● Public sector — One-stop-shop adviser
AI agent for the one-stop-shop adviser
An assistant that answers the public about the procedures of the main public bodies and prepares the support — the adviser stays the trusted contact. Hosted in France, on a resource isolated for each centre, the public's data never leaves the European Union. The AI agent assists, the adviser decides.
Hosted in France
The public's data protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
One-stop-shop assistant · hosted in France
Someone needs a new vehicle registration document after moving house, and has no account with the national agency.
Path prepared. Steps: create an account with the national agency or sign in through the national identity service, the “change of address” online procedure, documents to gather (registration document, proof of address less than six months old, identity document).
A printable summary is ready to hand over.
⛓ Source · official online procedures + the person's documents
A printable summary is ready to hand over.
⛓ Source · official online procedures + the person's documents
She finds French hard to read. What does she need for her housing benefit claim?
Summary ready: for housing benefit she needs her tenancy agreement, her bank details and the amount of her income. The summary also lists the steps still to take.
Plain-language rewriting, an easy-read transcript or a translation: I hand over to the Accessibility and inclusion agent — those capabilities are not part of this offer.
✎ Action · summary handed over — the adviser approves
Plain-language rewriting, an easy-read transcript or a translation: I hand over to the Accessibility and inclusion agent — those capabilities are not part of this offer.
✎ Action · summary handed over — the adviser 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 adviser decidesThe AI agent assists, the human supports
✦ In brief
In a public service one-stop shop, a Blue Lemon Agent agent assists the adviser with the recurring requests — answers on family benefits, health insurance, pensions, tax, employment services or vehicle registration procedures, step-by-step help with online services, pre-filling of the everyday forms. It runs on local inference or is hosted in France on a resource dedicated to and isolated for each centre: the public's 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 adviser stays the trusted contact; the time won back goes to supporting people who struggle with digital services. Live within a few weeks. Your public-sector staff write to it from Microsoft Teams, Slack or their email, and your users reach it on WhatsApp Business, the website chat or email — with no account to create and nothing to install. Reaching the administration from the tool people already have means less non-take-up of rights and equal access to the service. These connections are included in every plan, at no extra cost, within the number of connections your level includes.
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 one-stop shops — and why they hesitate
An adviser covers the procedures of dozens of public bodies — family benefits, health insurance, pensions, tax, employment services, vehicle registration — without being able to be an expert in each, facing people who often struggle with digital services and highly sensitive personal data.
! The issue
The adviser is caught between people who expect concrete help on very varied procedures and the impossibility of mastering every public body's rules in detail. Yet most consumer AI tools would amount to entrusting identities, income, family situations and people's files to a third party, often hosted outside Europe and subject to the Cloud Act.
✓ Our answer
For supporting the public, 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, every step approved by the person and the adviser: the time saved looking up information is never paid for in lost confidentiality. The aim is not to replace the adviser, but to give them back time for human support.
The decisive point
Protecting the public's data: sovereignty & compliance
A one-stop shop sees its users' most personal files. Here is how the architecture of our agents protects them, centre by centre.
Local inference
The agent can run on a machine at the centre: 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 public's data: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
The public's data: architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
One isolated resource per organisation
No pooling of data: an environment strictly dedicated to your centre, 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 procedure is submitted 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 — particularly relevant as soon as social or health data is processed. 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 bodyVaunoy-sur-Sèvre France Services centre — run by the Val de Sèvre community of municipalities (fictional public body)
- Service
- Inter-municipal France Services centre — helping users with procedures of the partner agencies: family benefits, health insurance, pensions, tax, employment service, vehicle registration
- Headcount
- 4 staff: 2 full-time France Services advisers, 1 half-time adviser, 1 front-desk officer — 3.5 full-time equivalents
- Public served
- 11,400 inhabitants across 14 rural municipalities; one user in three says they are not comfortable with online procedures
- Volume
- 470 visits a month, 62 % of them without an appointment; six agencies account for 84 % of requests; 7 use cases ready to deploy
- Tools in place
- France Services tracking portal, shared diary, service mailbox and six years of in-house guidance sheets — the agent plugs into them, nothing is replaced
- Who decides
- The centre manager sets how the service is organised; the lead adviser approves every summary handed to a user; the community council votes the resources
- Room for improvement
- 43 minutes on average to handle a common vehicle registration procedure; 19 days' wait for a support appointment; 118 cases left incomplete last year for want of a single document
The Vaunoy-sur-Sèvre France Services centre wants to give back to face-to-face support the time that hunting for a procedure takes, and to see just as many users while sending them away with their procedure filed. The agent is hosted in France on an isolated resource dedicated to the centre and plugs into the tracking portal, the shared diary and the guidance sheets: it prepares the complete case, the lead adviser approves, the user signs their procedure and leaves knowing how to do it again. The exchanges below span one quarter, from the first enquiry to the review.
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.
One-stop-shop assistant · hosted in France
I have read your 1,410 visits from the quarter and your six years of guidance sheets. I draw a gap your activity report did not show: 38 % of your visits concern just nine procedures, and they are the ones that take you longest to prepare.
The nine, in order: change of address on a vehicle registration document, housing benefit claim, declaring a change of family circumstances to the family benefits agency, health insurance entitlement certificate, pension record statement, income tax return, monthly employment service update, permit renewal, transport cover for medical care.
Why that gap sets the gain: these nine procedures are not difficult, they are scattered. Your advisers are not looking for a complicated answer, they are looking for the right page at the right agency, several times a day. Your own records show 12 minutes on average for an enquiry of this kind.
What I have already written, and it is waiting for you this morning: the nine sheets are done — the competent agency named, the documents required, the stated processing times, the date on which I checked the rule — and the matching form comes pre-filled with what the user gives at the desk. An enquiry goes from 12 minutes to 3: across 210 enquiries a month, 31 hours returned to the service.
The strength of a dated sheet, and it is an argument to use with your funder: every answer carries the agency with authority and the date it was checked. A dated answer can be verified, challenged and corrected in one morning, for all three advisers at once — that is what protects a centre, and no answer given from memory allows it.
What I propose: the lead adviser reviews the nine sheets, I apply them at the front desk from tomorrow, and I bring you the next nine on Friday — they cover a further 21 % of visits, 59 % of the total. Each sheet is corrected in one word: this is your own material, dated and set straight. quarterly-overview_1410-visits.pdf9 procedures account for 38 % of visits
⛓ Sourced · 1,410 visits over the quarter, the centre's guidance sheets, agencies' official pages
The nine, in order: change of address on a vehicle registration document, housing benefit claim, declaring a change of family circumstances to the family benefits agency, health insurance entitlement certificate, pension record statement, income tax return, monthly employment service update, permit renewal, transport cover for medical care.
Why that gap sets the gain: these nine procedures are not difficult, they are scattered. Your advisers are not looking for a complicated answer, they are looking for the right page at the right agency, several times a day. Your own records show 12 minutes on average for an enquiry of this kind.
What I have already written, and it is waiting for you this morning: the nine sheets are done — the competent agency named, the documents required, the stated processing times, the date on which I checked the rule — and the matching form comes pre-filled with what the user gives at the desk. An enquiry goes from 12 minutes to 3: across 210 enquiries a month, 31 hours returned to the service.
The strength of a dated sheet, and it is an argument to use with your funder: every answer carries the agency with authority and the date it was checked. A dated answer can be verified, challenged and corrected in one morning, for all three advisers at once — that is what protects a centre, and no answer given from memory allows it.
What I propose: the lead adviser reviews the nine sheets, I apply them at the front desk from tomorrow, and I bring you the next nine on Friday — they cover a further 21 % of visits, 59 % of the total. Each sheet is corrected in one word: this is your own material, dated and set straight. quarterly-overview_1410-visits.pdf9 procedures account for 38 % of visits
⛓ Sourced · 1,410 visits over the quarter, the centre's guidance sheets, agencies' official pages
A user comes in: she has just moved, she needs to update her vehicle registration document and she has no account on the registration agency's site. What do we do?
The path is ready, the form is pre-filled on screen and her three documents are checked. She leaves with her application filed today, not with a list of things to do.
What I have already done while she waited:
· I opened the exact online procedure at the right agency — an online procedure is one carried out entirely online, with no paper form and nothing to post. Finding it takes twelve minutes when you go looking; it is on screen.
· I checked her three documents: current registration document, proof of address, identity document. Her proof of address is four months old: it is valid. Without that check she would have gone home for a utility bill — and those return trips are what makes up your 118 incomplete cases from last year.
· I pre-filled the form with what she presented at the desk, every field carrying the document it comes from.
· I spared her creating an account: she already holds a national digital identity — the scheme that lets you use credentials you already hold with one government service, tax for instance, to sign in elsewhere without a new password. One user in two has one without knowing it, and I check before having anyone create an account.
The one act that stays hers, and it is a strength: she signs in to her own account and validates. Access to a personal account belongs to its holder, and it is that click which makes the application hers — enforceable against the administration, and open to challenge by her if the answer is slow. Everything before that click is already done.
The time this shifts: handling a procedure of this kind drops from 60 % to 10 % of the time it took — 43 minutes become 7. Those 43 minutes are preparation and search time, not time spent with her: that does not fall, it is precisely where we put it back.
What I propose: I print the summary with the tracking number as soon as it is filed, and I book her a follow-up appointment in three weeks — the same act brought your incomplete cases down from 29 to 9 a quarter. The adviser decides whether to offer it; it takes one click. user-path_change-of-address.pdfForm pre-filled, 3 documents checked, one page to hand over
⛓ Sourced · the agency's official online procedure, documents presented at the desk, form pre-filled on screen
What I have already done while she waited:
· I opened the exact online procedure at the right agency — an online procedure is one carried out entirely online, with no paper form and nothing to post. Finding it takes twelve minutes when you go looking; it is on screen.
· I checked her three documents: current registration document, proof of address, identity document. Her proof of address is four months old: it is valid. Without that check she would have gone home for a utility bill — and those return trips are what makes up your 118 incomplete cases from last year.
· I pre-filled the form with what she presented at the desk, every field carrying the document it comes from.
· I spared her creating an account: she already holds a national digital identity — the scheme that lets you use credentials you already hold with one government service, tax for instance, to sign in elsewhere without a new password. One user in two has one without knowing it, and I check before having anyone create an account.
The one act that stays hers, and it is a strength: she signs in to her own account and validates. Access to a personal account belongs to its holder, and it is that click which makes the application hers — enforceable against the administration, and open to challenge by her if the answer is slow. Everything before that click is already done.
The time this shifts: handling a procedure of this kind drops from 60 % to 10 % of the time it took — 43 minutes become 7. Those 43 minutes are preparation and search time, not time spent with her: that does not fall, it is precisely where we put it back.
What I propose: I print the summary with the tracking number as soon as it is filed, and I book her a follow-up appointment in three weeks — the same act brought your incomplete cases down from 29 to 9 a quarter. The adviser decides whether to offer it; it takes one click. user-path_change-of-address.pdfForm pre-filled, 3 documents checked, one page to hand over
⛓ Sourced · the agency's official online procedure, documents presented at the desk, form pre-filled on screen
A lot of people leave without knowing they were entitled to something else. We do not have time to ask everyone everything.
Over the quarter, 41 situations where a possible entitlement was not being claimed — and for each of them the claim is already written and pre-filled, ready to be signed the day the adviser asks the question.
Non-take-up means not claiming an entitlement you might be eligible for: because you did not know, because the procedure discouraged you, or because nobody asked the right question. It is the first blind spot of a fast front desk, and the heaviest figure in your activity report.
How I spot them, and it is mechanical: the details a user already gives for the procedure in hand — household composition, employment situation, age, tenant status — are exactly the ones that open other entitlements. Nobody has to ask again, they are already on the table.
The three most frequent families at your centre:
· 17 situations where housing benefit looked possible and was not being claimed, spotted during an entirely different procedure.
· 13 situations where the user came about their pension record without having claimed supplementary health cover.
· 11 situations where a change of family circumstances had been declared to no agency at all — and an undeclared change always ends in an overpayment to be repaid.
Who opens the entitlement, and why it is your best argument: the decision belongs to the agency, and I hand it over fully prepared. I write the question the adviser puts to the user, I attach the pre-filled claim and the list of documents, and the agency decides — by a reasoned, dated decision the user can challenge if they disagree. That challengeability is what makes an entitlement hold; an entitlement opened without a reasoned decision would not survive three weeks of appeal.
The figure that does not flatter me: of the 41 flags, 29 led to a question asked by the adviser and only 12 to a procedure actually started. 12 out of 41. The cause is measured and it is not the spotting: a bare flag cost twelve extra minutes per visit, and those twelve minutes are not there every day.
What I did with it, and I replayed it over your quarter: the flag no longer travels alone, it travels with the claim already filled in and the documents gathered — the adviser's part falls from 12 minutes to 2. On the family of undeclared changes of circumstances, procedures started rise from 4 out of 11 to 11 out of 11, without a minute more per visit.
What I propose: start with that family, because that is where the user quietly loses money, and extend to the other two whenever the manager says so. The 41 claims are written: they are waiting for a signature, not for development work. non-take-up_41-situations-flagged.pdf41 claims pre-filled, 12 adviser minutes down to 2
⛓ Sourced · details gathered at the desk, follow-up by the advisers, flagging log
Non-take-up means not claiming an entitlement you might be eligible for: because you did not know, because the procedure discouraged you, or because nobody asked the right question. It is the first blind spot of a fast front desk, and the heaviest figure in your activity report.
How I spot them, and it is mechanical: the details a user already gives for the procedure in hand — household composition, employment situation, age, tenant status — are exactly the ones that open other entitlements. Nobody has to ask again, they are already on the table.
The three most frequent families at your centre:
· 17 situations where housing benefit looked possible and was not being claimed, spotted during an entirely different procedure.
· 13 situations where the user came about their pension record without having claimed supplementary health cover.
· 11 situations where a change of family circumstances had been declared to no agency at all — and an undeclared change always ends in an overpayment to be repaid.
Who opens the entitlement, and why it is your best argument: the decision belongs to the agency, and I hand it over fully prepared. I write the question the adviser puts to the user, I attach the pre-filled claim and the list of documents, and the agency decides — by a reasoned, dated decision the user can challenge if they disagree. That challengeability is what makes an entitlement hold; an entitlement opened without a reasoned decision would not survive three weeks of appeal.
The figure that does not flatter me: of the 41 flags, 29 led to a question asked by the adviser and only 12 to a procedure actually started. 12 out of 41. The cause is measured and it is not the spotting: a bare flag cost twelve extra minutes per visit, and those twelve minutes are not there every day.
What I did with it, and I replayed it over your quarter: the flag no longer travels alone, it travels with the claim already filled in and the documents gathered — the adviser's part falls from 12 minutes to 2. On the family of undeclared changes of circumstances, procedures started rise from 4 out of 11 to 11 out of 11, without a minute more per visit.
What I propose: start with that family, because that is where the user quietly loses money, and extend to the other two whenever the manager says so. The 41 claims are written: they are waiting for a signature, not for development work. non-take-up_41-situations-flagged.pdf41 claims pre-filled, 12 adviser minutes down to 2
⛓ Sourced · details gathered at the desk, follow-up by the advisers, flagging log
Concretely, during a support session, what does the agent do and what does the adviser do?
I prepare before, prompt during, summarise after — and the procedure goes off the same day. The adviser supports, the user validates: the three roles interlock, they do not replace one another.
Before the session — that is where your time went: I identify the exact procedure, the competent agency, the documents required, the stated processing times, and the three traps that make this particular procedure fail. On a change of address for a vehicle registration document: a proof of address that is too old, a missing transfer code when the vehicle has just been bought, and a used name that does not match the civil status record. The form is pre-filled before the user sits down, every field carrying the document it comes from.
During the session: I display the current step and what is expected in the open field. The adviser explains, the user validates. When an error message appears, I say what it means in everyday wording and what has to be corrected — that is where support sessions used to get stuck, on a message nobody understood.
After: the summary printed, with the tracking number, the filing date, what remains to be done and when to follow up. And I track the case until the agency replies, without being asked.
The act I leave to the user, and it is worth it: signing in to their own account and validating. Access to a personal account belongs to its holder — and rightly so: an application validated by the user belongs to them, it is enforceable and it can be challenged. Everything before that click is already done when they sit down.
The benchmark: handling a common procedure drops from 60 % to 10 % — 43 minutes become 7 minutes of preparation. Across your 140 monthly support sessions, 84 hours returned to the service every month.
What I propose: start with the three procedures where your advisers spend the most time, and measure for a month before extending. The manager picks the three; the other six are ready already and are waiting for one word. step-by-step-support_and-known-traps.pdf3 traps named, 43 min → 7 min of preparation
✎ Framework · support session outline, the centre's time records, the three traps identified on the procedures followed
Before the session — that is where your time went: I identify the exact procedure, the competent agency, the documents required, the stated processing times, and the three traps that make this particular procedure fail. On a change of address for a vehicle registration document: a proof of address that is too old, a missing transfer code when the vehicle has just been bought, and a used name that does not match the civil status record. The form is pre-filled before the user sits down, every field carrying the document it comes from.
During the session: I display the current step and what is expected in the open field. The adviser explains, the user validates. When an error message appears, I say what it means in everyday wording and what has to be corrected — that is where support sessions used to get stuck, on a message nobody understood.
After: the summary printed, with the tracking number, the filing date, what remains to be done and when to follow up. And I track the case until the agency replies, without being asked.
The act I leave to the user, and it is worth it: signing in to their own account and validating. Access to a personal account belongs to its holder — and rightly so: an application validated by the user belongs to them, it is enforceable and it can be challenged. Everything before that click is already done when they sit down.
The benchmark: handling a common procedure drops from 60 % to 10 % — 43 minutes become 7 minutes of preparation. Across your 140 monthly support sessions, 84 hours returned to the service every month.
What I propose: start with the three procedures where your advisers spend the most time, and measure for a month before extending. The manager picks the three; the other six are ready already and are waiting for one word. step-by-step-support_and-known-traps.pdf3 traps named, 43 min → 7 min of preparation
✎ Framework · support session outline, the centre's time records, the three traps identified on the procedures followed
And if a user flatly asks us to do the procedure for them? That happens every day.
Then we do it, and 61 users out of 67 left with the procedure filed the same day. Three paths, each with its written act, from simplest to most formal.
· Path 1 — the user validates, the adviser guides, I have prepared everything. Four cases in five, and it is the one that makes them autonomous. 48 procedures filed this quarter by that path.
· Path 2 — a mandated relative. Many users come with a child, a neighbour, a carer. I draft the mandate — the written act by which one person authorises another to act for them, on a named procedure and until a set date — the user signs it, and the relative acts without ambiguity. Five minutes, 9 procedures this quarter.
· Path 3 — the public official keys in before the user's eyes, with their written mandate. Where the user cannot type — tremor, illiteracy, eyesight —, the adviser keys in with a dated mandate limited to that procedure and filed with the case. I draft that mandate in thirty seconds: the written act protects the public official as much as the user. 4 procedures this quarter.
What I do in all three cases: documents checked, form pre-filled, mandate drafted, summary printed, case tracked to the reply. The only thing that changes from one path to the next is the hand that validates — and there is always a written act saying which.
The 6 who did not leave with a filing: they were missing a document. All 6 had an appointment booked before they left and a named list of what to bring back. None left without a solution.
What I propose: that the manager approves a single mandate template for paths 2 and 3. Today each adviser drafts their own, and two of them carry no end date — exactly the detail that turns against a service on appeal. The template is written, dated, limited to one procedure, revocable in one word: it is waiting for a signature. who-keys-in-who-signs-who-approves.pdf3 paths, 61 procedures filed the same day
✎ Framework · the centre's practices, outcome of the 67 requests over the quarter
· Path 1 — the user validates, the adviser guides, I have prepared everything. Four cases in five, and it is the one that makes them autonomous. 48 procedures filed this quarter by that path.
· Path 2 — a mandated relative. Many users come with a child, a neighbour, a carer. I draft the mandate — the written act by which one person authorises another to act for them, on a named procedure and until a set date — the user signs it, and the relative acts without ambiguity. Five minutes, 9 procedures this quarter.
· Path 3 — the public official keys in before the user's eyes, with their written mandate. Where the user cannot type — tremor, illiteracy, eyesight —, the adviser keys in with a dated mandate limited to that procedure and filed with the case. I draft that mandate in thirty seconds: the written act protects the public official as much as the user. 4 procedures this quarter.
What I do in all three cases: documents checked, form pre-filled, mandate drafted, summary printed, case tracked to the reply. The only thing that changes from one path to the next is the hand that validates — and there is always a written act saying which.
The 6 who did not leave with a filing: they were missing a document. All 6 had an appointment booked before they left and a named list of what to bring back. None left without a solution.
What I propose: that the manager approves a single mandate template for paths 2 and 3. Today each adviser drafts their own, and two of them carry no end date — exactly the detail that turns against a service on appeal. The template is written, dated, limited to one procedure, revocable in one word: it is waiting for a signature. who-keys-in-who-signs-who-approves.pdf3 paths, 61 procedures filed the same day
✎ Framework · the centre's practices, outcome of the 67 requests over the quarter
All of this means you read our users' documents. Where exactly do they go?
They do not leave your centre. I run on an isolated resource hosted in France, dedicated to your centre alone — and your documents are copied nowhere.
The two routes, and you choose:
· Local inference — the model computes on a machine installed on your premises: the content of a tax notice crosses no external network to be read.
· An isolated resource hosted in France, under French law, dedicated to your centre — no pooling with any other France Services centre, nor with any other client.
What that changes for the people you receive:
· No user data outside the European Union, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity — a guarantee the large American hosts do not offer, even when installed in Europe.
· Nothing you entrust to me trains a model. What I read from a case serves that case.
· Encryption in transit and at rest, role-based access — rights follow the job: the front-desk officer sees the diary and the guidance sheets, the adviser sees the documents handed in, and every opening is logged.
· Complete logging: who opened what, when, and what was approved. That log is what lets you answer in one page a user who asks what you hold about them.
The retention period, your data protection officer's first question: documents handed in to prepare a procedure are deleted as soon as the summary is issued, and at the latest within 30 days. A France Services centre has no reason to build a permanent file on its users — you are not the agency, you are the support.
The figure that makes this concrete: your community council wrote hosting in France into its last two IT contracts, and the question comes back at every renewal of an agreement with your six partner agencies, which carry 84 % of your requests. You answer it in writing, in one page, without reservation.
What I propose: that I keep up to date the information notice displayed in the waiting room — what is processed, by whom, for how long, and how to ask for deletion — re-dated at every change. Your community council has its answer before the question is put. technical-framework_where-user-documents-live.pdfHosted in France, deleted within 30 days, 0 transfer outside the EU in the target architecture
✎ Framework · deployment architecture, retention periods, access log
The two routes, and you choose:
· Local inference — the model computes on a machine installed on your premises: the content of a tax notice crosses no external network to be read.
· An isolated resource hosted in France, under French law, dedicated to your centre — no pooling with any other France Services centre, nor with any other client.
What that changes for the people you receive:
· No user data outside the European Union, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity — a guarantee the large American hosts do not offer, even when installed in Europe.
· Nothing you entrust to me trains a model. What I read from a case serves that case.
· Encryption in transit and at rest, role-based access — rights follow the job: the front-desk officer sees the diary and the guidance sheets, the adviser sees the documents handed in, and every opening is logged.
· Complete logging: who opened what, when, and what was approved. That log is what lets you answer in one page a user who asks what you hold about them.
The retention period, your data protection officer's first question: documents handed in to prepare a procedure are deleted as soon as the summary is issued, and at the latest within 30 days. A France Services centre has no reason to build a permanent file on its users — you are not the agency, you are the support.
The figure that makes this concrete: your community council wrote hosting in France into its last two IT contracts, and the question comes back at every renewal of an agreement with your six partner agencies, which carry 84 % of your requests. You answer it in writing, in one page, without reservation.
What I propose: that I keep up to date the information notice displayed in the waiting room — what is processed, by whom, for how long, and how to ask for deletion — re-dated at every change. Your community council has its answer before the question is put. technical-framework_where-user-documents-live.pdfHosted in France, deleted within 30 days, 0 transfer outside the EU in the target architecture
✎ Framework · deployment architecture, retention periods, access log
A user arrives with a folder: tax notice, tenancy agreement, payslips, letters. He no longer knows what is for what.
The folder is sorted in four minutes: nine documents recognised, two to re-photograph, one missing — and the housing benefit form is already filled in on screen when he sits down.
What I read, document by document: last year's tax notice, the tenancy agreement signed in March, three payslips, an entitlement certificate, a letter from the family benefits agency, bank details, an electricity bill from November.
What I draw from it: for his housing benefit claim, seven of these documents are needed, two are irrelevant — the entitlement certificate and the letter — and what is missing is the declaration of his employment situation since his contract changed in January. Without it the case would have come back in six weeks: he obtains it online during the session, I have opened the page.
The two payslips photographed at an angle, and here is what I do with them: the taxable net is cut off. I have them re-scanned at the front desk — forty seconds — and I re-read them straight away. The figure entered on the form comes from the payslip, never from a calculation on the other two: a reconstructed figure is indistinguishable from a read one, and it is the one that would get the case rejected six weeks later. Over the quarter, 46 documents were re-scanned this way and 46 were re-read within the minute.
The pre-filling: the form carries what I actually read, and every field shows the document it comes from — the adviser checks a line at a glance. Fields left empty are highlighted and carry the document that will fill them, with where the user obtains it.
The time this shifts: building a case spanning several agencies drops from 100 % to 20 % of the time it took — 2 h 30 becomes 30 minutes. Across your 24 cases of this kind a month, 48 hours returned to the service.
What I propose: that the lead adviser reviews the form before it is handed over — she approves, always, and it is that check which makes the case defensible — and that the user leaves with the claim filed and a dated list of what remains to be supplied. reading-the-documents_9-recognised-2-unreadable.pdf9 documents read in 4 minutes, every field carries its own
⛓ Sourced · documents presented by the user, agency form, the centre's guidance sheets
What I read, document by document: last year's tax notice, the tenancy agreement signed in March, three payslips, an entitlement certificate, a letter from the family benefits agency, bank details, an electricity bill from November.
What I draw from it: for his housing benefit claim, seven of these documents are needed, two are irrelevant — the entitlement certificate and the letter — and what is missing is the declaration of his employment situation since his contract changed in January. Without it the case would have come back in six weeks: he obtains it online during the session, I have opened the page.
The two payslips photographed at an angle, and here is what I do with them: the taxable net is cut off. I have them re-scanned at the front desk — forty seconds — and I re-read them straight away. The figure entered on the form comes from the payslip, never from a calculation on the other two: a reconstructed figure is indistinguishable from a read one, and it is the one that would get the case rejected six weeks later. Over the quarter, 46 documents were re-scanned this way and 46 were re-read within the minute.
The pre-filling: the form carries what I actually read, and every field shows the document it comes from — the adviser checks a line at a glance. Fields left empty are highlighted and carry the document that will fill them, with where the user obtains it.
The time this shifts: building a case spanning several agencies drops from 100 % to 20 % of the time it took — 2 h 30 becomes 30 minutes. Across your 24 cases of this kind a month, 48 hours returned to the service.
What I propose: that the lead adviser reviews the form before it is handed over — she approves, always, and it is that check which makes the case defensible — and that the user leaves with the claim filed and a dated list of what remains to be supplied. reading-the-documents_9-recognised-2-unreadable.pdf9 documents read in 4 minutes, every field carries its own
⛓ Sourced · documents presented by the user, agency form, the centre's guidance sheets
Our users often leave having forgotten half of what we told them.
They leave with one page, written to be re-read alone at home — and your incomplete cases have gone from 29 to 9 a quarter.
What the summary carries, and nothing else:
· What was done today, with the date and the tracking number where there is one.
· What is left to do, in order, one action per line, in short sentences.
· The documents to bring back, named as they are named on the document itself — « your 2026 tax notice », not « proof of income », because nobody recognises a proof of income in their folder.
· The deadline, and what happens if it passes: the case is not lost, it is re-examined, and saying so prevents the anxiety that makes people give up.
· Who to contact, with the centre's name, opening hours and phone number.
What it changed, measured over the quarter: 29 cases left incomplete in the previous quarter, 9 in this one. The figure that does not flatter me: paper alone was not enough for those 9. Their cause lies elsewhere: a document to be obtained from a third party — a former employer, a landlord, a notary — whose timescale is neither yours nor mine.
What I did with it, and it is already running: for those 9 cases I wrote the request to be sent to the third party, ready for the user to sign, and I flag to the adviser the one whose deadline is approaching, five days ahead. Over the last six weeks, 4 of the 9 obtained their document through that letter.
The figure over a full year: your 118 incomplete cases from last year would be 36 at this rate. These are not cases, they are people coming back a second and a third time for the same thing — and they are your Tuesday morning queues.
What I propose: that the summary also goes out by message to those who ask for it, in addition to the paper copy, never instead of it. Some of your users do not open email: paper remains the reference, and that is a choice about equality, not about convenience. summary-handed-to-the-user.pdfOne page, 5 sections, 29 incomplete cases → 9
⛓ Sourced · summaries issued over the quarter, cases tracked through to complete filing, causes recorded
What the summary carries, and nothing else:
· What was done today, with the date and the tracking number where there is one.
· What is left to do, in order, one action per line, in short sentences.
· The documents to bring back, named as they are named on the document itself — « your 2026 tax notice », not « proof of income », because nobody recognises a proof of income in their folder.
· The deadline, and what happens if it passes: the case is not lost, it is re-examined, and saying so prevents the anxiety that makes people give up.
· Who to contact, with the centre's name, opening hours and phone number.
What it changed, measured over the quarter: 29 cases left incomplete in the previous quarter, 9 in this one. The figure that does not flatter me: paper alone was not enough for those 9. Their cause lies elsewhere: a document to be obtained from a third party — a former employer, a landlord, a notary — whose timescale is neither yours nor mine.
What I did with it, and it is already running: for those 9 cases I wrote the request to be sent to the third party, ready for the user to sign, and I flag to the adviser the one whose deadline is approaching, five days ahead. Over the last six weeks, 4 of the 9 obtained their document through that letter.
The figure over a full year: your 118 incomplete cases from last year would be 36 at this rate. These are not cases, they are people coming back a second and a third time for the same thing — and they are your Tuesday morning queues.
What I propose: that the summary also goes out by message to those who ask for it, in addition to the paper copy, never instead of it. Some of your users do not open email: paper remains the reference, and that is a choice about equality, not about convenience. summary-handed-to-the-user.pdfOne page, 5 sections, 29 incomplete cases → 9
⛓ Sourced · summaries issued over the quarter, cases tracked through to complete filing, causes recorded
That leaves the cases sitting at the agency. We have no time to chase them, and users call us.
The 34 chasers are written and ready to go — 34 cases beyond the agency's stated processing time. What is missing is a signature: a message sent in your centre's name commits your centre. Here is the mandate.
What the count shows, and it was nowhere: 34 cases exceed the stated processing time, 11 of them by more than two months. Three agencies account for 26 of the 34 — this is not general slowness, it is three specific channels, and that is information worth having for your steering committee.
The mandate, exactly as written:
· Scope: chasing cases followed by the centre that exceed the agency's stated processing time. Nothing else.
· Cap: 30 chasers a week, so as not to swamp a counterpart who does answer.
· Message: a single template approved by the manager, factual: the reference, the filing date, the stated time, a request for a progress update. No formal notice, no threatening phrasing — your counterparts are public service colleagues.
· Visibility: everything is copied to the manager, who sees what goes out the same day.
· Duration: until 31/03/2027, renewed by a fresh written act.
· Withdrawal: one word from the manager and it all stops, with no reason to be given.
And the second strand, the one that touches users: calling back the 9 users whose case is incomplete and whose deadline is approaching. Same logic: approved template, 30 a week at most, copied to the manager, stopped in one word.
What it is worth in time: done by hand, these chasers amounted to 4 hours a week nobody had — so they were not done. This is not time saved, it is work that did not exist and now will.
The decision stays with the manager: you sign, I start on Monday with the 26 cases at the three slowest agencies, and I hand you the first statement on Friday. mandate_agency-chasers-and-user-reminders.pdf34 cases, 30 chasers a week, revocable in one word
⛓ Sourced · the centre's tracking portal, processing times stated by the agencies, cases awaiting one document
What the count shows, and it was nowhere: 34 cases exceed the stated processing time, 11 of them by more than two months. Three agencies account for 26 of the 34 — this is not general slowness, it is three specific channels, and that is information worth having for your steering committee.
The mandate, exactly as written:
· Scope: chasing cases followed by the centre that exceed the agency's stated processing time. Nothing else.
· Cap: 30 chasers a week, so as not to swamp a counterpart who does answer.
· Message: a single template approved by the manager, factual: the reference, the filing date, the stated time, a request for a progress update. No formal notice, no threatening phrasing — your counterparts are public service colleagues.
· Visibility: everything is copied to the manager, who sees what goes out the same day.
· Duration: until 31/03/2027, renewed by a fresh written act.
· Withdrawal: one word from the manager and it all stops, with no reason to be given.
And the second strand, the one that touches users: calling back the 9 users whose case is incomplete and whose deadline is approaching. Same logic: approved template, 30 a week at most, copied to the manager, stopped in one word.
What it is worth in time: done by hand, these chasers amounted to 4 hours a week nobody had — so they were not done. This is not time saved, it is work that did not exist and now will.
The decision stays with the manager: you sign, I start on Monday with the 26 cases at the three slowest agencies, and I hand you the first statement on Friday. mandate_agency-chasers-and-user-reminders.pdf34 cases, 30 chasers a week, revocable in one word
⛓ Sourced · the centre's tracking portal, processing times stated by the agencies, cases awaiting one document
One user reads French with difficulty. She comes about housing benefit and leaves every time without having understood.
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: what I do here is spot the sheet that blocks, gather the material that agent needs, and hand it to you ready to go.
What I spotted on housing benefit, and it is measured: three expressions make the session fail every time — « proof of income », « family quotient », « sworn statement ». All three are in your current sheet, and of the 34 sessions this quarter where the user left without having understood, 29 came up against one of them.
What I hand over for each term, and that is the real work: the agency's original wording, its dated source, and the constraint that forbids distorting it. That is what counts — it is what stands before the agency, and it is what keeps her a remedy if the agency disputes what she understood. A rewording served on its own would be more dangerous than a difficult sentence: it would lose the exact word an appeal rests on.
The list I hand you: 60 administrative terms taken from your nine main procedures, each with its source sheet, its dated source and the number of sessions it made fail. It is an order ready to go, not an adapted text — and the list is yours: it is corrected in one word and it holds for every one of your sheets.
The figure that does not flatter me, and it matters: of 34 lists handed over this quarter, 5 had to be redone by the adviser. The cause is real and it is not technical: I quoted the original wording without its update date — an undated wording stands against nothing, and nobody can work from it.
What I did with it: every term now carries the date of the day I checked the rule with the agency, and I ran it back over the quarter's 34 lists: the 5 redos would have fallen to 1. Your advisers checked it, it is in service, and the following month's measurement does give one redo in 34.
What the second visit costs you, in figures: all 34 of those sessions produced a second visit to the desk — about 7 hours over the quarter, and that is the figure to set against the dedicated agent's quotation.
What I propose: that your 18 most requested guidance sheets go to the Accessibility, Easy Read and Translation agent — a dedicated agent, and it is not part of what this agent includes —, each with its material already gathered, its dated source and the count of what it costs today. All 18 orders are written: what is missing is a decision on your side, not preparation work. terms-to-adapt_60-terms-and-their-source.pdfMaterial handed to the dedicated agent · 5 redos in 34 down to 1
⛓ Sourced · the centre's guidance sheets, rewrites made by the advisers, approved glossary
What I spotted on housing benefit, and it is measured: three expressions make the session fail every time — « proof of income », « family quotient », « sworn statement ». All three are in your current sheet, and of the 34 sessions this quarter where the user left without having understood, 29 came up against one of them.
What I hand over for each term, and that is the real work: the agency's original wording, its dated source, and the constraint that forbids distorting it. That is what counts — it is what stands before the agency, and it is what keeps her a remedy if the agency disputes what she understood. A rewording served on its own would be more dangerous than a difficult sentence: it would lose the exact word an appeal rests on.
The list I hand you: 60 administrative terms taken from your nine main procedures, each with its source sheet, its dated source and the number of sessions it made fail. It is an order ready to go, not an adapted text — and the list is yours: it is corrected in one word and it holds for every one of your sheets.
The figure that does not flatter me, and it matters: of 34 lists handed over this quarter, 5 had to be redone by the adviser. The cause is real and it is not technical: I quoted the original wording without its update date — an undated wording stands against nothing, and nobody can work from it.
What I did with it: every term now carries the date of the day I checked the rule with the agency, and I ran it back over the quarter's 34 lists: the 5 redos would have fallen to 1. Your advisers checked it, it is in service, and the following month's measurement does give one redo in 34.
What the second visit costs you, in figures: all 34 of those sessions produced a second visit to the desk — about 7 hours over the quarter, and that is the figure to set against the dedicated agent's quotation.
What I propose: that your 18 most requested guidance sheets go to the Accessibility, Easy Read and Translation agent — a dedicated agent, and it is not part of what this agent includes —, each with its material already gathered, its dated source and the count of what it costs today. All 18 orders are written: what is missing is a decision on your side, not preparation work. terms-to-adapt_60-terms-and-their-source.pdfMaterial handed to the dedicated agent · 5 redos in 34 down to 1
⛓ Sourced · the centre's guidance sheets, rewrites made by the advisers, approved glossary
Does she know she is talking to a machine? We want no ambiguity about that.
She knows from the first sentence, in one short sentence, and the disclosure is built into the welcome message — nobody can switch it off.
What is announced, at every interaction: that she is dealing with an artificial intelligence system, that a human adviser is available and that she may ask for one at any time, without having to justify herself. Over the quarter, 23 users asked, and all 23 were seen by an adviser the same day.
The basis, and it is dated: since 2 August 2026, Article 50(1) of the European regulation on artificial intelligence requires that any person interacting with an AI system be informed, except where it is obvious. At your centre it is not: the user walks into a public service and speaks to someone.
What that spares you, very concretely: the question asked in the community council — « are you replacing staff with a machine without saying so? » The answer is written, dated, displayed in the waiting room, and I have drafted it in French. For the plain-language version that goes with it, pass this content to the Accessibility, Easy Read and Translation agent — a dedicated agent, and it is not part of what this agent includes.
And accessibility, which goes with it: what I produce follows the accessibility rules applying to public digital services — they require in particular that text be readable by a screen reader, that contrast be sufficient, and that nothing essential rests on colour alone. Concretely: summaries print at 12 point minimum, structured by headings, with no decorative tables, and a large-print version prints in one click. You print about thirty a month across 470 visits, and the layout had to be redone by hand every time.
What I propose: have the disclosure and the display notice reviewed by your community council at its next sitting, then display them in the waiting room, in French. Both texts are written and fit on one page. A centre that announces its own use of AI will never face the charge of having hidden it. ai-disclosure-accessibility-and-human-recourse.pdfAnnounced at every interaction, a human available at any time
✎ Framework · welcome message, accessibility rules for public digital services, display notice
What is announced, at every interaction: that she is dealing with an artificial intelligence system, that a human adviser is available and that she may ask for one at any time, without having to justify herself. Over the quarter, 23 users asked, and all 23 were seen by an adviser the same day.
The basis, and it is dated: since 2 August 2026, Article 50(1) of the European regulation on artificial intelligence requires that any person interacting with an AI system be informed, except where it is obvious. At your centre it is not: the user walks into a public service and speaks to someone.
What that spares you, very concretely: the question asked in the community council — « are you replacing staff with a machine without saying so? » The answer is written, dated, displayed in the waiting room, and I have drafted it in French. For the plain-language version that goes with it, pass this content to the Accessibility, Easy Read and Translation agent — a dedicated agent, and it is not part of what this agent includes.
And accessibility, which goes with it: what I produce follows the accessibility rules applying to public digital services — they require in particular that text be readable by a screen reader, that contrast be sufficient, and that nothing essential rests on colour alone. Concretely: summaries print at 12 point minimum, structured by headings, with no decorative tables, and a large-print version prints in one click. You print about thirty a month across 470 visits, and the layout had to be redone by hand every time.
What I propose: have the disclosure and the display notice reviewed by your community council at its next sitting, then display them in the waiting room, in French. Both texts are written and fit on one page. A centre that announces its own use of AI will never face the charge of having hidden it. ai-disclosure-accessibility-and-human-recourse.pdfAnnounced at every interaction, a human available at any time
✎ Framework · welcome message, accessibility rules for public digital services, display notice
And how can we be sure two users in the same situation get the same answer?
Because the answer comes from the same dated sheet, whichever adviser, whichever day, however busy the desk — and I measure it for you every quarter.
What you had before, and it is nobody's fault: three staff, three ways of explaining the same procedure, and six years of guidance sheets, some of them three reforms out of date. Across the nine most requested procedures I found four sheets giving an obsolete list of documents. A user seen on Monday did not get the same list as a user seen on Thursday.
What changed: one sheet per procedure, dated on the day I checked the rule with the agency, the same for all three advisers. And when a rule changes, the sheet changes for everyone the same morning, with a note of what moved.
The measurement, across your nine procedures: I compared 60 answers given this quarter against the reference sheet. 52 match exactly, 6 differ on a detail without consequence — a rounded processing time, a document called by another name — and 2 genuinely diverged. Both concerned the same procedure and came from a sheet I had not yet revised: it is revised, and both users were called back within the week.
What I measure, and why I measure it on answers rather than on people: a per-adviser counter would become a target, the target would shorten interviews, and you would lose both the measure and the quality of welcome it claimed to watch over. That is a design choice, not a prohibition: if your authority wants the individual indicator, I write it, I run it over the past quarter and I bring you the three conditions to be met — prior information of staff, consultation of the representative bodies, proportionality of what is measured. The decision is theirs, and I hand it over fully prepared in a week rather than in a committee.
What I propose: a quarterly consistency check on the nine main procedures, handed to the manager and the steering committee, with the 2 divergences and what was done about each. Equal treatment is proved with figures or it is not proved at all — and that is exactly what a funder wants to see.
What you had before, and it is nobody's fault: three staff, three ways of explaining the same procedure, and six years of guidance sheets, some of them three reforms out of date. Across the nine most requested procedures I found four sheets giving an obsolete list of documents. A user seen on Monday did not get the same list as a user seen on Thursday.
What changed: one sheet per procedure, dated on the day I checked the rule with the agency, the same for all three advisers. And when a rule changes, the sheet changes for everyone the same morning, with a note of what moved.
The measurement, across your nine procedures: I compared 60 answers given this quarter against the reference sheet. 52 match exactly, 6 differ on a detail without consequence — a rounded processing time, a document called by another name — and 2 genuinely diverged. Both concerned the same procedure and came from a sheet I had not yet revised: it is revised, and both users were called back within the week.
What I measure, and why I measure it on answers rather than on people: a per-adviser counter would become a target, the target would shorten interviews, and you would lose both the measure and the quality of welcome it claimed to watch over. That is a design choice, not a prohibition: if your authority wants the individual indicator, I write it, I run it over the past quarter and I bring you the three conditions to be met — prior information of staff, consultation of the representative bodies, proportionality of what is measured. The decision is theirs, and I hand it over fully prepared in a week rather than in a committee.
What I propose: a quarterly consistency check on the nine main procedures, handed to the manager and the steering committee, with the 2 divergences and what was done about each. Equal treatment is proved with figures or it is not proved at all — and that is exactly what a funder wants to see.
Our real problem is the wait: 19 days for a support appointment. People give up.
Six days at the latest count, and without a single extra slot in the diary. This is not added capacity, it is triage at the door.
What I found reading your incoming requests: you receive about 130 appointment requests a month, by phone, by message and at the desk. 47 of them did not need an appointment — an enquiry answered it, or a document could be dropped off at the desk without waiting. They were taking up a third of your support slots.
What I now do when a request arrives:
· I triage: which procedure, which agency, which documents the user already has, and whether there is a deadline.
· I answer within the hour the requests that an enquiry settles — with the dated sheet, the list of documents and the form already pre-filled, which saves the journey.
· I offer three slots where support really is needed, reserving the long slots for multi-agency cases, the only ones that need them.
· I escalate to the adviser requests where a deadline is approaching: they go to the front of the queue, and that is a service decision, not a calculation.
The result over the quarter: average wait cut from 19 days to 6 · missed appointments down from 14 % to 5 %, because a user who received the list of documents turns up with the documents · and 47 journeys avoided a month by people who sometimes drive 25 kilometres to see you. Across fourteen rural municipalities, an avoided journey is not a comfort detail.
What I guarantee at the door, and it is what holds the six days: every user who asks to be seen gets their slot, even where an enquiry would do. The triage is on slots, never on people: over the quarter, 0 appointments refused and 0 cancelled.
What I propose: opening two weekly slots reserved for cases spotted as non-take-up, the ones where the user was asking for nothing. That is the one group who will never book an appointment on their own, and your 41 pre-filled claims are waiting for exactly those two slots. incoming-requests-and-appointments.pdf130 requests, 47 settled without a journey, 19 d → 6 d
⛓ Sourced · incoming requests over the quarter, shared diary, record of appointments attended
What I found reading your incoming requests: you receive about 130 appointment requests a month, by phone, by message and at the desk. 47 of them did not need an appointment — an enquiry answered it, or a document could be dropped off at the desk without waiting. They were taking up a third of your support slots.
What I now do when a request arrives:
· I triage: which procedure, which agency, which documents the user already has, and whether there is a deadline.
· I answer within the hour the requests that an enquiry settles — with the dated sheet, the list of documents and the form already pre-filled, which saves the journey.
· I offer three slots where support really is needed, reserving the long slots for multi-agency cases, the only ones that need them.
· I escalate to the adviser requests where a deadline is approaching: they go to the front of the queue, and that is a service decision, not a calculation.
The result over the quarter: average wait cut from 19 days to 6 · missed appointments down from 14 % to 5 %, because a user who received the list of documents turns up with the documents · and 47 journeys avoided a month by people who sometimes drive 25 kilometres to see you. Across fourteen rural municipalities, an avoided journey is not a comfort detail.
What I guarantee at the door, and it is what holds the six days: every user who asks to be seen gets their slot, even where an enquiry would do. The triage is on slots, never on people: over the quarter, 0 appointments refused and 0 cancelled.
What I propose: opening two weekly slots reserved for cases spotted as non-take-up, the ones where the user was asking for nothing. That is the one group who will never book an appointment on their own, and your 41 pre-filled claims are waiting for exactly those two slots. incoming-requests-and-appointments.pdf130 requests, 47 settled without a journey, 19 d → 6 d
⛓ Sourced · incoming requests over the quarter, shared diary, record of appointments attended
What do you do without asking us? We want the exact list, to give to the elected members.
Three actions, not one more, and each undoes as fast as it does. Here they are, worded as they will appear in your report to the elected members.
· I acknowledge an incoming request within the hour, with the centre's name, its opening hours and what the person should bring if they come. And the reverse holds: if they ask not to be contacted by message any more, that applies immediately and to everything — except paper, which remains the reference channel.
· I date and file every summary issued, so the adviser can find in ten seconds what was said to a user three weeks ago. A summary is corrected and reissued in one word.
· I delete the documents handed in as soon as the summary is issued, and at the latest within 30 days. That one does not undo, and deliberately so: a deletion you can reverse is not a deletion.
Everything else waits for a named decision: handing a summary to a user, sending a chaser to an agency, calling a user back, opening a procedure, amending a guidance sheet. Over the quarter: 1,410 visits, 0 decisions taken without human validation — the measure is in the log, not in a promise.
Why these three: each protects the relationship with the user rather than an indicator. Acknowledging within the hour stops an anxious person calling three times — your chasing calls have fallen by 41 % since. Filing avoids making someone repeat what they have already explained, and being made to repeat yourself is what most reliably discourages people who are uneasy with digital services. Deleting stops a support service turning into a database.
What I propose: that this list appears in the notice displayed in the waiting room and in the report to the community council — I have drafted it, one page, in French. Three actions is a number an elected member remembers, and an elected member who knows what the agent does alone does not ask for it to be stopped. I propose reopening this list every quarter: the day a fourth action is justified, it goes through you, not through a setting.
· I acknowledge an incoming request within the hour, with the centre's name, its opening hours and what the person should bring if they come. And the reverse holds: if they ask not to be contacted by message any more, that applies immediately and to everything — except paper, which remains the reference channel.
· I date and file every summary issued, so the adviser can find in ten seconds what was said to a user three weeks ago. A summary is corrected and reissued in one word.
· I delete the documents handed in as soon as the summary is issued, and at the latest within 30 days. That one does not undo, and deliberately so: a deletion you can reverse is not a deletion.
Everything else waits for a named decision: handing a summary to a user, sending a chaser to an agency, calling a user back, opening a procedure, amending a guidance sheet. Over the quarter: 1,410 visits, 0 decisions taken without human validation — the measure is in the log, not in a promise.
Why these three: each protects the relationship with the user rather than an indicator. Acknowledging within the hour stops an anxious person calling three times — your chasing calls have fallen by 41 % since. Filing avoids making someone repeat what they have already explained, and being made to repeat yourself is what most reliably discourages people who are uneasy with digital services. Deleting stops a support service turning into a database.
What I propose: that this list appears in the notice displayed in the waiting room and in the report to the community council — I have drafted it, one page, in French. Three actions is a number an elected member remembers, and an elected member who knows what the agent does alone does not ask for it to be stopped. I propose reopening this list every quarter: the day a fourth action is justified, it goes through you, not through a setting.
One quarter on, what has it returned to the service, very concretely?
163 hours a month — the equivalent of one full-time member of staff. And let me say at once what that figure is not: it is not a post to be cut, and no post has been created either.
Where the 163 hours come from, item by item, on your own records:
· Multi-agency enquiries: 12 minutes → 3, across 210 a month — 31 hours.
· Support on a common procedure: 43 minutes → 7, i.e. 60 % → 10 %, across 140 a month — 84 hours.
· Building a multi-agency case: 2 h 30 → 30 minutes, i.e. 100 % → 20 %, across 24 a month — 48 hours.
What those hours became, and it is the only question that matters: +38 % of time spent on long one-to-one support — the people who will never complete a procedure alone —, two outreach sessions opened in the most remote municipalities, and the appointment wait cut from 19 days to 6. That is a support post you could not afford to create, not a post to hand back.
The other measured effects: 61 procedures filed the same day out of 67 requests · 29 incomplete cases a quarter down to 9 · 41 non-take-up situations fully prepared, the adviser's part cut from 12 minutes to 2 · 47 journeys avoided a month · one redo in 34 lists handed over, instead of five.
And the figure that does not flatter me, published without being asked: of 1,410 visits, 87 produced an answer the adviser corrected before handing it over — 6 %. The cause is identified: 61 of those 87 concerned two agencies whose rules changed during the quarter, and my sheets followed a few days behind. What I did with it: those two agencies are now re-checked every Monday morning rather than as things come up. Over the last month, corrections fell back to 2 %. An answer corrected by an adviser is not a failure — it is exactly the role she has been given.
What I propose for the coming quarter: present this review to the community council in hours returned, never in hours saved — the file is written, ten pages, every figure pointing back to its record. I add the one expense that really moves: 47 journeys avoided a month, about 1,100 kilometres your users no longer drive. A public service does not sell productivity, it gives staff time back to those who need it — and that is the sentence your elected members will remember. quarterly-review_time-returned-to-support.pdf163 h/month, 6 % corrections down to 2 %
⛓ Sourced · the centre's time records, tracking portal, log of corrections made by the advisers
Where the 163 hours come from, item by item, on your own records:
· Multi-agency enquiries: 12 minutes → 3, across 210 a month — 31 hours.
· Support on a common procedure: 43 minutes → 7, i.e. 60 % → 10 %, across 140 a month — 84 hours.
· Building a multi-agency case: 2 h 30 → 30 minutes, i.e. 100 % → 20 %, across 24 a month — 48 hours.
What those hours became, and it is the only question that matters: +38 % of time spent on long one-to-one support — the people who will never complete a procedure alone —, two outreach sessions opened in the most remote municipalities, and the appointment wait cut from 19 days to 6. That is a support post you could not afford to create, not a post to hand back.
The other measured effects: 61 procedures filed the same day out of 67 requests · 29 incomplete cases a quarter down to 9 · 41 non-take-up situations fully prepared, the adviser's part cut from 12 minutes to 2 · 47 journeys avoided a month · one redo in 34 lists handed over, instead of five.
And the figure that does not flatter me, published without being asked: of 1,410 visits, 87 produced an answer the adviser corrected before handing it over — 6 %. The cause is identified: 61 of those 87 concerned two agencies whose rules changed during the quarter, and my sheets followed a few days behind. What I did with it: those two agencies are now re-checked every Monday morning rather than as things come up. Over the last month, corrections fell back to 2 %. An answer corrected by an adviser is not a failure — it is exactly the role she has been given.
What I propose for the coming quarter: present this review to the community council in hours returned, never in hours saved — the file is written, ten pages, every figure pointing back to its record. I add the one expense that really moves: 47 journeys avoided a month, about 1,100 kilometres your users no longer drive. A public service does not sell productivity, it gives staff time back to those who need it — and that is the sentence your elected members will remember. quarterly-review_time-returned-to-support.pdf163 h/month, 6 % corrections down to 2 %
⛓ Sourced · the centre's time records, tracking portal, log of corrections made by the advisers
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 a one-stop shop
Each use corresponds to an agent we deploy. All work in support, subject to the adviser's approval.
Included in your agent
The 6 capabilities essential to this promise are included, at no extra cost.
From 765 € incl. VAT / month
Answers across public bodies
Answer questions on family benefits, health insurance, pensions, tax, employment services or vehicle registration, and direct people.
Step-by-step help with procedures
Prepare the support for online procedures, step by step, and tackle unclaimed entitlements.
Pre-filling the forms
Pre-fill the everyday forms from the person's documents, under the adviser's control.
Guidance through the online services
Guide the person step by step through the official portals, never acting in their place.
Summary of documents and steps
Generate a clear summary of the documents gathered and the steps remaining, to hand to the person. The reminder of the documents still needed can go out by text message.
Appointment booking
Qualify incoming requests and schedule support appointments at the centre.
Controls and safeguards
These 5 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
Offer accessibility, easy read, multichannel access and genuine human recourse
Measure resolution, response time, non-take-up, drop-offs and satisfaction
Other needs our agents cover
Each card says where the matching agent stands: available, on quote, or still being architected.
Multichannel administrative information
RAG on procedures; web, phone and front desk; 24/7; multilingual. Sourced answers.
On quote View the agent page →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 one-stop shop win back?
By automating the search for information, the step-by-step explanations and the summaries, a centre can aim for a clear reduction in time spent on repetitive requests — reinvested in human support.
Supporting an everyday vehicle registration procedure
Explaining a procedure step by step
Assembling a file across several public bodies
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 one-stop-shop agent (answers across public bodies, help with procedures, form pre-filling), installed and operated for you. Choose according to how you are organised — available by direct award below the public procurement thresholds.
Agility
Setup + controlled subscription
7,430 € incl. VAT setup
then 765 € 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
1,180 € 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
11,621 € incl. VAT setup
then 1,005 € incl. VAT/month · + hardware from 2,989 € (one-off purchase, in addition) — a sovereign computer installed on your premises, maintained remotely. Models run locally, your data returned at the end of the contract. 36-month commitment.
- Hardware installed on your premises (you own it)
- French / European AI models run locally
- Secure remote maintenance (Pro support included)
Not included in the packages: AI consumption (model tokens), re-invoiced at real cost with no margin, and tracked in real time in your client area. Maintenance and supervision subscription for an initial term of 12 months for the Agility package, 24 months for the Serenity package and 36 months for the 100% Sovereign package, renewable; support levels (SLA 72 h / 24 h / 4 h) optional. Bespoke development, additional integrations or exceptional volumes are quoted separately. Support Monday to Friday, 9am to 6pm. Prices include VAT at 20%: as a public body that is not VAT-registered, you cannot reclaim it.
AI model:
none of the AI models offered currently carries a fixed surcharge. When the selected model carries a cost, that cost is shown when you choose it, before you order, and re-invoiced at the cost incurred, with no mark-up; usage is billed at the publisher's price. Publishers' prices are published in US dollars: the amount re-invoiced is the amount in euros actually borne by Blue Lemon Agent on the publisher's invoice, at that invoice's exchange rate, with no commission or mark-up.
Included components and additional components Components included in the base offer: the Blue Lemon Agent software foundation, the AI models listed in the order journey, the standard channels (Microsoft Teams, Slack, WhatsApp Business, email, website chat, calendars, Microsoft 365 / Google Workspace, file storage, market VoIP telephony, professional social-media pages and accounts, Google Business Profile), hosting in France for the package chosen, backups, supervision, updates and support. If adapting the AI agent to your constraints, your needs or your requests requires other paid components — a third-party publisher's software licence, paid API access to one of your applications, hosting of health data, for which French law requires an HDS-certified host (art. L. 1111-8 of the French Public Health Code), SecNumCloud-qualified hosting, a speech synthesis service, particular hardware —, they are offered to you as an option or on quotation and re-invoiced at the cost incurred; nothing is committed without your written agreement. Where the artificial intelligence model you choose entails an additional cost, that cost is shown to you before you order and re-invoiced to you at the cost incurred, with no margin.
What to expect
Go-live
2 to 3 weeks
Agent designed, channels connected, team trained.
Steady state
4 to 7 weeks
After a few weeks of real use, once the agent's behaviour matches what you expect.
Indicative estimate, adjusted to the options you keep. It is not a delivery commitment.
Our commitment
Four guarantees that matter to a one-stop shop
The public's data never leaves the centreLocal inference or an isolated resource hosted in France; no data entrusted to a foreign third party.
Data in France, under French lawThe public's data: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The adviser keeps the decisionThe AI agent prepares replies, forms and summaries that can be checked; no procedure is submitted automatically.
Human oversight & traceabilityAnswers across public bodies: every change to the service is logged, in line with the requirements of the AI Act.
Frequently asked questions
Your questions, our answers
Does the agent carry out the procedures in the person's place?
No: it guides and prepares; the person and the adviser approve every step. No administrative decision is automated.
Does the public's data leave the centre?
No: local inference or hosting in France, an isolated resource, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
Does the agent fill in the forms for the person?
It pre-fills and explains, but the person approves and signs their own procedures; the adviser keeps the human support.
Does it work for people who do not speak French?
It answers in French and prepares the summary of documents and steps; the adviser stays the trusted contact for the support itself. Plain-language rewriting, easy-read transcripts and translation belong to a separate agent, Accessibility and inclusion, ordered on its own.
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 and handing over to your teams.
Do we need a technical team in-house?
No: the agent is designed, installed and operated for you — integration, supervision and updates included. Your teams concentrate on their work, with guided familiarisation.
Do we have to change software?
No. The agent relies on your templates and complements your existing tools, with no migration imposed. We adapt the integration to your environment.
Which tools can users use to reach the agent?
The ones they already have. The agent answers on WhatsApp Business, the website chat and email: users have no account to create and no application to install. This is a lever for access to the service — WhatsApp and the telephone reach people an online form never does, which reduces the non-take-up of rights and serves equal access. Internally, your public-sector staff talk to the agent from Microsoft Teams, Slack or their email, without switching tools. Oversight runs from a web dashboard. These connectors rely on open standards, including the MCP protocol; they are included in every plan, at no extra cost, within the number of connections your level includes. Only the fees charged by the platforms themselves — WhatsApp Business bills per conversation, and the gateway bills the text messages sent — are passed on at actual cost, with no margin, outside the subscription.
Can the agent notify users by text message?
Yes, as an option — and it is the channel that reaches the people the others never do. A text message arrives on an ordinary phone: no email address, no account, no app, no data plan and no smartphone. Reminding people by text message of a support appointment or of a document still missing means fewer entitlements going unclaimed and more equal access to the service. It is an outbound notification channel, not a conversation channel: users do not write to the agent there — they reach you on WhatsApp Business, the website chat or email. The message announces without detailing — never the reason, never the detail of the situation — and never contains a payment link, a login or a password: users reach the service the way they usually do. Delivery goes through a gateway established in the European Union; the messages sent are passed on at actual cost, with no margin, and the channel itself is an optional module, set out on the pricing page.
Related roles
Professions and solutions related to supporting the public
Let's talk
Let's size up the potential in your centre
A few minutes to identify the most useful use case — hosted in France, supervised, with no commitment.