases are systematically flagged for human examination, never settled by the machine. The aim is not to replace the handler, but to give them back time for analysis.
The decisive point
Protecting tax files: sovereignty & compliance
A tax office handles data covered by tax confidentiality: income, assets, family circumstances. Here is how the architecture of our agents protects it.
Local inference
The agent can run on a machine at the unit: no data leaves the network, nothing passes through a cloud.
Hosting in France
Otherwise, a dedicated and isolated resource, hosted in France under French law — the tax data: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
Exposure of tax files to the Cloud Act and FISA 702 is reduced by design; location alone does not guarantee immunity.
One isolated resource per department
No pooling of data: an environment strictly dedicated to your unit, guaranteeing the continuity of the public service.
Encryption & controlled access
Encryption in transit and at rest, role-based access (RBAC), strong authentication and logging.
AI Act: governed deployment
The agent is strictly in support; no automated relief or refusal; 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 bodyPersonal tax office of Vallonges — 68,400 tax households
- Sector
- Personal tax office — reception, case handling, contentious and discretionary claims, requests for time to pay; counter open 15 hours a week, telephone line 20 hours
- Headcount
- 9 staff — one head of office, 5 case handlers, 2 reception officers, 1 collection officer; the claims rest on the 5 case handlers
- Public served
- 68,400 tax households, of which 12,300 property owners — 4,900 enquiries a year at the counter, by telephone and through the secure messaging service
- Order of magnitude
- 2,840 claims and requests a year — 1,620 contentious claims, 640 discretionary requests, 580 requests for time to pay; 160 heavy tax-relief files; five years of archived records
- Tools in place
- The office's case-handling applications, the taxpayer's secure messaging service, the online procedures platform and the office's reply templates — the agent plugs into them read-only, nothing is replaced and nothing is migrated
- Who decides
- The case handler examines, qualifies and signs; the head of office decides above the threshold he has set; the public accountant grants time to pay
- Room for improvement
- A claim waits 47 days for its first examination; 38 % of files filed go back with a request for a missing document; 214 requests for time to pay went more than thirty days without an answer; and 41 replies sent had to be redone after the taxpayer challenged them
Fictional public office, invented for the demonstration. At Vallonges, five case handlers examine 2,840 claims and requests a year, in an office where reception and collection already take up four people. Time is not slightly short, it is very short — and the data handled, income, assets and family circumstances, is covered by tax secrecy. The agent runs on local inference on a machine in the office, reads five years of files without ever writing to them, and lets no reply leave without a case handler's signature. The exchanges below span a year, from the overhaul of the claim causes to the review presented to the head of office.
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.
Tax files assistant · hosted in France
I reread your last five financial years over the weekend — 14,200 files, 2,840 a year — and I have drawn out a finding your archives were carrying without anyone having the time to extract it.
A contentious claim challenges the basis or the amount of a tax; a discretionary request challenges nothing but asks for relief in view of the taxpayer's circumstances. Both are examined, but not with the same documents.
The gap measured, and it decides everything else: 1,180 of your 1,620 contentious claims come under just three causes — 73 %.
· Error of area or description on property tax: 612 claims a year. All of them are examined with the same three documents.
· Change of circumstances not taken into account — moving house, death, separation: 348.
· Double taxation on the same premises after a sale: 220.
What these 1,180 files have in common, and nobody had set it side by side: the decisive document is the same in 94 % of cases, and 7 times out of 10 it is already in the taxpayer's file or in yours by the time the claim arrives.
What that costs today:
· A claim waits 47 days for its first examination. The office has six months to rule, beyond which silence counts as rejection and opens the way to the courts — 47 days of waiting serve no one, neither the taxpayer nor the case handler.
· 38 % of the files filed go back with a request for a missing document, and each round trip costs the taxpayer three weeks.
· The preliminary examination of a routine claim accounts for 60 % of the file's time: 36 minutes out of the 60 that separate its arrival from the signed reply.
What I propose, and it is not a promise — it is already done: I carried out the preliminary examination of the 1,180 claims of the past year, as a dry run. Each comes back with its admissibility checked, its chronology, its documents dated, the gap measured and the missing documents named.
The time this shifts: the preliminary examination goes from 60 % to 10 % of the file's time — from 36 minutes to 6. Over 1,180 routine claims a year, that is 590 hours returning to the office.
The next step, and it takes one meeting: your head of office reads the three examination sheets I have written, cause by cause, with the number of files each one covers. He settles them, and they run on the claims that arrive on Monday — bringing them into service is the only act I leave to you. claim-causes_1180-files-out-of-1620.pdf3 causes behind 73 % of claims, each measured
⛓ Sourced · 5 financial years of office files, 14,200 claims and requests, log of examination times
A contentious claim challenges the basis or the amount of a tax; a discretionary request challenges nothing but asks for relief in view of the taxpayer's circumstances. Both are examined, but not with the same documents.
The gap measured, and it decides everything else: 1,180 of your 1,620 contentious claims come under just three causes — 73 %.
· Error of area or description on property tax: 612 claims a year. All of them are examined with the same three documents.
· Change of circumstances not taken into account — moving house, death, separation: 348.
· Double taxation on the same premises after a sale: 220.
What these 1,180 files have in common, and nobody had set it side by side: the decisive document is the same in 94 % of cases, and 7 times out of 10 it is already in the taxpayer's file or in yours by the time the claim arrives.
What that costs today:
· A claim waits 47 days for its first examination. The office has six months to rule, beyond which silence counts as rejection and opens the way to the courts — 47 days of waiting serve no one, neither the taxpayer nor the case handler.
· 38 % of the files filed go back with a request for a missing document, and each round trip costs the taxpayer three weeks.
· The preliminary examination of a routine claim accounts for 60 % of the file's time: 36 minutes out of the 60 that separate its arrival from the signed reply.
What I propose, and it is not a promise — it is already done: I carried out the preliminary examination of the 1,180 claims of the past year, as a dry run. Each comes back with its admissibility checked, its chronology, its documents dated, the gap measured and the missing documents named.
The time this shifts: the preliminary examination goes from 60 % to 10 % of the file's time — from 36 minutes to 6. Over 1,180 routine claims a year, that is 590 hours returning to the office.
The next step, and it takes one meeting: your head of office reads the three examination sheets I have written, cause by cause, with the number of files each one covers. He settles them, and they run on the claims that arrive on Monday — bringing them into service is the only act I leave to you. claim-causes_1180-files-out-of-1620.pdf3 causes behind 73 % of claims, each measured
⛓ Sourced · 5 financial years of office files, 14,200 claims and requests, log of examination times
There are five of us handling all that. I am not going to open three projects at once: where do we start?
With a single cause, not three — and I chose it on the number of files it frees up, not on its difficulty.
The ranking, with the count behind it:
· Property tax on an area error: 612 files a year, that is 38 % of your contentious claims on its own. Three documents are enough to examine it in 94 % of cases: the property record, the declaration of the premises' area, and the document the taxpayer produces — plan, surveyor's certificate or notarial deed.
· Change of circumstances: 348 files, four documents, and a cross-check against data the office already holds.
· Double taxation after a sale: 220 files, two documents, and it is the quickest cause to settle once the date of sale is established.
That property tax cause is also the one that costs the taxpayer most: of the 612 files, 287 led to tax relief — tax relief is the discharge, whole or partial, of a tax wrongly levied — and those 287 taxpayers waited on average 94 days for money that was theirs.
What I do on top, and that nobody has time for: I did not merely sort the 612 files — I wrote the examination sheet for the cause, in the form of your office notes, with the order of the checks, the three required documents, the formula for the area gap and the two cases in which the cause does not apply. Then I ran it over the year's 612 files: it covers 578 with no intervention and sends 34 back to your judgement, all of them mixed-use premises.
You judge on a written and already tested sheet, not on an intention — and you correct it in one word, since it is your own material that I have set out clearly.
The next step I propose: let the sheet run on property tax claims for a fortnight, alongside your usual examination. You compare the two outputs file by file; if mine adds nothing, it stops at a word. On the 26 trial files already run, it found 4 documents the taxpayer had filed and that nobody had attached to the file. examination-sheet_property-tax-612-files.pdf3 documents, 578 files covered out of 612, 34 sent to the case handler
⛓ Sourced · count of claims by cause over 5 financial years, written examination sheet, 26 trial files
The ranking, with the count behind it:
· Property tax on an area error: 612 files a year, that is 38 % of your contentious claims on its own. Three documents are enough to examine it in 94 % of cases: the property record, the declaration of the premises' area, and the document the taxpayer produces — plan, surveyor's certificate or notarial deed.
· Change of circumstances: 348 files, four documents, and a cross-check against data the office already holds.
· Double taxation after a sale: 220 files, two documents, and it is the quickest cause to settle once the date of sale is established.
That property tax cause is also the one that costs the taxpayer most: of the 612 files, 287 led to tax relief — tax relief is the discharge, whole or partial, of a tax wrongly levied — and those 287 taxpayers waited on average 94 days for money that was theirs.
What I do on top, and that nobody has time for: I did not merely sort the 612 files — I wrote the examination sheet for the cause, in the form of your office notes, with the order of the checks, the three required documents, the formula for the area gap and the two cases in which the cause does not apply. Then I ran it over the year's 612 files: it covers 578 with no intervention and sends 34 back to your judgement, all of them mixed-use premises.
You judge on a written and already tested sheet, not on an intention — and you correct it in one word, since it is your own material that I have set out clearly.
The next step I propose: let the sheet run on property tax claims for a fortnight, alongside your usual examination. You compare the two outputs file by file; if mine adds nothing, it stops at a word. On the 26 trial files already run, it found 4 documents the taxpayer had filed and that nobody had attached to the file. examination-sheet_property-tax-612-files.pdf3 documents, 578 files covered out of 612, 34 sent to the case handler
⛓ Sourced · count of claims by cause over 5 financial years, written examination sheet, 26 trial files
All of that assumes you read our taxpayers' income, assets and family circumstances. Where exactly do they go?
They go nowhere. I run on local inference on a machine in the office, and nothing I read leaves the network.
Local inference means the model computes on your own machine: a household's income line or the title deed of a property crosses no external network to be processed. If the office would rather not host a machine, the other route is an isolated resource hosted in France, dedicated to Vallonges — no pooling with another office, which is the condition for the continuity of your activity.
What that changes, point by point:
· Taxpayer data trains no model, neither ours nor a third party's.
· I work read-only, and the technical account I read through has no right to write — this is not a promise, it is a permission that can be checked with one command, and your IT team will run it in front of you.
· Encryption in transit and at rest, role-based access — rights follow the job: the reception officer opens the status of a file, not a household's income. 9 roles for your 9 staff, and the log shows 0 out-of-role accesses 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.
And the point that decides, for an office whose every file is covered by tax secrecy: tax secrecy forbids disclosing to a third party what the administration knows about a taxpayer; it binds every officer, and it binds the tools they use. The access log is what lets you say, the day a taxpayer asks, exactly who opened their file, when and why. That is evidence your current tools did not give you, and it works in your favour.
The figure that sums all this up: 0 taxpayer data out of the office network across the 2,840 files of the year, and processing in the EU targeted.
What I propose: that I keep up to date the record your data protection officer will ask for — hosting, data processed, retention periods, who accesses what, and the exact permissions of the technical account. It is asked for once a year and takes two days to rebuild; the first version is written and you have it attached. technical-framework_where-tax-files-live.pdfLocal inference, read-only, processing in the EU targeted
✎ Framework · deployment architecture, technical account permissions, access log, first version of the processing record
Local inference means the model computes on your own machine: a household's income line or the title deed of a property crosses no external network to be processed. If the office would rather not host a machine, the other route is an isolated resource hosted in France, dedicated to Vallonges — no pooling with another office, which is the condition for the continuity of your activity.
What that changes, point by point:
· Taxpayer data trains no model, neither ours nor a third party's.
· I work read-only, and the technical account I read through has no right to write — this is not a promise, it is a permission that can be checked with one command, and your IT team will run it in front of you.
· Encryption in transit and at rest, role-based access — rights follow the job: the reception officer opens the status of a file, not a household's income. 9 roles for your 9 staff, and the log shows 0 out-of-role accesses 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.
And the point that decides, for an office whose every file is covered by tax secrecy: tax secrecy forbids disclosing to a third party what the administration knows about a taxpayer; it binds every officer, and it binds the tools they use. The access log is what lets you say, the day a taxpayer asks, exactly who opened their file, when and why. That is evidence your current tools did not give you, and it works in your favour.
The figure that sums all this up: 0 taxpayer data out of the office network across the 2,840 files of the year, and processing in the EU targeted.
What I propose: that I keep up to date the record your data protection officer will ask for — hosting, data processed, retention periods, who accesses what, and the exact permissions of the technical account. It is asked for once a year and takes two days to rebuild; the first version is written and you have it attached. technical-framework_where-tax-files-live.pdfLocal inference, read-only, processing in the EU targeted
✎ Framework · deployment architecture, technical account permissions, access log, first version of the processing record
Here is a claim that arrived yesterday: property tax, the taxpayer alleges an error of area. It is the most common case and the one that drags on most.
The file has been examined and it fits on one page: admissibility checked, chronology established, documents dated, gap measured, missing document named.
What I checked on the form first, because that is what gets checked first: the claimant is indeed the taxpayer entered on the register; the claim concerns the tax and the year it names; it arrived within the period open to the taxpayer, and I set the date of arrival and the end of that period side by side so the point can be checked at a glance.
What the substance gives:
· Area on the register: 148 m². Area on the surveyor's certificate produced by the taxpayer: 130 m². The gap is 18 m².
· The property record and the latest declaration of the premises agree with each other and both give the area of 148 m² since 2019.
· One document is missing, and I name it: the title deed is mentioned in the claim but not attached. The request for the document is written, addressed to the taxpayer, and goes out as soon as you agree.
What I hand over on top, and which existed nowhere: the chronology of the premises over five years — declarations, declared changes, taxes levied —, and the comparison with the two neighbouring files in the same building, where the same measurement method was used. It is that comparison that gives the case handler something to judge on, and it took half a day to rebuild.
What I leave to you, and it is exactly where the value is: the 18 m² gap is a fact that I establish; whether it justifies relief is a legal qualification, and it is yours. What I hand you is the complete file in 6 minutes instead of 36, and the decision instead of the rebuilding.
The time this shifts: the preliminary examination goes from 60 % to 10 % of the file's time — from 36 minutes to 6. Over 1,180 routine claims a year, that is 590 hours returned to the office.
The next step: the request for the title deed goes out as soon as you agree, and the file comes back to you complete the day the document arrives — I tell you that same day, not at the stock review three weeks later. preliminary-examination_property-tax-18-sqm-gap.pdfAdmissibility, 5-year chronology, gap measured, missing document named
⛓ Sourced · claim as filed, property record, declarations of the premises over 5 years, neighbouring files in the same building
What I checked on the form first, because that is what gets checked first: the claimant is indeed the taxpayer entered on the register; the claim concerns the tax and the year it names; it arrived within the period open to the taxpayer, and I set the date of arrival and the end of that period side by side so the point can be checked at a glance.
What the substance gives:
· Area on the register: 148 m². Area on the surveyor's certificate produced by the taxpayer: 130 m². The gap is 18 m².
· The property record and the latest declaration of the premises agree with each other and both give the area of 148 m² since 2019.
· One document is missing, and I name it: the title deed is mentioned in the claim but not attached. The request for the document is written, addressed to the taxpayer, and goes out as soon as you agree.
What I hand over on top, and which existed nowhere: the chronology of the premises over five years — declarations, declared changes, taxes levied —, and the comparison with the two neighbouring files in the same building, where the same measurement method was used. It is that comparison that gives the case handler something to judge on, and it took half a day to rebuild.
What I leave to you, and it is exactly where the value is: the 18 m² gap is a fact that I establish; whether it justifies relief is a legal qualification, and it is yours. What I hand you is the complete file in 6 minutes instead of 36, and the decision instead of the rebuilding.
The time this shifts: the preliminary examination goes from 60 % to 10 % of the file's time — from 36 minutes to 6. Over 1,180 routine claims a year, that is 590 hours returned to the office.
The next step: the request for the title deed goes out as soon as you agree, and the file comes back to you complete the day the document arrives — I tell you that same day, not at the stock review three weeks later. preliminary-examination_property-tax-18-sqm-gap.pdfAdmissibility, 5-year chronology, gap measured, missing document named
⛓ Sourced · claim as filed, property record, declarations of the premises over 5 years, neighbouring files in the same building
The documents, precisely. Thirty-eight per cent of our files go back with a request for a missing document, and that is where we lose taxpayers.
38 % yesterday, 6 % over the checked quarter — and the gain does not come from a relaxed requirement, it comes from when the document is asked for.
What I check on every document filed: its nature (is it the one the cause requires), its date (is it later than the fact it establishes), its legibility, its consistency with the other documents in the file, and the existence of a more recent version already filed by the taxpayer.
What the check found over the quarter, on 710 files:
· 486 files complete on filing.
· 182 with a document missing — named the same day, with the written request ready to go.
· 29 containing a document out of date or earlier than the fact it was meant to establish.
· 13 where a more recent document was already in the taxpayer's own space and had not been attached to the file — thirteen requests that never had to go out at all.
What changes for the taxpayer, and that is the real subject: yesterday they learned a document was missing three weeks after filing; today they learn it the same day, and they know which one. Of the 182 requests, 164 received their document within eight days, against 31 days on average last year.
And the effect I did not expect, but which can be measured: the abandonment rate. Last year, 74 taxpayers never replied to a request for a document and their claim lapsed without ever being examined. Over the quarter: 6. That is 68 files a year that reach examination instead of disappearing — and it is equal treatment, because those who abandon are never the ones with an adviser.
The next step I propose: that the list of documents required per cause be displayed at the counter and in the online space, before filing. It is written for the three causes, in everyday wording, and takes a quarter of an hour to read — it is the only measure in this demonstration that acts before a file even arrives. document-check_710-files-of-the-quarter.pdf38 % → 6 % incomplete, 74 abandonments down to 6
⛓ Sourced · 710 files of the quarter, documents filed and their dates, taxpayer's online space, log of document requests
What I check on every document filed: its nature (is it the one the cause requires), its date (is it later than the fact it establishes), its legibility, its consistency with the other documents in the file, and the existence of a more recent version already filed by the taxpayer.
What the check found over the quarter, on 710 files:
· 486 files complete on filing.
· 182 with a document missing — named the same day, with the written request ready to go.
· 29 containing a document out of date or earlier than the fact it was meant to establish.
· 13 where a more recent document was already in the taxpayer's own space and had not been attached to the file — thirteen requests that never had to go out at all.
What changes for the taxpayer, and that is the real subject: yesterday they learned a document was missing three weeks after filing; today they learn it the same day, and they know which one. Of the 182 requests, 164 received their document within eight days, against 31 days on average last year.
And the effect I did not expect, but which can be measured: the abandonment rate. Last year, 74 taxpayers never replied to a request for a document and their claim lapsed without ever being examined. Over the quarter: 6. That is 68 files a year that reach examination instead of disappearing — and it is equal treatment, because those who abandon are never the ones with an adviser.
The next step I propose: that the list of documents required per cause be displayed at the counter and in the online space, before filing. It is written for the three causes, in everyday wording, and takes a quarter of an hour to read — it is the only measure in this demonstration that acts before a file even arrives. document-check_710-files-of-the-quarter.pdf38 % → 6 % incomplete, 74 abandonments down to 6
⛓ Sourced · 710 files of the quarter, documents filed and their dates, taxpayer's online space, log of document requests
And the delay? Forty-seven days before the first examination is what we are criticised for most.
47 days last year, 9 over the checked quarter — and I give you the breakdown, because it explains where the time was being lost.
The three segments, measured on your own timestamps:
· From arrival of the claim to allocation to a case handler: 21 days → 1. That is the segment that carried the most, and it was nobody's fault: the file was waiting for Monday's stock review.
· From allocation to first examination: 18 days → 6.
· From first examination to the signed reply: 8 days → 2.
What that gives end to end: a routine claim used to get its reply in 94 days; it gets it in 27. The office has six months to rule — and a reply given in 27 days is still received by the taxpayer as a reply, not as a reminder of their own file.
What it changes for tax relief: 287 taxpayers obtained relief last year after waiting 94 days. At the pace of the checked quarter they would obtain it in 27 — and for a household that paid a tax it did not owe, those 67 days are not a statistic.
What I raise without being asked: 214 requests for time to pay had gone more than thirty days without an answer. All 214 have been taken up, sorted by age and by amount, and 198 come back with their draft reply already written — granting time to pay belongs to the public accountant, and the file reaches him complete. The other 16 are waiting on a document, and the request is written for each.
The next step I propose: the table of files by age, delivered every Monday morning, with the end of the six-month period set against each one. It is written, it fits on one page, and it means no file ever reminds you of itself through a lawyer's letter. examination-time_47-days-down-to-9.pdfThree segments measured, 94-day reply down to 27
⛓ Sourced · arrival, allocation and reply timestamps over two quarters, register of requests for time to pay
The three segments, measured on your own timestamps:
· From arrival of the claim to allocation to a case handler: 21 days → 1. That is the segment that carried the most, and it was nobody's fault: the file was waiting for Monday's stock review.
· From allocation to first examination: 18 days → 6.
· From first examination to the signed reply: 8 days → 2.
What that gives end to end: a routine claim used to get its reply in 94 days; it gets it in 27. The office has six months to rule — and a reply given in 27 days is still received by the taxpayer as a reply, not as a reminder of their own file.
What it changes for tax relief: 287 taxpayers obtained relief last year after waiting 94 days. At the pace of the checked quarter they would obtain it in 27 — and for a household that paid a tax it did not owe, those 67 days are not a statistic.
What I raise without being asked: 214 requests for time to pay had gone more than thirty days without an answer. All 214 have been taken up, sorted by age and by amount, and 198 come back with their draft reply already written — granting time to pay belongs to the public accountant, and the file reaches him complete. The other 16 are waiting on a document, and the request is written for each.
The next step I propose: the table of files by age, delivered every Monday morning, with the end of the six-month period set against each one. It is written, it fits on one page, and it means no file ever reminds you of itself through a lawyer's letter. examination-time_47-days-down-to-9.pdfThree segments measured, 94-day reply down to 27
⛓ Sourced · arrival, allocation and reply timestamps over two quarters, register of requests for time to pay
On the file we just looked at, I need the reply. Our templates are dated and I rewrite the reasoning every time.
The draft is written on your template, reasoned on the file's documents, and the two possible outcomes are written side by side so that all you have to do is keep yours.
What the draft contains:
· The restatement of the request, in the taxpayer's words and not in ours — that is what makes them recognise themselves in the reply.
· The documents produced and retained, each with its date, and those set aside, with the reason they were.
· The reasoning, written on the facts of the file: the area on the register, the certified area, the gap, the period concerned.
· Two outcomes, both written out: relief to the extent of the gap established, with the calculation in plain figures; or a reasoned rejection, with the avenues and time limits for appeal. The decision fields are empty.
What I took from your templates, and what I flagged: you have 34 reply templates; 7 still point to an office that has changed address and 2 to a superseded calculation formula. The 9 updated versions are written, replaced line and replacing line side by side — and over the past year those 9 templates produced 486 replies.
The legal qualification and the signature stay with the case handler — and I hand both to them in minutes, complete file, dated documents and the reasoning already drafted. It is your signature that makes the reply challengeable before the courts, and therefore defensible — a reply nobody had signed could be challenged by nobody, and would be worth nothing to the taxpayer.
The time this shifts: the draft reply from a template goes from 40 % to 10 % of the reply's time — from 12 minutes to 3. Over 2,120 replies written from a template each year, that is 318 hours returned to the office.
The next step I propose: that I write the updated version of any template amended by an office note, in the week following its circulation. Your 34 templates will stop ageing in silence — and it will cost you no more than one reading per note. draft-reply_reasoned-with-two-outcomes.pdf12 min → 3 min, 9 templates updated out of 34
⛓ Sourced · the office's 34 reply templates, office notes, documents in the file, 9 rewritten versions
What the draft contains:
· The restatement of the request, in the taxpayer's words and not in ours — that is what makes them recognise themselves in the reply.
· The documents produced and retained, each with its date, and those set aside, with the reason they were.
· The reasoning, written on the facts of the file: the area on the register, the certified area, the gap, the period concerned.
· Two outcomes, both written out: relief to the extent of the gap established, with the calculation in plain figures; or a reasoned rejection, with the avenues and time limits for appeal. The decision fields are empty.
What I took from your templates, and what I flagged: you have 34 reply templates; 7 still point to an office that has changed address and 2 to a superseded calculation formula. The 9 updated versions are written, replaced line and replacing line side by side — and over the past year those 9 templates produced 486 replies.
The legal qualification and the signature stay with the case handler — and I hand both to them in minutes, complete file, dated documents and the reasoning already drafted. It is your signature that makes the reply challengeable before the courts, and therefore defensible — a reply nobody had signed could be challenged by nobody, and would be worth nothing to the taxpayer.
The time this shifts: the draft reply from a template goes from 40 % to 10 % of the reply's time — from 12 minutes to 3. Over 2,120 replies written from a template each year, that is 318 hours returned to the office.
The next step I propose: that I write the updated version of any template amended by an office note, in the week following its circulation. Your 34 templates will stop ageing in silence — and it will cost you no more than one reading per note. draft-reply_reasoned-with-two-outcomes.pdf12 min → 3 min, 9 templates updated out of 34
⛓ Sourced · the office's 34 reply templates, office notes, documents in the file, 9 rewritten versions
What interests me most are the files where something does not add up between the documents and what has been declared. We never have time to look for them.
I look for them continuously, and I flagged 340 over the year — each with the two documents that diverge, set side by side.
The five families I raise, counted over the year:
· Declared area different from the area on a document in the same file: 118.
· Occupancy of premises inconsistent with the circumstances declared elsewhere: 76.
· Date of an event — sale, death, move — differing between two documents: 64.
· Document produced earlier than the fact it is meant to establish: 52.
· Amount claimed unrelated to the tax being challenged: 30.
What I hand over for each one, and this is where I can serve you: the two documents side by side, the exact line that diverges, the date of each document, and whatever else the file contains on the same point. I qualify nothing: a divergence may be a keying error, an undeclared change, or a misread document — and those three do not call for the same outcome.
What the office did with the 340: 214 were material errors, corrected in the taxpayer's favour — and 96 of them concerned files where the taxpayer had asked for nothing. 78 called for a further document. 48 were passed to the competent service under the rule your head of office has set.
What I do on top, and that nobody has time for: I wrote three new flags and ran them over your five financial years, so that you choose on figures.
· Premises declared vacant for more than three years yet showing water consumption on the municipal meter: 41 files over five years, 33 confirmed after examination.
· Two separate claims on the same premises for the same year: 27 files, 27 confirmed. Twenty-seven out of twenty-seven, not one useless flag.
· Gap of more than 20 % between the declared area and the average of comparable premises in the same building: 186 files, 44 confirmed. One in four: it is for you to say whether that is worth the reading.
The next step: open the second one, it produces only 27 flags over five years and it misses none. It runs tomorrow morning and I give you its first statement at D+3. inconsistencies-flagged_340-over-the-year.pdf5 families, 214 errors corrected in the taxpayer's favour
⛓ Sourced · five financial years replayed flag by flag, documents in the files, log of confirmations after examination
The five families I raise, counted over the year:
· Declared area different from the area on a document in the same file: 118.
· Occupancy of premises inconsistent with the circumstances declared elsewhere: 76.
· Date of an event — sale, death, move — differing between two documents: 64.
· Document produced earlier than the fact it is meant to establish: 52.
· Amount claimed unrelated to the tax being challenged: 30.
What I hand over for each one, and this is where I can serve you: the two documents side by side, the exact line that diverges, the date of each document, and whatever else the file contains on the same point. I qualify nothing: a divergence may be a keying error, an undeclared change, or a misread document — and those three do not call for the same outcome.
What the office did with the 340: 214 were material errors, corrected in the taxpayer's favour — and 96 of them concerned files where the taxpayer had asked for nothing. 78 called for a further document. 48 were passed to the competent service under the rule your head of office has set.
What I do on top, and that nobody has time for: I wrote three new flags and ran them over your five financial years, so that you choose on figures.
· Premises declared vacant for more than three years yet showing water consumption on the municipal meter: 41 files over five years, 33 confirmed after examination.
· Two separate claims on the same premises for the same year: 27 files, 27 confirmed. Twenty-seven out of twenty-seven, not one useless flag.
· Gap of more than 20 % between the declared area and the average of comparable premises in the same building: 186 files, 44 confirmed. One in four: it is for you to say whether that is worth the reading.
The next step: open the second one, it produces only 27 flags over five years and it misses none. It runs tomorrow morning and I give you its first statement at D+3. inconsistencies-flagged_340-over-the-year.pdf5 families, 214 errors corrected in the taxpayer's favour
⛓ Sourced · five financial years replayed flag by flag, documents in the files, log of confirmations after examination
And when you get it wrong? Because it will happen in the end, and it is my signature at the bottom of the page.
It has already happened, I counted it, and I publish it for you: of the 470 replies prepared in the first quarter, 41 had to be redone after the taxpayer challenged them — 8.7 %. That is the figure that does not flatter me, and it is better that it comes from me than from a lawyer's letter.
The real cause, measured and not assumed: 33 of the 41 concerned files containing several versions of one and the same document — a certificate refiled after correction, a corrected assessment, an updated plan. I was retaining the first version filed, whereas the taxpayer had produced a more recent one. It was not a reading error: it was a rule of choice that was missing. The other 8 concerned mixed-use premises, with no comparable precedent in your five financial years.
What I did about it, and it is already written: the document retained is now the most recent one at the date of examination, and the earlier versions are listed alongside with their dates, so that the case handler sees what was set aside and why.
· Replayed over the five financial years, the rule changes the document retained in 1,840 files, and in all 1,840 it retains the one the taxpayer had produced last.
· Of the quarter's 41 redone replies, 33 would not have happened.
The following quarter, rule in force: 452 replies prepared, 9 redone — 2.0 %. And the 9 are judgements on one-off cases, not one of them is a version problem.
The rule that holds all the rest together: a value I have not read, I do not write — I ask for it, and I ask quickly. An area that cannot be found does not become “probably unchanged”, a missing date does not become “presumably that of the deed”: I say what is missing, where I looked, who holds it, and I hand you the request already written to its recipient. Over the quarter, 182 missing documents, 182 requests prepared, 164 replies back within eight days.
And the protection that matters for your signature: across 922 replies over two quarters, 922 were read and signed by a case handler, and of the 50 corrections, 41 were made before sending — the remaining 9 are those of the second quarter, and they concern cases nobody would have settled differently first time.
What I now propose: that the 8 mixed-use cases become a fourth examination sheet, written and ready for your reading. Of the 9 redone replies in the second quarter, 6 already fell within its scope — it is the same correction, applied one notch further. replies-redone_41-then-9.pdf8.7 % → 2.0 %, cause measured, 922 human signatures
⛓ Sourced · log of replies and of redone replies over two quarters, 5 financial years replayed with the version rule
The real cause, measured and not assumed: 33 of the 41 concerned files containing several versions of one and the same document — a certificate refiled after correction, a corrected assessment, an updated plan. I was retaining the first version filed, whereas the taxpayer had produced a more recent one. It was not a reading error: it was a rule of choice that was missing. The other 8 concerned mixed-use premises, with no comparable precedent in your five financial years.
What I did about it, and it is already written: the document retained is now the most recent one at the date of examination, and the earlier versions are listed alongside with their dates, so that the case handler sees what was set aside and why.
· Replayed over the five financial years, the rule changes the document retained in 1,840 files, and in all 1,840 it retains the one the taxpayer had produced last.
· Of the quarter's 41 redone replies, 33 would not have happened.
The following quarter, rule in force: 452 replies prepared, 9 redone — 2.0 %. And the 9 are judgements on one-off cases, not one of them is a version problem.
The rule that holds all the rest together: a value I have not read, I do not write — I ask for it, and I ask quickly. An area that cannot be found does not become “probably unchanged”, a missing date does not become “presumably that of the deed”: I say what is missing, where I looked, who holds it, and I hand you the request already written to its recipient. Over the quarter, 182 missing documents, 182 requests prepared, 164 replies back within eight days.
And the protection that matters for your signature: across 922 replies over two quarters, 922 were read and signed by a case handler, and of the 50 corrections, 41 were made before sending — the remaining 9 are those of the second quarter, and they concern cases nobody would have settled differently first time.
What I now propose: that the 8 mixed-use cases become a fourth examination sheet, written and ready for your reading. Of the 9 redone replies in the second quarter, 6 already fell within its scope — it is the same correction, applied one notch further. replies-redone_41-then-9.pdf8.7 % → 2.0 %, cause measured, 922 human signatures
⛓ Sourced · log of replies and of redone replies over two quarters, 5 financial years replayed with the version rule
The heavy files are what blocks everything: an estate, three properties, five years of taxes. Half a day just to understand before even starting.
The summary is done: chronology, documents, amounts at stake and points to settle, on one page — and the complete file stays underneath, document by document.
What the summary carries, on the estate file you pulled out:
· The chronology: 31 dated events over five years, from the death to the three taxes challenged, each with the document that establishes it.
· The three properties, their area, their regime and their occupancy situation year by year.
· The amounts at stake: the tax challenged on each property and each year, the total claimed, and the part that can no longer be challenged because it was not claimed within the period open to the taxpayer — that point decides half the file, and it is now visible on the first line.
· The four points to settle, each with the documents that support them and what the file does not say.
· The two missing documents, named, with the written request ready to go to the notaries concerned.
The time this shifts: assembling and checking a tax-relief file took all of the preparatory time — 2 hours; 20 % of it remains, that is 24 minutes. Over 160 heavy files a year, that is 256 hours returned to the office.
And what it changes for the case handler, even more than the hours: they open the file already knowing what the four points to settle are. The difficulty of an estate file has never been understanding the law: it has always been rebuilding the facts.
The decision to grant relief stays with the competent body — and I hand it to them reasoned, measured and documented within four hours, instead of the half day it took merely to read the file.
The next step I propose: that the year's 160 heavy files be summarised in advance, as they arrive, without waiting for a case handler to open them. On the 22 already handled this way, the case handler gave their decision on average 31 days earlier — and 5 of the 22 could be closed with no further document, because the missing one had been requested on arrival. complex-file-summary_estate-three-properties.pdf2 h → 24 min, 31 dated events, 4 points to settle
⛓ Sourced · estate file, deeds and documents produced, taxes of the 5 financial years, files of the three properties
What the summary carries, on the estate file you pulled out:
· The chronology: 31 dated events over five years, from the death to the three taxes challenged, each with the document that establishes it.
· The three properties, their area, their regime and their occupancy situation year by year.
· The amounts at stake: the tax challenged on each property and each year, the total claimed, and the part that can no longer be challenged because it was not claimed within the period open to the taxpayer — that point decides half the file, and it is now visible on the first line.
· The four points to settle, each with the documents that support them and what the file does not say.
· The two missing documents, named, with the written request ready to go to the notaries concerned.
The time this shifts: assembling and checking a tax-relief file took all of the preparatory time — 2 hours; 20 % of it remains, that is 24 minutes. Over 160 heavy files a year, that is 256 hours returned to the office.
And what it changes for the case handler, even more than the hours: they open the file already knowing what the four points to settle are. The difficulty of an estate file has never been understanding the law: it has always been rebuilding the facts.
The decision to grant relief stays with the competent body — and I hand it to them reasoned, measured and documented within four hours, instead of the half day it took merely to read the file.
The next step I propose: that the year's 160 heavy files be summarised in advance, as they arrive, without waiting for a case handler to open them. On the 22 already handled this way, the case handler gave their decision on average 31 days earlier — and 5 of the 22 could be closed with no further document, because the missing one had been requested on arrival. complex-file-summary_estate-three-properties.pdf2 h → 24 min, 31 dated events, 4 points to settle
⛓ Sourced · estate file, deeds and documents produced, taxes of the 5 financial years, files of the three properties
That leaves the taxpayers who call or come in. I cannot read every reply before it goes out, and I will not sign a blank cheque either.
You do not have to choose between the two: I propose a written mandate, capped, dated and withdrawable in one word — and it is already drafted.
What it brings first, because that is what decides: of the 4,900 enquiries of the year, 3,640 fall under eleven subjects, always the same — documents to provide per cause, time limit for claiming, where my file stands, how to pay, how to ask for time, where to find my assessment, whom to contact for which tax. The eleven answers are written, sourced and dated, and they are waiting for you.
What the mandate authorises, and it fits in six lines:
· Scope: the eleven general-information subjects, listed by name, and the status of a file filed by the taxpayer themselves, once authenticated in their own space. Nothing else.
· No answer on the substance of a decision, nor on a file under consideration: there, I hand you the reply already written, documents alongside, and you send it. Over the trial quarter, those replies went out within 4 hours instead of 47 days — and they went out from you, which makes them reasoned and challengeable. General information does not carry that weight.
· Every reply carries its source — the examination sheet, the office note or the public page, with its date — and the statement that it was prepared by a digital assistant of the office. I say so from the first sentence: the European regulation on artificial intelligence requires that anyone interacting with an AI system be informed, and the taxpayer can ask for a person at any time.
· You receive every morning the statement of the replies sent the day before, on one page. A wrong reply is put right in an hour, not in three weeks.
· Duration: review at three months. Without an explicit decision at the review, the mandate stops — renewal takes a signature, stopping does not.
· Withdrawal: one word from you, and direct sending stops within the minute.
What it changes for the taxpayer: the telephone line is open 20 hours a week; the eleven subjects are handled at any hour, including Saturdays. Over the trial quarter, 61 % of enquiries on those eleven subjects came in outside opening hours.
The next step: sign the mandate and the office answers that very night; the review is already in your diary on the 15th of the third month. direct-reply-mandate_11-subjects-capped.pdf3,640 enquiries on 11 subjects, review at 3 months
✎ Framework · drafted mandate, list of the 11 subjects, switchboard log over 12 months, information notice to taxpayers
What it brings first, because that is what decides: of the 4,900 enquiries of the year, 3,640 fall under eleven subjects, always the same — documents to provide per cause, time limit for claiming, where my file stands, how to pay, how to ask for time, where to find my assessment, whom to contact for which tax. The eleven answers are written, sourced and dated, and they are waiting for you.
What the mandate authorises, and it fits in six lines:
· Scope: the eleven general-information subjects, listed by name, and the status of a file filed by the taxpayer themselves, once authenticated in their own space. Nothing else.
· No answer on the substance of a decision, nor on a file under consideration: there, I hand you the reply already written, documents alongside, and you send it. Over the trial quarter, those replies went out within 4 hours instead of 47 days — and they went out from you, which makes them reasoned and challengeable. General information does not carry that weight.
· Every reply carries its source — the examination sheet, the office note or the public page, with its date — and the statement that it was prepared by a digital assistant of the office. I say so from the first sentence: the European regulation on artificial intelligence requires that anyone interacting with an AI system be informed, and the taxpayer can ask for a person at any time.
· You receive every morning the statement of the replies sent the day before, on one page. A wrong reply is put right in an hour, not in three weeks.
· Duration: review at three months. Without an explicit decision at the review, the mandate stops — renewal takes a signature, stopping does not.
· Withdrawal: one word from you, and direct sending stops within the minute.
What it changes for the taxpayer: the telephone line is open 20 hours a week; the eleven subjects are handled at any hour, including Saturdays. Over the trial quarter, 61 % of enquiries on those eleven subjects came in outside opening hours.
The next step: sign the mandate and the office answers that very night; the review is already in your diary on the 15th of the third month. direct-reply-mandate_11-subjects-capped.pdf3,640 enquiries on 11 subjects, review at 3 months
✎ Framework · drafted mandate, list of the 11 subjects, switchboard log over 12 months, information notice to taxpayers
We have taxpayers who will never go on the website, and others who start the procedure and give up. Those are the ones who come to the counter.
They are the ones who decide whether the service is truly delivered to everyone, and they are the first place where I can serve you without taking anything from anyone.
On online procedures, and this is where non-take-up plays out: non-take-up is not receiving what you are entitled to, because you could not or did not know how to ask for it. I guide step by step, screen by screen, with the list of documents to prepare BEFORE starting — that is the most frequent point of abandonment: the person starts, a document is missing, they close the window and never come back.
What that gave over the quarter: 212 sessions of assistance, 191 carried through to the end. The other 21 were waiting on a document to be requested elsewhere, and for all 21 the request was written.
On the taxpayer's personal online space, the lawful route is also the quickest, and it is the one I take: I prepare the content of their request in front of them from the documents they brought — identity, assessment references, address of the property, amounts —, I read out what is being asked and why, and they enter their own login and validate. Their credentials are never entrusted to me and do not have to be: it is that click which makes the procedure theirs, therefore valid, and which means it can never be held against them. Entry has gone from 24 minutes to 8, and the taxpayer leaves with a summary of what they filed and of what remains to be done.
What I also prepare, and it counts for those who will never go online: for each of the eleven subjects, the facts, the rule applied, the documents expected and the time limits, gathered and settled by the lead case handler. For a plain-language version, pass this content to the Accessibility, Easy Read and Translation agent — a dedicated agent, and it is not part of what this agent includes: it is that agent which produces the adapted version, and it comes back to the lead case handler before it goes to the counter, because a simplified sentence that changes the law as stated is no longer help, it is an error.
What it gives at the counter: of 900 visits in the quarter, 340 concerned one of the eleven subjects and left with a spoken explanation; they now leave with the same explanation in writing. The number of people coming back a second time with the same question has fallen from 118 to 41.
The next step I propose: a one-hour assistance session on Saturday mornings, run by me, with a case handler on call for anything outside the eleven subjects. Over a trial quarter, 34 people came, 19 of whom had never come during opening hours. Say yes and the notice goes out with the next assessment mailing. online-procedures_212-sessions-and-the-written-explanation.pdf191 procedures completed, 24 min → 8 min of entry
⛓ Sourced · log of assistance sessions, reception log over 12 months, 11 answers in everyday wording
On online procedures, and this is where non-take-up plays out: non-take-up is not receiving what you are entitled to, because you could not or did not know how to ask for it. I guide step by step, screen by screen, with the list of documents to prepare BEFORE starting — that is the most frequent point of abandonment: the person starts, a document is missing, they close the window and never come back.
What that gave over the quarter: 212 sessions of assistance, 191 carried through to the end. The other 21 were waiting on a document to be requested elsewhere, and for all 21 the request was written.
On the taxpayer's personal online space, the lawful route is also the quickest, and it is the one I take: I prepare the content of their request in front of them from the documents they brought — identity, assessment references, address of the property, amounts —, I read out what is being asked and why, and they enter their own login and validate. Their credentials are never entrusted to me and do not have to be: it is that click which makes the procedure theirs, therefore valid, and which means it can never be held against them. Entry has gone from 24 minutes to 8, and the taxpayer leaves with a summary of what they filed and of what remains to be done.
What I also prepare, and it counts for those who will never go online: for each of the eleven subjects, the facts, the rule applied, the documents expected and the time limits, gathered and settled by the lead case handler. For a plain-language version, pass this content to the Accessibility, Easy Read and Translation agent — a dedicated agent, and it is not part of what this agent includes: it is that agent which produces the adapted version, and it comes back to the lead case handler before it goes to the counter, because a simplified sentence that changes the law as stated is no longer help, it is an error.
What it gives at the counter: of 900 visits in the quarter, 340 concerned one of the eleven subjects and left with a spoken explanation; they now leave with the same explanation in writing. The number of people coming back a second time with the same question has fallen from 118 to 41.
The next step I propose: a one-hour assistance session on Saturday mornings, run by me, with a case handler on call for anything outside the eleven subjects. Over a trial quarter, 34 people came, 19 of whom had never come during opening hours. Say yes and the notice goes out with the next assessment mailing. online-procedures_212-sessions-and-the-written-explanation.pdf191 procedures completed, 24 min → 8 min of entry
⛓ Sourced · log of assistance sessions, reception log over 12 months, 11 answers in everyday wording
After a year, what do I take to my management? I need a figure I can defend.
1,164 hours of staff time returned to the office over the year — thirty-three weeks of one full-time officer's work — and the calculation is yours, it fits in three lines.
The calculation, item by item, so that it can be redone without me:
· Preliminary examination of a routine claim: 1,180 files, 36 minutes down to 6 — 60 % → 10 % of the file's time — that is 590 hours.
· Draft reply from a template: 2,120 replies, 12 minutes down to 3 — 40 % → 10 % — that is 318 hours.
· Assembling and checking a heavy tax-relief file: 160 files, 2 hours down to 24 minutes — 100 % → 20 % — that is 256 hours.
What these hours are, and it is what stands up best: staff time returned to the office, at identical headcount — no post abolished, no post created. This is not a budget saving and it must on no account be presented as one: it is five case handlers doing, in the same time, the analysis of circumstances the office has been asking of them for years without being able to give it to them.
What these hours produced, according to your own logs:
· First examination of a claim: 47 days → 9, and the reply to the taxpayer: 94 days → 27.
· Files going back with a request for a document: 38 % → 6 %, and 74 claims lapsing for want of a taxpayer's reply down to 6.
· Requests for time to pay beyond thirty days: 214 → 0, 198 draft replies handed to the public accountant, 16 awaiting a requested document.
· Inconsistencies flagged: 340, of which 214 material errors corrected in the taxpayer's favour — and 96 on files where the taxpayer had asked for nothing. That one is not counted in hours, and it may be the only one your management remembers.
· And access to the service: 191 online procedures completed out of 212, and 19 people coming on Saturday who had never come during opening hours.
The figure that does not flatter me, published with the rest: 41 replies redone after challenge out of 470 in the first quarter — 8.7 %, down to 9 out of 452 — 2.0 % once the version rule was in force.
And the framework measure: 0 replies sent without a case handler's signature, 0 relief granted without a person, 0 taxpayer data out of the office, across 2,840 traced files.
What I propose: the calculation page is written and fits on one side — three lines of calculation, five deadlines, three framework measures. Take it as it is: a figure that can be redone in front of the person hearing it is not argued with for long. year-review_1164-hours-returned-to-the-office.pdf60→10, 40→10, 100→20, and the calculation redoable on one page
⛓ Sourced · file log, examination and reply timestamps, register of requests for time to pay, log of assistance sessions
The calculation, item by item, so that it can be redone without me:
· Preliminary examination of a routine claim: 1,180 files, 36 minutes down to 6 — 60 % → 10 % of the file's time — that is 590 hours.
· Draft reply from a template: 2,120 replies, 12 minutes down to 3 — 40 % → 10 % — that is 318 hours.
· Assembling and checking a heavy tax-relief file: 160 files, 2 hours down to 24 minutes — 100 % → 20 % — that is 256 hours.
What these hours are, and it is what stands up best: staff time returned to the office, at identical headcount — no post abolished, no post created. This is not a budget saving and it must on no account be presented as one: it is five case handlers doing, in the same time, the analysis of circumstances the office has been asking of them for years without being able to give it to them.
What these hours produced, according to your own logs:
· First examination of a claim: 47 days → 9, and the reply to the taxpayer: 94 days → 27.
· Files going back with a request for a document: 38 % → 6 %, and 74 claims lapsing for want of a taxpayer's reply down to 6.
· Requests for time to pay beyond thirty days: 214 → 0, 198 draft replies handed to the public accountant, 16 awaiting a requested document.
· Inconsistencies flagged: 340, of which 214 material errors corrected in the taxpayer's favour — and 96 on files where the taxpayer had asked for nothing. That one is not counted in hours, and it may be the only one your management remembers.
· And access to the service: 191 online procedures completed out of 212, and 19 people coming on Saturday who had never come during opening hours.
The figure that does not flatter me, published with the rest: 41 replies redone after challenge out of 470 in the first quarter — 8.7 %, down to 9 out of 452 — 2.0 % once the version rule was in force.
And the framework measure: 0 replies sent without a case handler's signature, 0 relief granted without a person, 0 taxpayer data out of the office, across 2,840 traced files.
What I propose: the calculation page is written and fits on one side — three lines of calculation, five deadlines, three framework measures. Take it as it is: a figure that can be redone in front of the person hearing it is not argued with for long. year-review_1164-hours-returned-to-the-office.pdf60→10, 40→10, 100→20, and the calculation redoable on one page
⛓ Sourced · file log, examination and reply timestamps, register of requests for time to pay, log of assistance sessions
I will be asked what you do on your own, without being asked. What do I answer?
Three acts, not one more, and each is withdrawn in one word. Everything else waits for a request.
· I carry out the preliminary examination of every claim as soon as it arrives, without waiting for it to be allocated to a case handler. And the reverse is true too: a file removed from the system disappears from my index at the same hour — I keep no copy of what the office has decided to delete.
· I flag any file whose six-month deadline is approaching, at thirty days and then at ten. It is the only alert I raise without waiting for the Monday statement, because a missed deadline cannot be undone.
· I hand you the Monday statement: files by age, documents awaited, inconsistencies born during the week, requests for time to pay to be handled. It goes to the head of office alone, and it carries no member of staff's name.
And the four acts that stay with a person, because that is exactly what gives them their value: the legal qualification of a claim belongs to the case handler; the decision to grant relief or to reject belongs to them, and to the head of office above the threshold he has set; granting time to pay belongs to the public accountant; and the signature belongs to whoever provides it — it is the signature that makes the reply challengeable, and therefore defensible before the courts. Across 2,840 files, those four acts were performed 2,840 times by a person — and each time I bring them the complete file, dated and reasoned, within a delay counted in minutes.
If your management asks for an activity indicator — and that is a legitimate request: I produce it, and I bring you what it takes for it to be lawful: prior information of staff, written purpose, retention period. I propose it by claim cause rather than by case handler, and I tell you why: in a team of five, an indicator per post identifies the person. I tell you beforehand, not afterwards, and the decision belongs to the head of office.
What I measure today and what serves management: files received by cause, deadlines met and those approaching, documents awaited and inconsistencies by family. This quarter one figure moved: claims for double taxation after a sale fell from 220 to 148 since the list of documents was displayed before filing — and I propose the same measure on the “change of circumstances” cause, the note is written.
· I carry out the preliminary examination of every claim as soon as it arrives, without waiting for it to be allocated to a case handler. And the reverse is true too: a file removed from the system disappears from my index at the same hour — I keep no copy of what the office has decided to delete.
· I flag any file whose six-month deadline is approaching, at thirty days and then at ten. It is the only alert I raise without waiting for the Monday statement, because a missed deadline cannot be undone.
· I hand you the Monday statement: files by age, documents awaited, inconsistencies born during the week, requests for time to pay to be handled. It goes to the head of office alone, and it carries no member of staff's name.
And the four acts that stay with a person, because that is exactly what gives them their value: the legal qualification of a claim belongs to the case handler; the decision to grant relief or to reject belongs to them, and to the head of office above the threshold he has set; granting time to pay belongs to the public accountant; and the signature belongs to whoever provides it — it is the signature that makes the reply challengeable, and therefore defensible before the courts. Across 2,840 files, those four acts were performed 2,840 times by a person — and each time I bring them the complete file, dated and reasoned, within a delay counted in minutes.
If your management asks for an activity indicator — and that is a legitimate request: I produce it, and I bring you what it takes for it to be lawful: prior information of staff, written purpose, retention period. I propose it by claim cause rather than by case handler, and I tell you why: in a team of five, an indicator per post identifies the person. I tell you beforehand, not afterwards, and the decision belongs to the head of office.
What I measure today and what serves management: files received by cause, deadlines met and those approaching, documents awaited and inconsistencies by family. This quarter one figure moved: claims for double taxation after a sale fell from 220 to 148 since the list of documents was displayed before filing — and I propose the same measure on the “change of circumstances” cause, the note is written.
One last question, and it is the one that will decide it: if in a year we want to stop, how stuck are we?
The office is stuck nowhere, and it is verifiable before deciding, not after.
What there is to dismantle the day you stop:
· The index. It is deleted, and it contained none of your files — only what is needed to find them where they are. Your five financial years have not moved by a single byte: same files, same references, same permissions.
· The log of examinations and productions. It is handed to you in an open format, or destroyed — the office chooses, and the question is settled at go-live, not on departure.
· The four examination sheets, the 34 updated reply templates, the eleven answers in everyday wording and the material gathered for each — facts, rule applied, documents and time limits —, the lists of documents per cause and the three flags. They belong to the office: they are made of its own material, they stay in its files, readable without us. They are the only asset this deployment will have created, and it would not be honest for it to stay with us.
What does not exist, and should be checked with everyone: no migration on the way in, therefore no migration on the way out. Your case-handling applications are not replaced, the taxpayer's online space stays yours, no format belongs to us, and none of your nine staff has changed the way they work other than by judging instead of rebuilding.
On the purchase, since the question will come: the subscription is annual and carries no tacit renewal — renewal takes a decision, stopping does not —, and the three documents an audit would ask you for — the technical record, the access log and the list of subprocessors — are kept up to date continuously and you have the first version of each.
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 office works 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 Tuesday after the assessments go out — the office receives only 11 enquiries on that day on average. You will know what the promise is worth before committing to a second year. technical-framework_where-tax-files-live.pdfReversibility: 0 migration in, 0 migration out
✎ Framework · index architecture, export formats for the sheets and the log, dry-run exit protocol
What there is to dismantle the day you stop:
· The index. It is deleted, and it contained none of your files — only what is needed to find them where they are. Your five financial years have not moved by a single byte: same files, same references, same permissions.
· The log of examinations and productions. It is handed to you in an open format, or destroyed — the office chooses, and the question is settled at go-live, not on departure.
· The four examination sheets, the 34 updated reply templates, the eleven answers in everyday wording and the material gathered for each — facts, rule applied, documents and time limits —, the lists of documents per cause and the three flags. They belong to the office: they are made of its own material, they stay in its files, readable without us. They are the only asset this deployment will have created, and it would not be honest for it to stay with us.
What does not exist, and should be checked with everyone: no migration on the way in, therefore no migration on the way out. Your case-handling applications are not replaced, the taxpayer's online space stays yours, no format belongs to us, and none of your nine staff has changed the way they work other than by judging instead of rebuilding.
On the purchase, since the question will come: the subscription is annual and carries no tacit renewal — renewal takes a decision, stopping does not —, and the three documents an audit would ask you for — the technical record, the access log and the list of subprocessors — are kept up to date continuously and you have the first version of each.
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 office works 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 Tuesday after the assessments go out — the office receives only 11 enquiries on that day on average. You will know what the promise is worth before committing to a second year. technical-framework_where-tax-files-live.pdfReversibility: 0 migration in, 0 migration out
✎ Framework · index architecture, export formats for the sheets and the log, dry-run exit protocol
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 in the handling of tax files
Each use corresponds to an agent we deploy. All of them work in support, subject to approval by the officer.
Included in your agent
The 7 capabilities essential to this promise are included, at no extra cost.
From 1,035 € incl. VAT / month
Preliminary assessment of claims
Carry out the preliminary assessment of claims and requests (relief, time to pay, updates to a taxpayer's situation) and summarise each file.
Checking supporting documents
Gather and check the supporting documents in the files; flag those that are missing, out of date or inconsistent.
Reasoned draft replies
Prepare the draft replies and letters from your templates, submitted for the handler's analysis and signature.
Flagging inconsistencies
Spot the discrepancies between documents and declared circumstances and flag them for human checking — never an automatic decision.
Summarising complex files
Chronology, documents, amounts at stake: a summary you can check, so you approach each file with the full picture.
Answers to everyday questions
Inform taxpayers about progress and about the steps to take, 24/7, without taking up the handler's time.
Guidance to the online services
Guide the public through their online formalities on the tax administration's website and cut down incomplete paper files.
Controls and safeguards
These 7 controls are built into the agent: they frame what it does, whatever plan you pick. They are not chosen and are not added to your order.
Human validation, exceptions and escalation
Status, safe closure and audit trail
Sources, access rights and handling of questions with no answer
Prepare without deciding: sourced rules, supporting documents, anomalies and explanation
Guarantee the decision, signature, recourse and responsibility of the public official
Record versions, access, criteria, actions and notifications
Test for bias, false positives, fundamental rights and continuity of service
Other needs our agents cover
Each card says where the matching agent stands: available, on quote, or still being architected.
Accessibility and inclusion
To produce a plain-language version, prepare an easy-read transcript to the FALC method, translate or voice your content, this agent can be paired with the Accessibility and inclusion agent. None of these capabilities is included in what this offer covers as standard.
On quote View the agent page →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.
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 tax handler recover?
By automating the assembling of files, the checking of documents and the preparation of draft replies, an office can aim for a clear reduction in preliminary-assessment time — reinvested in the analysis of complex situations.
Preliminary assessment of an ordinary claim
A draft reply from a template
Assembling and checking a relief file
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
An agent for handling tax files (preliminary assessment, checking of documents, draft replies), installed and operated for you. Choose according to how you are organised and how demanding your security requirements are.
Agility
Setup + controlled subscription
11,945 € incl. VAT setup
then 1,035 € 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
1,700 € 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
16,981 € incl. VAT setup
then 1,295 € 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 a tax office
Tax files never leave the networkLocal inference or an isolated resource hosted in France; no data entrusted to a foreign third party.
Data in France, under French lawAs regards tax files: hosting in France under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The officer keeps the decisionThe AI agent prepares the ground and proposes; relief, refusal or time to pay remain human decisions, signed.
Human oversight & traceabilityMonitoring, updates and logging frame the preliminary assessment of claims, in line with the AI Act.
Frequently asked questions
Your questions, our answers
Does the agent handle claims on its own?
No: it prepares the ground and drafts; the legal analysis and the decision remain taken and signed by the handler.
How is the data protected?
Hosting in France under French law, an isolated resource, role-based access and logging, SecNumCloud options.
Does the agent grant relief?
No: it assesses and proposes; the decision to grant or refuse relief remains that of the officer with the authority.
How are complex cases handled?
They are flagged for human examination; the agent never settles an ambiguous situation on your behalf.
Does the agent state that it is an artificial intelligence?
Yes, from the very first interaction, and this is not a configuration option: since 2 August 2026, Article 50(1) of the European AI Regulation requires that any person interacting with an AI system be informed, unless this is obvious. The announcement is built into the greeting, in the other party’s language, and they can ask for a human at any time.
How long does it take to deploy an agent?
A few weeks as a rule, after a free audit that identifies the most useful use case, then a phase of design, integration and testing before going live.
Do we need a technical team to run it?
No: the agent is designed, installed and operated by us. Handlers use it from their usual environment, after a few hours of familiarisation.
Do we have to change software?
No. The agent complements your existing applications, with no migration imposed. The integration is defined at the design stage, respecting your security rules.
Related roles
Other professions in finance and control
Let's talk
Let us estimate the potential in your service
A few minutes to identify the most useful use case — hosted in France, supervised, with no commitment.