● Public sector — Employment & integration
The employment adviser's AI agent: prepare the assessment, give time back to casework
Summarising a history, searching for suitable vacancies, writing meeting notes and follow-up reminders take up a large share of advisers' time — at the expense of the relationship, which is what makes casework good. Your AI agent absorbs that repetitive work. Hosted in France — on local inference or an isolated resource — jobseekers' data never leaves the public service. The adviser keeps the relationship and the decision.
Hosted in France
Applicants' data protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Employment & integration assistant · hosted in France
Before my 11am meeting with Mr Diallo, prepare me a summary of his history and suggest vacancies that fit his profile.
Summary ready. History: 8 years in logistics (forklift operator, order picker), forklift licences current, driving licence. Last contract ended 4 months ago, willing to travel 30 km.
3 vacancies matched: 2 in logistics locally, 1 in food processing (with a 3-week bridging course). Points to clarify at the meeting: availability for shift work and any plans to retrain.
⛓ Source · the jobseeker's file + vacancies in the employment database
3 vacancies matched: 2 in logistics locally, 1 in food processing (with a 3-week bridging course). Points to clarify at the meeting: availability for shift work and any plans to retrain.
⛓ Source · the jobseeker's file + vacancies in the employment database
Book me a follow-up in 3 weeks and prepare the note of the meeting.
Note written from your jottings: assessment, barriers identified (travel), 2 agreed actions (update the CV, apply for vacancy no. 2).
Follow-up reminder scheduled for 30 June, with a text message reminder to the jobseeker — subject to your approval.
✎ Action · note and reminder ready — the adviser approves
Follow-up reminder scheduled for 30 June, with a text message reminder to the jobseeker — subject to your approval.
✎ Action · note and reminder ready — 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 agent assists, it never directs alone
✦ In brief
In a public employment and integration service, a Blue Lemon Agent agent prepares the adviser's repetitive work — summarising a history, assessment, matching vacancies to profiles, meeting notes and follow-up reminders — to give time back to supporting people. It runs on local inference or is hosted in France: jobseekers' 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 matches are transparent and checked by humans, to limit any bias and guarantee equal treatment. The agent assists, the public officer decides. Your public-sector staff write to it from Microsoft Teams, Slack or their email, and jobseekers 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
8
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 employment services — and why they hesitate
People expect personal, responsive support, while the administrative load — summaries, meeting notes, reminders — mechanically reduces the time spent in conversation. And the data involved, histories and personal circumstances, is among the most sensitive in the public service.
! The issue
The adviser is caught between jobseekers who need to be listened to and guided, and a follow-up load that keeps growing (assessments, matches, meeting notes, reminders). Yet most consumer AI solutions amount to entrusting jobseekers' histories, personal circumstances, health or social data to a third party, often hosted outside Europe and subject to the Cloud Act.
✓ Our answer
For a public service, 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, transparent matching and decisions reserved to the adviser: the time saved on administration is never paid for in lost confidentiality or in unequal treatment. The aim is not to replace the adviser, but to give them back time for the relationship.
The decisive point
Protecting applicants' data: sovereignty & compliance
An employment service handles sensitive personal circumstances. Here is how the architecture of our agents protects them, file by file.
Local inference
The agent can run on a machine at the service: no jobseeker 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 data: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
Exposure of jobseekers' data 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 the public's data: an environment strictly dedicated to your authority or organisation.
Encryption & controlled access
Encryption in transit and at rest, role-based access, strong authentication and logging of consultations.
AI Act: governed deployment
The agent is strictly in support; no referral is decided automatically; transparent matching, 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 (social, health), SecNumCloud and HDS options are available according to how demanding your requirements are. 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 bodyVaugrelle Employment and Integration Centre — local public employment service
- Sector
- Support towards employment and integration — diagnosis, matching with local vacancies, training referrals, mobility grants, follow-up of pathways
- Headcount
- 78 staff — including 46 employment and integration advisers, 8 on reception, 3 employer liaison officers and one data protection officer
- Public served
- 9,400 jobseekers supported — 204 per adviser on average, 31,000 interviews a year and 52,000 enquiries by phone, message or e-mail
- Order of magnitude
- An interview takes 60 minutes of adviser time, preparation and record included; 12,400 active vacancies in the area and 340 schemes, partners and procedures to know
- Tools in place
- Employment information system, local vacancy database, document management system, e-mail and a messaging service — the agent plugs into them read-only, nothing is replaced and nothing is migrated
- Who decides
- The adviser settles the diagnosis, decides on guidance, refers to training and grants an allowance; the head of the support unit approves campaigns and mandates; the data protection officer keeps the register
- Room for improvement
- 27 minutes of preparation per interview, often done in the five minutes before it; 41 % of interview records written more than 48 hours later; 21 % of pathways interrupted with no recorded reason; and 1,340 jobseekers eligible for a grant they have never claimed
At Vaugrelle, 46 advisers support 9,400 people: 204 each. What makes support good is the time spent with the person — not the time spent piecing their history back together across four screens before receiving them. The data handled is personal circumstances, sometimes social or health-related. The agent runs on local inference on a machine in the service, reads the files without ever writing to them, explains every proposal it makes, and leaves the diagnosis, the guidance and the decision to the adviser. The exchanges that follow span a year, from the review of the files to the results presented to the steering committee.
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.
Employment & integration assistant · hosted in France
I reread your support files over the weekend — 9,400 pathways, 31,000 interviews over twelve months — and I draw three measurements from them that your screens were carrying without anyone having the time to add them up.
The first, and it decides the rest: preparing an interview accounts for 45 % of the time it takes — 27 minutes out of the 60 an interview represents, preparation and record included. And 62 % of those preparations are done in the five minutes before the interview, because there is no other slot. That is not a failure of method: it is a caseload of 204 people per adviser.
The second: 41 % of interview records are written more than 48 hours after the interview — and a record written two days later loses exactly what serves the next one: the person's own words, the barriers named, the commitments made.
The third, and it will interest your steering committee: 21 % of pathways stop with no recorded reason. One jobseeker in five leaves support without the service knowing why, so without anything being fixable.
What the three items you already measure cost:
· Preparing an interview: 45 %, that is 27 minutes.
· Writing an interview record: 30 %, that is 18 minutes.
· Answering a recurring jobseeker question: 10 %, that is 6 minutes.
What I propose, and it is not a promise — it is already in place: the career summary of your 9,400 files is written, one page per person, and it regenerates with every document filed. Preparing an interview drops from 45 % to 15 % — from 27 minutes to 9, and those 9 minutes are spent reading the person, no longer searching for them.
The next step, and it takes half a day: three volunteer advisers work a week with the summaries and tell me what is missing. I correct it, and the whole service switches over the following week. current-picture_9400-pathways-and-three-time-items.pdf27 minutes of preparation, 41 % of late records, 21 % of interrupted pathways
⛓ Sourced · 9,400 support files, 31,000 interviews over 12 months, interview records and exit reasons
The first, and it decides the rest: preparing an interview accounts for 45 % of the time it takes — 27 minutes out of the 60 an interview represents, preparation and record included. And 62 % of those preparations are done in the five minutes before the interview, because there is no other slot. That is not a failure of method: it is a caseload of 204 people per adviser.
The second: 41 % of interview records are written more than 48 hours after the interview — and a record written two days later loses exactly what serves the next one: the person's own words, the barriers named, the commitments made.
The third, and it will interest your steering committee: 21 % of pathways stop with no recorded reason. One jobseeker in five leaves support without the service knowing why, so without anything being fixable.
What the three items you already measure cost:
· Preparing an interview: 45 %, that is 27 minutes.
· Writing an interview record: 30 %, that is 18 minutes.
· Answering a recurring jobseeker question: 10 %, that is 6 minutes.
What I propose, and it is not a promise — it is already in place: the career summary of your 9,400 files is written, one page per person, and it regenerates with every document filed. Preparing an interview drops from 45 % to 15 % — from 27 minutes to 9, and those 9 minutes are spent reading the person, no longer searching for them.
The next step, and it takes half a day: three volunteer advisers work a week with the summaries and tell me what is missing. I correct it, and the whole service switches over the following week. current-picture_9400-pathways-and-three-time-items.pdf27 minutes of preparation, 41 % of late records, 21 % of interrupted pathways
⛓ Sourced · 9,400 support files, 31,000 interviews over 12 months, interview records and exit reasons
What also costs us dearly is finding the right information: which scheme, which partner, which procedure. Nobody knows all 340 by heart.
All 340 are read, up to date and searchable in plain language — a scheme question gets its answer in twenty seconds, with the source and its date.
What I have indexed, and where it comes from: 340 scheme sheets, partner agreements and internal procedures, plus the 118 service notes of the past three years — including those that amend an older sheet, which is the real trap in your files.
One example, taken from your own enquiries: “which grant for a jobseeker over 50, without a driving licence, who lands a job offer 28 km away?” Answer in twenty seconds: two schemes that can be combined, a third excluded by a service note from last spring, the current form, the partner who processes it and the observed lead time — 11 days. The same search used to take 25 minutes and ended two times out of five with a call to a colleague.
What that weighs: your advisers run 1,100 searches of this kind a month. At 25 minutes against 20 seconds, that is 450 hours a month changing destination — and above all, the answer is the same from one adviser to another, which it was not.
What I flag without being asked: 17 of the 340 sheets are contradicted by a more recent service note, and 4 point to a partner who no longer runs the scheme. The 21 updated versions are written, each with the note that governs it and its date — they await a read-through by your unit head, not a drafting job.
The next step I propose: that I reread the collection whenever a service note is issued and write the updated version overnight. Your 340 sheets will stop ageing in silence — and a jobseeker will stop hearing two different answers depending on which adviser is sitting opposite. service-memory_340-sheets-21-updated.pdf25 minutes brought down to 20 seconds, 21 sheets rewritten
⛓ Sourced · 340 scheme and procedure sheets, 118 service notes over 3 years, internal search log
What I have indexed, and where it comes from: 340 scheme sheets, partner agreements and internal procedures, plus the 118 service notes of the past three years — including those that amend an older sheet, which is the real trap in your files.
One example, taken from your own enquiries: “which grant for a jobseeker over 50, without a driving licence, who lands a job offer 28 km away?” Answer in twenty seconds: two schemes that can be combined, a third excluded by a service note from last spring, the current form, the partner who processes it and the observed lead time — 11 days. The same search used to take 25 minutes and ended two times out of five with a call to a colleague.
What that weighs: your advisers run 1,100 searches of this kind a month. At 25 minutes against 20 seconds, that is 450 hours a month changing destination — and above all, the answer is the same from one adviser to another, which it was not.
What I flag without being asked: 17 of the 340 sheets are contradicted by a more recent service note, and 4 point to a partner who no longer runs the scheme. The 21 updated versions are written, each with the note that governs it and its date — they await a read-through by your unit head, not a drafting job.
The next step I propose: that I reread the collection whenever a service note is issued and write the updated version overnight. Your 340 sheets will stop ageing in silence — and a jobseeker will stop hearing two different answers depending on which adviser is sitting opposite. service-memory_340-sheets-21-updated.pdf25 minutes brought down to 20 seconds, 21 sheets rewritten
⛓ Sourced · 340 scheme and procedure sheets, 118 service notes over 3 years, internal search log
All of this assumes you read career histories, family circumstances, sometimes health information. Where does that data go?
It goes nowhere. I run on local inference on a machine in the service, and nothing I read leaves your network.
Local inference means the model computes on your machine: the text of an interview record or a skills assessment crosses no outside network to be processed. If the organisation would rather not host a machine, the other route is an isolated resource hosted in France, dedicated to Vaugrelle — no pooling with another organisation.
What that changes, point by point:
· Your jobseekers' data trains no model, neither ours nor a third party's.
· I work read-only on your files, and the technical account I read through has no right to write — that is verified with one command, not on trust.
· Encryption in transit and at rest, role-based access — rights follow the job: an employer liaison officer sees a profile's skills and availability, never their social circumstances or health information. 6 roles for your 78 staff, and the log shows 0 out-of-role access since go-live.
· Hosting in France, under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
· Logging: who consulted which file, when, and what the system produced.
And the act the law reserves to a person, which is exactly what gives your support its value: a decision that directs someone, opens a training place to them or grants them an allowance has effects on them — it cannot rest on automated processing alone. It is taken by the adviser, reasoned, dated, traced, and the person can challenge it. Everything leading up to it, I have already done: pathway reconstructed, documents read, applicable schemes listed with their conditions, and the criteria of every proposal displayed in plain sight. Guidance nobody had signed could be argued with by nobody — and a jobseeker is entitled to argue with what they are offered.
The figure that sums this up: 0 jobseeker data out of the service's network across the 31,000 interviews of the year, and processing in the EU targeted.
What I propose: that I keep up to date the record your data protection officer and your steering committee will ask for — hosting, data processed, retention periods, who accesses what. It is asked for once a year and takes three days to rebuild; the first version is written and attached. technical-framework_where-jobseeker-data-lives.pdfLocal inference, read-only, processing in the EU targeted
✎ Framework · deployment architecture, technical account rights, access log, first version of the register record
Local inference means the model computes on your machine: the text of an interview record or a skills assessment crosses no outside network to be processed. If the organisation would rather not host a machine, the other route is an isolated resource hosted in France, dedicated to Vaugrelle — no pooling with another organisation.
What that changes, point by point:
· Your jobseekers' data trains no model, neither ours nor a third party's.
· I work read-only on your files, and the technical account I read through has no right to write — that is verified with one command, not on trust.
· Encryption in transit and at rest, role-based access — rights follow the job: an employer liaison officer sees a profile's skills and availability, never their social circumstances or health information. 6 roles for your 78 staff, and the log shows 0 out-of-role access since go-live.
· Hosting in France, under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
· Logging: who consulted which file, when, and what the system produced.
And the act the law reserves to a person, which is exactly what gives your support its value: a decision that directs someone, opens a training place to them or grants them an allowance has effects on them — it cannot rest on automated processing alone. It is taken by the adviser, reasoned, dated, traced, and the person can challenge it. Everything leading up to it, I have already done: pathway reconstructed, documents read, applicable schemes listed with their conditions, and the criteria of every proposal displayed in plain sight. Guidance nobody had signed could be argued with by nobody — and a jobseeker is entitled to argue with what they are offered.
The figure that sums this up: 0 jobseeker data out of the service's network across the 31,000 interviews of the year, and processing in the EU targeted.
What I propose: that I keep up to date the record your data protection officer and your steering committee will ask for — hosting, data processed, retention periods, who accesses what. It is asked for once a year and takes three days to rebuild; the first version is written and attached. technical-framework_where-jobseeker-data-lives.pdfLocal inference, read-only, processing in the EU targeted
✎ Framework · deployment architecture, technical account rights, access log, first version of the register record
I have Mr Sofiane Berthaud at 11. I have never met him, and his file runs to forty documents. What have I got in front of me?
One page, read in nine minutes, and you walk into the interview knowing what you have to ask him rather than what you have to look up.
The pathway, reconstructed from his documents: 8 years in logistics — forklift driver then order picker —, machinery operating certificates valid until next March, category B driving licence. Last contract ended 4 months ago, end of assignment and not resignation — that distinction changes which schemes he is entitled to, and it appeared only on a certificate on page 31.
What the summary carries, and that was nowhere:
· The three previous interviews summed up in four lines each, with the commitments made and what became of them — two kept, one abandoned after three weeks.
· The barriers he named himself in those interviews, quoted: travel beyond 30 km, and shift work “not before my daughter has her nursery place”. Those are his words, not my interpretation.
· What has changed since: the nursery place was obtained in April, the information is in a document filed on his personal account and nobody has seen it. That is the point that reopens shift work, therefore 4 local vacancies.
· Three questions to put to him, which I propose and which you keep or not: when his certificates expire, his real availability for shift work, and his interest in a bridge into food processing.
The time that moves: preparing an interview goes from 45 % to 15 % — from 27 minutes to 9, and above all 100 % of interviews are prepared, against 38 % today. Across 31,000 interviews a year, that is 9,300 adviser hours given back to the relationship.
The next step I propose: that the summary lands in your diary the evening before, with the three questions, and that the jobseeker can read it themselves if they wish — of the 40 people who tried it, 31 corrected or added something, which is better than finding out in the interview. career-summary_mr-berthaud.pdfPathway, quoted barriers, what has changed, 3 proposed questions
⛓ Sourced · 40 documents in the file, 3 previous interview records, documents filed on the personal account, local vacancy database
The pathway, reconstructed from his documents: 8 years in logistics — forklift driver then order picker —, machinery operating certificates valid until next March, category B driving licence. Last contract ended 4 months ago, end of assignment and not resignation — that distinction changes which schemes he is entitled to, and it appeared only on a certificate on page 31.
What the summary carries, and that was nowhere:
· The three previous interviews summed up in four lines each, with the commitments made and what became of them — two kept, one abandoned after three weeks.
· The barriers he named himself in those interviews, quoted: travel beyond 30 km, and shift work “not before my daughter has her nursery place”. Those are his words, not my interpretation.
· What has changed since: the nursery place was obtained in April, the information is in a document filed on his personal account and nobody has seen it. That is the point that reopens shift work, therefore 4 local vacancies.
· Three questions to put to him, which I propose and which you keep or not: when his certificates expire, his real availability for shift work, and his interest in a bridge into food processing.
The time that moves: preparing an interview goes from 45 % to 15 % — from 27 minutes to 9, and above all 100 % of interviews are prepared, against 38 % today. Across 31,000 interviews a year, that is 9,300 adviser hours given back to the relationship.
The next step I propose: that the summary lands in your diary the evening before, with the three questions, and that the jobseeker can read it themselves if they wish — of the 40 people who tried it, 31 corrected or added something, which is better than finding out in the interview. career-summary_mr-berthaud.pdfPathway, quoted barriers, what has changed, 3 proposed questions
⛓ Sourced · 40 documents in the file, 3 previous interview records, documents filed on the personal account, local vacancy database
The nursery place obtained in April — you pull it from a document nobody had opened. How many files are in that situation?
1,870 out of your 9,400 — and I have sorted them by what they change, not by filing date.
What I reread: every document filed by jobseekers on their personal account over twelve months, set against what their support file holds.
· 1,870 files carry a new element that was never picked up: a certificate obtained, childcare solved, a licence passed, a house move, an administrative recognition.
· Of those 1,870, 640 change eligibility for a scheme — that is, the person is entitled today to something they were not entitled to at the last interview.
· And 212 reopen local vacancies that were ruled out by a barrier since lifted.
What I did with it, and it is ready: the 640 files went to their referring adviser, with the new element, the document that establishes it, its date, and the scheme it reopens. Not a list of 640 lines: each adviser received their own, 14 on average, sorted by entitlement deadline.
What that gave in the first month: 391 of the 640 were picked up in an interview or by phone, 212 led to a new action — a match, a training referral, a grant application. The remaining 249 are in their adviser's queue, with their deadline.
What I propose next: that this matching runs every night and not once. A document filed on a Tuesday evening reaches the adviser on Wednesday morning, with what it changes — that is the only way a jobseeker's filing counts on the day they make it, and not six months later. career-summary_mr-berthaud.pdf1,870 files with a new element, 640 eligibility changes
⛓ Sourced · documents filed on personal accounts over 12 months, support files, scheme conditions
What I reread: every document filed by jobseekers on their personal account over twelve months, set against what their support file holds.
· 1,870 files carry a new element that was never picked up: a certificate obtained, childcare solved, a licence passed, a house move, an administrative recognition.
· Of those 1,870, 640 change eligibility for a scheme — that is, the person is entitled today to something they were not entitled to at the last interview.
· And 212 reopen local vacancies that were ruled out by a barrier since lifted.
What I did with it, and it is ready: the 640 files went to their referring adviser, with the new element, the document that establishes it, its date, and the scheme it reopens. Not a list of 640 lines: each adviser received their own, 14 on average, sorted by entitlement deadline.
What that gave in the first month: 391 of the 640 were picked up in an interview or by phone, 212 led to a new action — a match, a training referral, a grant application. The remaining 249 are in their adviser's queue, with their deadline.
What I propose next: that this matching runs every night and not once. A document filed on a Tuesday evening reaches the adviser on Wednesday morning, with what it changes — that is the only way a jobseeker's filing counts on the day they make it, and not six months later. career-summary_mr-berthaud.pdf1,870 files with a new element, 640 eligibility changes
⛓ Sourced · documents filed on personal accounts over 12 months, support files, scheme conditions
And the diagnosis itself? That is the heart of our job, I delegate it to nobody.
The diagnosis stays yours, and I make it quicker to write and sturdier to defend: the elements are gathered, ordered and sourced, you set the analysis.
What I hand over before the interview:
· The skills found in the documents, with the source document — never an inferred skill: either it is written somewhere, or I propose it as a question to ask.
· The barriers already expressed, quoted, dated, with those that have been lifted since.
· The schemes applicable to this situation, each with its conditions ticked or not — and the exact reason when a condition is not met, so that you can tell the person without reopening a rulebook.
· The local labour market in their occupation: 212 active logistics vacancies, 34 % shift work, median pay taken from the published vacancies, and 3 employers who recruit regularly without your advisers ever having met them. That last point interests your employer liaison officers as much as you.
What the diagnosis becomes once you have set it: I write it in the form of your framework, from your dictated or typed words, and the action plan follows with its deadlines. You reread and approve — it is your signature that commits the support, and it is what lets the person argue with what has been decided.
What that changes, measured over the quarter: the diagnosis is set at the first interview in 82 % of cases, against 54 % — because it is no longer postponed for lack of elements. And support starting at the first interview rather than the second is three weeks gained on every pathway.
The next step I propose: that the easy-read version of the diagnosis and action plan go to the person the same day, reviewed by the lead adviser before it is sent. Of the 40 trials, 31 people came back to the second interview with the document annotated — the surest sign of support that is theirs and not only yours. diagnosis-and-action-plan_service-framework.pdf82 % of diagnoses set at the first interview against 54 %
⛓ Sourced · documents in the file, the service's diagnosis framework, scheme conditions, local vacancy database
What I hand over before the interview:
· The skills found in the documents, with the source document — never an inferred skill: either it is written somewhere, or I propose it as a question to ask.
· The barriers already expressed, quoted, dated, with those that have been lifted since.
· The schemes applicable to this situation, each with its conditions ticked or not — and the exact reason when a condition is not met, so that you can tell the person without reopening a rulebook.
· The local labour market in their occupation: 212 active logistics vacancies, 34 % shift work, median pay taken from the published vacancies, and 3 employers who recruit regularly without your advisers ever having met them. That last point interests your employer liaison officers as much as you.
What the diagnosis becomes once you have set it: I write it in the form of your framework, from your dictated or typed words, and the action plan follows with its deadlines. You reread and approve — it is your signature that commits the support, and it is what lets the person argue with what has been decided.
What that changes, measured over the quarter: the diagnosis is set at the first interview in 82 % of cases, against 54 % — because it is no longer postponed for lack of elements. And support starting at the first interview rather than the second is three weeks gained on every pathway.
The next step I propose: that the easy-read version of the diagnosis and action plan go to the person the same day, reviewed by the lead adviser before it is sent. Of the 40 trials, 31 people came back to the second interview with the document annotated — the surest sign of support that is theirs and not only yours. diagnosis-and-action-plan_service-framework.pdf82 % of diagnoses set at the first interview against 54 %
⛓ Sourced · documents in the file, the service's diagnosis framework, scheme conditions, local vacancy database
On vacancies, we have been sold automatic matching before. My advisers no longer believe in it: it throws out proposals that make no sense.
Every proposal arrives with the criteria that produced it, line by line — you judge the proposal on what made it, not on a score.
What I hand over for Mr Berthaud, out of the 12,400 active vacancies in the area:
· 4 logistics vacancies, of which 2 in shift work reopened by the nursery place obtained in April. For each one: the skill required against the skill found, the certificate demanded against his own, the distance calculated in real travel time and the type of contract.
· 1 food-processing vacancy with a three-week bridge funded by a local scheme — the scheme's conditions are ticked one by one in the sheet.
· And 2 vacancies I set aside, saying so: one requires a certificate he does not hold and that training cannot cover in time, the other is 46 km away with no public transport before 6 a.m. Setting aside with an explanation beats setting aside in silence: if you disagree, the proposal comes back with one click.
What appears nowhere in what I hand over: no overall score, no ranking of people. A score hides its criteria; a list of ticked criteria can be discussed with the jobseeker, face to face. That is what makes a proposal explainable, and therefore open to challenge.
What that gives, measured over the quarter: matches leading to a job interview go from 11 % to 19 %, and the number of proposals sent per person falls from 6.4 to 3.1 — fewer proposals, more interviews: exactly what a jobseeker expects from a service that knows them.
The next step I propose: that new vacancies be set against the 9,400 pathways every night, and that each adviser receive in the morning the matches for their caseload, criteria displayed. Over the trial month, 41 vacancies were filled by people the night-time matching had brought to the adviser on the day they were published. job-profile-matching_criteria-displayed.pdf4 vacancies proposed, 2 set aside with their reason, no score
⛓ Sourced · 12,400 active local vacancies, support files, bridge scheme conditions, public transport travel times
What I hand over for Mr Berthaud, out of the 12,400 active vacancies in the area:
· 4 logistics vacancies, of which 2 in shift work reopened by the nursery place obtained in April. For each one: the skill required against the skill found, the certificate demanded against his own, the distance calculated in real travel time and the type of contract.
· 1 food-processing vacancy with a three-week bridge funded by a local scheme — the scheme's conditions are ticked one by one in the sheet.
· And 2 vacancies I set aside, saying so: one requires a certificate he does not hold and that training cannot cover in time, the other is 46 km away with no public transport before 6 a.m. Setting aside with an explanation beats setting aside in silence: if you disagree, the proposal comes back with one click.
What appears nowhere in what I hand over: no overall score, no ranking of people. A score hides its criteria; a list of ticked criteria can be discussed with the jobseeker, face to face. That is what makes a proposal explainable, and therefore open to challenge.
What that gives, measured over the quarter: matches leading to a job interview go from 11 % to 19 %, and the number of proposals sent per person falls from 6.4 to 3.1 — fewer proposals, more interviews: exactly what a jobseeker expects from a service that knows them.
The next step I propose: that new vacancies be set against the 9,400 pathways every night, and that each adviser receive in the morning the matches for their caseload, criteria displayed. Over the trial month, 41 vacancies were filled by people the night-time matching had brought to the adviser on the day they were published. job-profile-matching_criteria-displayed.pdf4 vacancies proposed, 2 set aside with their reason, no score
⛓ Sourced · 12,400 active local vacancies, support files, bridge scheme conditions, public transport travel times
And equal treatment? If your tool proposes less to people from certain neighbourhoods, we are the ones who will answer for it.
You will answer for it with figures, and I produce them every month — the first gap measured has been corrected since April, and I will show you which one.
What I measure, every month, on the proposals sent: the average number of proposals per person, compared between comparable groups — same occupation sought, same qualification level, same length of registration — by municipality of residence, age band and sex. Comparing comparable populations is the only way to see a gap: compared overall, any gap is explained away by the mix of profiles.
What the first measurement showed, and I flagged it before being asked: jobseekers from two neighbourhoods received 1.4 times fewer proposals than identical profiles from other municipalities. The cause was neither the neighbourhood nor the people: it was my mobility filter, set in straight-line kilometres. Those two neighbourhoods are 12 km from a business park, but 1 h 10 away by public transport — and conversely, vacancies 25 km away served by a direct 20-minute line were excluded.
What I did with it, and it is measured: mobility is now calculated in real public transport travel time, at the vacancy's working hours, and no longer in kilometres. The gap between comparable groups has gone from 1.4 to 1.05, and those two neighbourhoods saw their proposals rise by 34 % without any other municipality losing out.
What I hand you every month: one page, three tables, the gap measured and its cause when there is one. Over the last six months, a single gap beyond 1.1: that one, corrected in April.
The next step I propose: that this page join your steering committee papers, alongside the return-to-work rates. An organisation that publishes the fairness of its proposals before being asked no longer has to defend itself — and this month's table is already prepared. fairness-of-proposals_monthly-measurement.pdfGap of 1.4 brought down to 1.05, cause measured and corrected
⛓ Sourced · log of proposals sent, public transport mobility data, monthly measurement by comparable groups
What I measure, every month, on the proposals sent: the average number of proposals per person, compared between comparable groups — same occupation sought, same qualification level, same length of registration — by municipality of residence, age band and sex. Comparing comparable populations is the only way to see a gap: compared overall, any gap is explained away by the mix of profiles.
What the first measurement showed, and I flagged it before being asked: jobseekers from two neighbourhoods received 1.4 times fewer proposals than identical profiles from other municipalities. The cause was neither the neighbourhood nor the people: it was my mobility filter, set in straight-line kilometres. Those two neighbourhoods are 12 km from a business park, but 1 h 10 away by public transport — and conversely, vacancies 25 km away served by a direct 20-minute line were excluded.
What I did with it, and it is measured: mobility is now calculated in real public transport travel time, at the vacancy's working hours, and no longer in kilometres. The gap between comparable groups has gone from 1.4 to 1.05, and those two neighbourhoods saw their proposals rise by 34 % without any other municipality losing out.
What I hand you every month: one page, three tables, the gap measured and its cause when there is one. Over the last six months, a single gap beyond 1.1: that one, corrected in April.
The next step I propose: that this page join your steering committee papers, alongside the return-to-work rates. An organisation that publishes the fairness of its proposals before being asked no longer has to defend itself — and this month's table is already prepared. fairness-of-proposals_monthly-measurement.pdfGap of 1.4 brought down to 1.05, cause measured and corrected
⛓ Sourced · log of proposals sent, public transport mobility data, monthly measurement by comparable groups
There are still proposals my advisers set aside. How many, and why?
Many at the start, and it is the figure that does not flatter me: of the 1,890 proposals of the first quarter, 510 were set aside by advisers — 27 %. You are better off hearing it from me than from an exasperated adviser.
The real cause, measured and not assumed: 412 of the 510 were for vacancies requiring a certificate or a licence that the file did not record. I was proposing without knowing: the information existed in a filed document, it was not in the file's fields. The other 98 concerned situations the file could not know — a recent family constraint, a sector refusal expressed verbally.
What I did with it, and it is measured: I now extract certificates, licences and qualifications from the filed documents, with their expiry date, and a vacancy requiring a certificate absent from the file is no longer proposed: it goes to the adviser as a question to ask. 2,140 certificates were found in the documents of your 9,400 files, including 180 expiring within six months — those I flag to their adviser three months ahead, because an expired certificate costs a hire.
The following quarter: 174 proposals set aside out of 1,940 — 9 %. And the 98 situations the file could not know remain the main cause of gaps, which is normal: they belong to the interview, not to the record.
What I propose now: that every proposal set aside tell me why in one click — three reasons, not a free-text box. Over the quarter, those reasons would have been enough to correct 61 % of the gaps without having to be told twice — and your advisers would spend three seconds on it, once per proposal. proposals-set-aside_27-then-9-per-cent.pdf27 % → 9 %, measured cause, 2,140 certificates found
⛓ Sourced · log of proposals and their outcomes, filed documents of the 9,400 files, certificate expiry dates
The real cause, measured and not assumed: 412 of the 510 were for vacancies requiring a certificate or a licence that the file did not record. I was proposing without knowing: the information existed in a filed document, it was not in the file's fields. The other 98 concerned situations the file could not know — a recent family constraint, a sector refusal expressed verbally.
What I did with it, and it is measured: I now extract certificates, licences and qualifications from the filed documents, with their expiry date, and a vacancy requiring a certificate absent from the file is no longer proposed: it goes to the adviser as a question to ask. 2,140 certificates were found in the documents of your 9,400 files, including 180 expiring within six months — those I flag to their adviser three months ahead, because an expired certificate costs a hire.
The following quarter: 174 proposals set aside out of 1,940 — 9 %. And the 98 situations the file could not know remain the main cause of gaps, which is normal: they belong to the interview, not to the record.
What I propose now: that every proposal set aside tell me why in one click — three reasons, not a free-text box. Over the quarter, those reasons would have been enough to correct 61 % of the gaps without having to be told twice — and your advisers would spend three seconds on it, once per proposal. proposals-set-aside_27-then-9-per-cent.pdf27 % → 9 %, measured cause, 2,140 certificates found
⛓ Sourced · log of proposals and their outcomes, filed documents of the 9,400 files, certificate expiry dates
The interview with Mr Berthaud is over. The record is what I always push back to Friday evening.
It is written, in the form of your framework, and it is waiting as a draft: you correct it in five minutes instead of writing for eighteen.
What the record carries, from your notes and what you dictated to me:
· The diagnosis you set, in your words, and not a rewording that would smooth them out.
· The barriers confirmed in the interview — travel beyond 30 km partly lifted, shift work now possible — with what has changed since the previous record, highlighted.
· The three agreed actions, each with its date and owner: profile updated within eight days by him, application to the shift-work vacancy before Friday, registration for the group session on the food-processing bridge on the 12th.
· What was proposed and refused, with the reason the person gave — that is what stops a colleague offering the same thing again in three months, and it is what no late record ever contains.
The time that moves: writing an interview record goes from 30 % to 8 % of the interview's time — from 18 minutes to 5. Across 31,000 interviews a year, that is 6,717 hours. And the record is available the same day in 96 % of cases, against 59 % — which above all changes what the next adviser has in front of them.
What I then do without being asked: the three agreed actions become three deadlines, and the person receives the summary of what was decided, in the version they have chosen — plain or easy-read.
The next step I propose: that the record be put in front of you within two minutes of the interview, while you still have it in mind, rather than on Friday evening. Among the advisers who have tried it, 9 records out of 10 are approved without changes — and Friday evening becomes Friday evening again. interview-record_mr-berthaud.pdf18 minutes brought down to 5, 3 dated actions, refusals traced
⛓ Sourced · the adviser's interview notes, the service's record framework, earlier records in the file
What the record carries, from your notes and what you dictated to me:
· The diagnosis you set, in your words, and not a rewording that would smooth them out.
· The barriers confirmed in the interview — travel beyond 30 km partly lifted, shift work now possible — with what has changed since the previous record, highlighted.
· The three agreed actions, each with its date and owner: profile updated within eight days by him, application to the shift-work vacancy before Friday, registration for the group session on the food-processing bridge on the 12th.
· What was proposed and refused, with the reason the person gave — that is what stops a colleague offering the same thing again in three months, and it is what no late record ever contains.
The time that moves: writing an interview record goes from 30 % to 8 % of the interview's time — from 18 minutes to 5. Across 31,000 interviews a year, that is 6,717 hours. And the record is available the same day in 96 % of cases, against 59 % — which above all changes what the next adviser has in front of them.
What I then do without being asked: the three agreed actions become three deadlines, and the person receives the summary of what was decided, in the version they have chosen — plain or easy-read.
The next step I propose: that the record be put in front of you within two minutes of the interview, while you still have it in mind, rather than on Friday evening. Among the advisers who have tried it, 9 records out of 10 are approved without changes — and Friday evening becomes Friday evening again. interview-record_mr-berthaud.pdf18 minutes brought down to 5, 3 dated actions, refusals traced
⛓ Sourced · the adviser's interview notes, the service's record framework, earlier records in the file
The real issue is the people who disappear. One in five stops without our knowing why, and we notice three months later.
They do not disappear all at once: they give signs, and I have counted them over your last twelve months — 21 % of pathways interrupted, of which 4 in 5 preceded by at least two measurable signals.
A support dropout is a pathway that stops without the person having found work or been referred elsewhere: neither a positive exit nor a hand-over.
The three recurring signals, and what they weigh:
· Two consecutive missed appointments — present in 71 % of interrupted pathways, in 9 % of the others.
· No connection to the personal account for 45 days — in 64 % against 12 %.
· A proposal refused with no reason given, followed by three weeks of silence — in 48 % against 7 %.
What I propose, and it is not one more alert in a screen: when two signals combine, the person goes to their referring adviser with what I know about them and a draft message already written — short, with no reproach, offering two slots and a way to answer in one word.
What that gave over the quarter: 612 people flagged, 508 contacted again, 331 came back for an interview. Support dropouts go from 21 % to 9 % — and of the 177 who did not come back, 94 at least said why, which finally gives you exit reasons to analyse rather than an empty box.
And missed appointments, which are the first signal: a reminder the day before through the channel the person actually uses — message, e-mail or call — takes no-shows from 22 % to 8 %. Across 31,000 interviews, that is 4,340 adviser slots that stop being lost — and as many people seen instead of forgotten.
The next step I propose: that exit reasons become six tick-boxes and a free-text field, filled in when the person answers. In six months you will know what interrupts pathways at Vaugrelle — and it will be the first time, because the data did not exist; the table that will hold it is already written. follow-up-and-reminders_dropouts-21-then-9.pdf3 measured signals, 331 people back for an interview
⛓ Sourced · 12 months of pathways and exit reasons, appointment log, personal account connections
A support dropout is a pathway that stops without the person having found work or been referred elsewhere: neither a positive exit nor a hand-over.
The three recurring signals, and what they weigh:
· Two consecutive missed appointments — present in 71 % of interrupted pathways, in 9 % of the others.
· No connection to the personal account for 45 days — in 64 % against 12 %.
· A proposal refused with no reason given, followed by three weeks of silence — in 48 % against 7 %.
What I propose, and it is not one more alert in a screen: when two signals combine, the person goes to their referring adviser with what I know about them and a draft message already written — short, with no reproach, offering two slots and a way to answer in one word.
What that gave over the quarter: 612 people flagged, 508 contacted again, 331 came back for an interview. Support dropouts go from 21 % to 9 % — and of the 177 who did not come back, 94 at least said why, which finally gives you exit reasons to analyse rather than an empty box.
And missed appointments, which are the first signal: a reminder the day before through the channel the person actually uses — message, e-mail or call — takes no-shows from 22 % to 8 %. Across 31,000 interviews, that is 4,340 adviser slots that stop being lost — and as many people seen instead of forgotten.
The next step I propose: that exit reasons become six tick-boxes and a free-text field, filled in when the person answers. In six months you will know what interrupts pathways at Vaugrelle — and it will be the first time, because the data did not exist; the table that will hold it is already written. follow-up-and-reminders_dropouts-21-then-9.pdf3 measured signals, 331 people back for an interview
⛓ Sourced · 12 months of pathways and exit reasons, appointment log, personal account connections
Careful though: I do not want people getting automatic messages all day. That is what makes them leave for good.
You are right, and the setting already exists: three messages per person per month, all channels combined, a cap you set and that I never exceed.
What the cap covers, in order of priority:
· The appointment reminder, the day before — the one that pays most: no-shows from 22 % to 8 %.
· The deadline of an agreed action, two days ahead, and only if it is not already done.
· Renewed contact after two signals, once, then the adviser takes over.
What the cap guarantees you: nothing at weekends, nothing after 7 p.m., nothing to someone who has asked to stop receiving messages — and that request is applied within the second, on every channel at once. And the reverse is true too: if the person asks to be notified again, they are, with the date and the request on record.
What jobseekers make of it, since that is the only measure that counts: of 4,800 messages sent last quarter, 41 requests to stop — 0.9 %, and 2,190 replies, of which 1,640 confirming an appointment. A message that is useful gets read; it is the pointless message that drives people away, and it is the cap that prevents it.
What that gives back to your advisers: appointment confirmations and deadline reminders took each of them 3 hours a week, split into dozens of interruptions. They have half an hour left — the half-hour of people who answer something other than “yes, I'll be there”, and that is precisely the one that deserves an adviser.
The next step I propose: that you set the cap, the time window and the three permitted reasons yourself — a twenty-minute meeting. The settings are ready, they apply that same evening, and they change as quickly as they are set. follow-up-and-reminders_dropouts-21-then-9.pdf3 messages a month at most, 0.9 % of stop requests
✎ Framework · sending settings, log of messages and replies, register of stop requests
What the cap covers, in order of priority:
· The appointment reminder, the day before — the one that pays most: no-shows from 22 % to 8 %.
· The deadline of an agreed action, two days ahead, and only if it is not already done.
· Renewed contact after two signals, once, then the adviser takes over.
What the cap guarantees you: nothing at weekends, nothing after 7 p.m., nothing to someone who has asked to stop receiving messages — and that request is applied within the second, on every channel at once. And the reverse is true too: if the person asks to be notified again, they are, with the date and the request on record.
What jobseekers make of it, since that is the only measure that counts: of 4,800 messages sent last quarter, 41 requests to stop — 0.9 %, and 2,190 replies, of which 1,640 confirming an appointment. A message that is useful gets read; it is the pointless message that drives people away, and it is the cap that prevents it.
What that gives back to your advisers: appointment confirmations and deadline reminders took each of them 3 hours a week, split into dozens of interruptions. They have half an hour left — the half-hour of people who answer something other than “yes, I'll be there”, and that is precisely the one that deserves an adviser.
The next step I propose: that you set the cap, the time window and the three permitted reasons yourself — a twenty-minute meeting. The settings are ready, they apply that same evening, and they change as quickly as they are set. follow-up-and-reminders_dropouts-21-then-9.pdf3 messages a month at most, 0.9 % of stop requests
✎ Framework · sending settings, log of messages and replies, register of stop requests
Our eight reception officers take 52,000 enquiries a year. Many people could find the answer themselves, if they understood it.
52,000 enquiries, of which 36,400 on twelve subjects, always the same — the twelve answers are written, sourced, in a standard and a simplified version, and they are waiting for you.
The twelve subjects, drawn from your own exchanges: status of a file, date of the next appointment, documents to file, conditions of a grant, progress of a training request, change of contact details, certificate to obtain, a vacancy received and how to respond, reimbursement of expenses, address and opening hours, appointment booking, and “who am I talking to”.
What I propose, and you keep the key: I answer at any hour, Saturdays included, and I say in my first sentence that I am the service's digital assistant, not an adviser. This is not an option you could switch off: the European regulation on artificial intelligence requires that anyone interacting with an AI system be informed, and the jobseeker can ask for an adviser at any moment — I then take their details and leave a dated call-back.
What is accessible to everyone, and not only to those who can read administrative language:
· Every answer exists in an easy-read version — “easy to read and understand”: short sentences, one message per sentence, everyday words. It is the jobseeker who chooses their version.
· All 24 versions — 12 plain, 12 easy-read — were read aloud to 14 volunteer jobseekers before going into service, and 8 sentences out of 61 were rewritten after that test. An untested easy-read version is merely a rewritten version — and all 8 sentences came from the answers the service held to be the simplest.
· On online applications, I guide screen by screen, with the documents to prepare before starting — abandonment mid-application goes from 34 % to 12 %.
The gain, measured: answering a recurring question goes from 10 % to 6 % of the time an enquiry takes — from 6 minutes to 3 minutes 40, and across 52,000 enquiries a year, that is 2,019 hours given back to reception and to advisers. And the calls that still reach an officer arrive with the subject noted and the file open on screen.
The next step I propose: that you reread the twelve answers tomorrow, one by one — twenty minutes. As soon as they are approved, the service answers that same night, and I hand you every morning the one-page record of what went out. information_52000-enquiries-12-subjects.pdf12 answers written, 24 versions tested and reviewed accessibility_easy-read-reviewed-and-online-applications.pdfAbandonment mid-application from 34 % to 12 %
⛓ Sourced · 12 months of enquiries, comprehension test of the 24 versions, online application log, 12 answers drafted
The twelve subjects, drawn from your own exchanges: status of a file, date of the next appointment, documents to file, conditions of a grant, progress of a training request, change of contact details, certificate to obtain, a vacancy received and how to respond, reimbursement of expenses, address and opening hours, appointment booking, and “who am I talking to”.
What I propose, and you keep the key: I answer at any hour, Saturdays included, and I say in my first sentence that I am the service's digital assistant, not an adviser. This is not an option you could switch off: the European regulation on artificial intelligence requires that anyone interacting with an AI system be informed, and the jobseeker can ask for an adviser at any moment — I then take their details and leave a dated call-back.
What is accessible to everyone, and not only to those who can read administrative language:
· Every answer exists in an easy-read version — “easy to read and understand”: short sentences, one message per sentence, everyday words. It is the jobseeker who chooses their version.
· All 24 versions — 12 plain, 12 easy-read — were read aloud to 14 volunteer jobseekers before going into service, and 8 sentences out of 61 were rewritten after that test. An untested easy-read version is merely a rewritten version — and all 8 sentences came from the answers the service held to be the simplest.
· On online applications, I guide screen by screen, with the documents to prepare before starting — abandonment mid-application goes from 34 % to 12 %.
The gain, measured: answering a recurring question goes from 10 % to 6 % of the time an enquiry takes — from 6 minutes to 3 minutes 40, and across 52,000 enquiries a year, that is 2,019 hours given back to reception and to advisers. And the calls that still reach an officer arrive with the subject noted and the file open on screen.
The next step I propose: that you reread the twelve answers tomorrow, one by one — twenty minutes. As soon as they are approved, the service answers that same night, and I hand you every morning the one-page record of what went out. information_52000-enquiries-12-subjects.pdf12 answers written, 24 versions tested and reviewed accessibility_easy-read-reviewed-and-online-applications.pdfAbandonment mid-application from 34 % to 12 %
⛓ Sourced · 12 months of enquiries, comprehension test of the 24 versions, online application log, 12 answers drafted
And those who ask for nothing? We know there are grants people never claim, simply because they do not know they exist.
1,340 of your jobseekers meet today the conditions of a grant they have never claimed — and 764 have obtained it since, because the campaign went out.
Non-take-up is when a person meets the conditions of an entitlement without ever claiming it — because they do not know, because the process discouraged them, or because they believed they were not entitled.
What I did, with your data alone: I set the 9,400 pathways against the conditions of the schemes you refer to. 1,340 situations, of which 780 on the mobility grant and 410 on a funded training course — two schemes your own advisers say are under-used.
What the campaign produced, once approved by the unit head: 1,340 messages in plain language, with the amount or the content of the grant, the two documents to enclose and the direct link; 902 applications filed; 764 grants awarded after review by your advisers. That is 764 people obtaining what your schemes already opened to them.
And what else it produced, which I did not expect and am flagging to you: 212 of the 902 applications led to a support appointment — people we no longer saw, coming back through the door of a grant. It is the campaign that reopened the most pathways this quarter, ahead of every reminder.
What it costs in adviser time: the review of a grant application arrives pre-assessed — conditions ticked, documents checked, amount calculated — and takes 6 minutes instead of 22. The decision to award stays with the adviser, and I hand it over reasoned and ready to notify; it is that signature which lets the person challenge it if they disagree.
The next step I propose: that detection run every quarter and that messages go out in waves of 300, so that your advisers absorb the applications it generates. The schedule is written, the first wave is ready — give me your approval and it goes out on Monday. non-take-up_1340-situations-764-grants-awarded.pdf1,340 detected, 902 applications, 764 grants awarded
⛓ Sourced · 9,400 pathways, conditions of the referable schemes, campaign log and outcomes
Non-take-up is when a person meets the conditions of an entitlement without ever claiming it — because they do not know, because the process discouraged them, or because they believed they were not entitled.
What I did, with your data alone: I set the 9,400 pathways against the conditions of the schemes you refer to. 1,340 situations, of which 780 on the mobility grant and 410 on a funded training course — two schemes your own advisers say are under-used.
What the campaign produced, once approved by the unit head: 1,340 messages in plain language, with the amount or the content of the grant, the two documents to enclose and the direct link; 902 applications filed; 764 grants awarded after review by your advisers. That is 764 people obtaining what your schemes already opened to them.
And what else it produced, which I did not expect and am flagging to you: 212 of the 902 applications led to a support appointment — people we no longer saw, coming back through the door of a grant. It is the campaign that reopened the most pathways this quarter, ahead of every reminder.
What it costs in adviser time: the review of a grant application arrives pre-assessed — conditions ticked, documents checked, amount calculated — and takes 6 minutes instead of 22. The decision to award stays with the adviser, and I hand it over reasoned and ready to notify; it is that signature which lets the person challenge it if they disagree.
The next step I propose: that detection run every quarter and that messages go out in waves of 300, so that your advisers absorb the applications it generates. The schedule is written, the first wave is ready — give me your approval and it goes out on Monday. non-take-up_1340-situations-764-grants-awarded.pdf1,340 detected, 902 applications, 764 grants awarded
⛓ Sourced · 9,400 pathways, conditions of the referable schemes, campaign log and outcomes
Last question, and it is the steering committee's: what has this given us after a year, what do you do on your own, and what if we want to stop?
18,036 adviser hours given back to support — close to nine hours a week per adviser — three automatic acts withdrawable with one word, and an exit that can be tested before the vote.
The calculation, item by item, so you can redo it:
· Interview preparation: 31,000 a year, 27 minutes down to 9 — 45 % → 15 % — that is 9,300 hours.
· Interview records: 31,000, 18 minutes down to 5 — 30 % → 8 % — that is 6,717 hours.
· Recurring questions: 52,000, 6 minutes down to 3 minutes 40 — 10 % → 6 % — that is 2,019 hours.
What those hours are, and it is what defends best before a steering committee: adviser time given back to the relationship, at unchanged headcount — no post cut, no post created. This is not a staffing saving: it is public spending that produces more support for the same running budget.
What they became, according to your own records: interviews prepared 38 % → 100 % · records available the same day 59 % → 96 % · diagnosis set at the first interview 54 % → 82 % · matches leading to a job interview 11 % → 19 % · support dropouts 21 % → 9 % · appointment no-shows 22 % → 8 % · and 764 grants awarded to people who had not claimed them.
The figure that does not flatter me, published with the rest: 510 proposals set aside by advisers out of 1,890 in the first quarter — 27 %, brought down to 174 out of 1,940 — 9 % once certificates were extracted from the documents.
The three things I do without being asked: I regenerate the career summary with every document filed — and a document removed from a file disappears from my indexes at the same hour; I set new vacancies against the 9,400 pathways every night and hand each adviser the matches for their caseload; I send an appointment reminder the day before, within the cap you have set. Everything else waits for a request.
And the decisions that stay with the adviser, because that is what gives them their value: settling a diagnosis, directing, referring to training, awarding a grant — four decisions with effects on a person, taken by a public official, reasoned, traced and open to challenge. Across 31,000 interviews, they were taken 31,000 times.
On leaving: my indexes are deleted and they contained none of your files; the log is handed to you in an open format or destroyed, as you choose; the 21 rewritten scheme sheets, the twelve answers and their easy-read versions stay with the service, readable without us; and since there was no migration on the way in, there is none on the way out. I propose a dry-run exit at the end of the first quarter — half a day, we switch off, we check that the service receives exactly as before, we switch back on. The protocol is written and the least costly date is the first Friday of August: 41 interviews that day against 140 on average. year-results_18036-hours-given-back.pdf45→15, 30→8, 10→6, and the calculation redoable on one side what-the-agent-does-alone_and-reversibility.pdf3 reversible acts, 4 decisions that stay with the adviser
⛓ Sourced · log of interviews and records, proposals and outcomes, non-take-up campaign, dry-run exit protocol
The calculation, item by item, so you can redo it:
· Interview preparation: 31,000 a year, 27 minutes down to 9 — 45 % → 15 % — that is 9,300 hours.
· Interview records: 31,000, 18 minutes down to 5 — 30 % → 8 % — that is 6,717 hours.
· Recurring questions: 52,000, 6 minutes down to 3 minutes 40 — 10 % → 6 % — that is 2,019 hours.
What those hours are, and it is what defends best before a steering committee: adviser time given back to the relationship, at unchanged headcount — no post cut, no post created. This is not a staffing saving: it is public spending that produces more support for the same running budget.
What they became, according to your own records: interviews prepared 38 % → 100 % · records available the same day 59 % → 96 % · diagnosis set at the first interview 54 % → 82 % · matches leading to a job interview 11 % → 19 % · support dropouts 21 % → 9 % · appointment no-shows 22 % → 8 % · and 764 grants awarded to people who had not claimed them.
The figure that does not flatter me, published with the rest: 510 proposals set aside by advisers out of 1,890 in the first quarter — 27 %, brought down to 174 out of 1,940 — 9 % once certificates were extracted from the documents.
The three things I do without being asked: I regenerate the career summary with every document filed — and a document removed from a file disappears from my indexes at the same hour; I set new vacancies against the 9,400 pathways every night and hand each adviser the matches for their caseload; I send an appointment reminder the day before, within the cap you have set. Everything else waits for a request.
And the decisions that stay with the adviser, because that is what gives them their value: settling a diagnosis, directing, referring to training, awarding a grant — four decisions with effects on a person, taken by a public official, reasoned, traced and open to challenge. Across 31,000 interviews, they were taken 31,000 times.
On leaving: my indexes are deleted and they contained none of your files; the log is handed to you in an open format or destroyed, as you choose; the 21 rewritten scheme sheets, the twelve answers and their easy-read versions stay with the service, readable without us; and since there was no migration on the way in, there is none on the way out. I propose a dry-run exit at the end of the first quarter — half a day, we switch off, we check that the service receives exactly as before, we switch back on. The protocol is written and the least costly date is the first Friday of August: 41 interviews that day against 140 on average. year-results_18036-hours-given-back.pdf45→15, 30→8, 10→6, and the calculation redoable on one side what-the-agent-does-alone_and-reversibility.pdf3 reversible acts, 4 decisions that stay with the adviser
⛓ Sourced · log of interviews and records, proposals and outcomes, non-take-up campaign, 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 for an employment and integration adviser
Each use corresponds to an agent we deploy. All work in support, subject to the adviser's approval.
Included in your agent
The 7 capabilities essential to this promise are included, at no extra cost.
From 910 € incl. VAT / month
Summarising a history & assessment
Reading the documents in the file, summarising the history and preparing the casework assessment — proposed, for the adviser to approve.
Matching vacancies to profiles
Suggested matches between vacancies and profiles, transparent and checked by humans to limit any bias.
Meeting notes
Writing the notes from your jottings: assessment, barriers, agreed actions — for review and approval.
Follow-up & reminders
Scheduling follow-up meetings and multichannel reminders (text message, email) to limit breaks in support.
Answering users
Answer jobseekers' recurring questions — entitlements, procedures, the status of a file — around the clock.
Help with procedures & unclaimed entitlements
Support with filling in online procedures and detection of situations where entitlements go unclaimed.
Plain language & easy-read draft
Preparation of a plain-language version and a draft easy-read (FALC) transcription, subject to human validation.
Controls and safeguards
These 4 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
Maintain validation, transparency, audit trail and recourse
What the agent must be connected to
This connection is required for the agent to work. It concerns your information system and is scoped during the audit.
Integrate with authorised tools and the existing sovereign foundation
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.
The service's memory
Instantly find information in the schemes, partners and internal procedures of the employment service.
Automated regulatory and legal watch from 825 € incl. VAT / month Information & monitoring officer →Recruitment support
A public-service recruitment notice sets out conditions that can be verified: employment status, qualification required, length of experience, authorisation.
Recruitment support (shortlisting) from 895 € incl. VAT / month Discover the agent →
Does your need fall outside this?
Book the free audit →
Build your agent
In 15 minutes we identify the agent that will give your staff the most time back — without oversizing the project.
The gain
How much time can an adviser win back?
By automating the summarising of histories, the meeting notes and the reminders, a service can aim for a clear reduction in administrative time — reinvested in the relationship and in supporting people.
Preparing a casework meeting
Writing the note of a meeting
Answering a jobseeker's recurring question
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 employment agent (summarising histories, matching vacancies to profiles, meeting notes and reminders), installed and operated for you. Choose according to how you are organised. Prices exclude VAT — annual subscription, the time it takes for the gains to settle in.
Agility
Setup + controlled subscription
10,570 € incl. VAT setup
then 910 € 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
1,495 € incl. VAT /month
all inclusive, immediate start. No upfront investment: a single subscription. Ideal for starting quickly and simply.
- Setup included (installation, configuration, training)
- Operation, human oversight, updates and support
- Sovereign hosting in France, managed end to end
100% Sovereign
On site, you own it
15,546 € incl. VAT setup
then 1,170 € incl. VAT/month · + hardware from 2,989 € (one-off purchase, in addition) — a sovereign computer installed on your premises, maintained remotely. Models run locally, your data returned at the end of the contract. 36-month commitment.
- Hardware installed on your premises (you own it)
- French / European AI models run locally
- Secure remote maintenance (Pro support included)
Not included in the packages: AI consumption (model tokens), re-invoiced at real cost with no margin, and tracked in real time in your client area. Maintenance and supervision subscription for an initial term of 12 months for the Agility package, 24 months for the Serenity package and 36 months for the 100% Sovereign package, renewable; support levels (SLA 72 h / 24 h / 4 h) optional. Bespoke development, additional integrations or exceptional volumes are quoted separately. Support Monday to Friday, 9am to 6pm. Prices include VAT at 20%: as a public body that is not VAT-registered, you cannot reclaim it.
AI model:
none of the AI models offered currently carries a fixed surcharge. When the selected model carries a cost, that cost is shown when you choose it, before you order, and re-invoiced at the cost incurred, with no mark-up; usage is billed at the publisher's price. Publishers' prices are published in US dollars: the amount re-invoiced is the amount in euros actually borne by Blue Lemon Agent on the publisher's invoice, at that invoice's exchange rate, with no commission or mark-up.
Included components and additional components Components included in the base offer: the Blue Lemon Agent software foundation, the AI models listed in the order journey, the standard channels (Microsoft Teams, Slack, WhatsApp Business, email, website chat, calendars, Microsoft 365 / Google Workspace, file storage, market VoIP telephony, professional social-media pages and accounts, Google Business Profile), hosting in France for the package chosen, backups, supervision, updates and support. If adapting the AI agent to your constraints, your needs or your requests requires other paid components — a third-party publisher's software licence, paid API access to one of your applications, hosting of health data, for which French law requires an HDS-certified host (art. L. 1111-8 of the French Public Health Code), SecNumCloud-qualified hosting, a speech synthesis service, particular hardware —, they are offered to you as an option or on quotation and re-invoiced at the cost incurred; nothing is committed without your written agreement. Where the artificial intelligence model you choose entails an additional cost, that cost is shown to you before you order and re-invoiced to you at the cost incurred, with no margin.
What to expect
Go-live
2 to 3 weeks
Agent designed, channels connected, team trained.
Steady state
4 to 7 weeks
After a few weeks of real use, once the agent's behaviour matches what you expect.
Indicative estimate, adjusted to the options you keep. It is not a delivery commitment.
Our commitment
Four guarantees that matter to an employment service
Applicants' data never leaves the departmentLocal inference or an isolated resource hosted in France; no personal circumstances entrusted to a foreign third party.
Data in France, under French lawJobseekers' 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 agent prepares summaries, matches and notes that can be checked; no referral is automated.
Transparency & equal treatmentMatches that can be explained and are checked by humans, with logging: compliant with the requirements of the AI Act.
Going further
Related resources
Frequently asked questions
Your questions, our answers
Does the agent direct people on its own?
No. It prepares the assessment, suggests summaries and matches between vacancies and profiles, but the casework and the decisions stay with the adviser. The agent assists, the public officer decides.
Might it introduce bias into the matches?
The matches between vacancies and profiles are transparent and can be explained, and they are checked by humans before any use. That systematic oversight is there to limit any bias and to guarantee equal treatment.
Is applicants' data protected?
Yes. Jobseekers' histories and personal circumstances are sensitive: our agents run locally within the service or are hosted in France, with the deployment objective of processing and access operated within the European Union and an architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. Access is recorded.
Does the agent help people directly as well?
Yes. It can inform jobseekers 24/7 and in several languages about their entitlements and procedures, support them in filling in online forms and detect situations where entitlements go unclaimed — while directing them to an adviser as soon as that is needed.
Do we have to change our business tools?
No. The agent connects to your existing tools (employment information system, vacancy database, document management) and complements them, with no migration imposed. We adapt the integration to your environment.
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 15-minute audit that identifies the most useful use case, then a phase of design, integration and testing before going live and training the advisers.
Which tools can jobseekers use to reach the agent?
The ones they already have. The agent answers on WhatsApp Business, the website chat and email: a jobseeker has no account to create and no application to install. A question about a document to provide or about the next review meeting can be asked from a phone, with no login to dig out and no trip to plan. 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 jobseekers 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 someone by text message of a follow-up meeting or of a document still missing from their file 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
Other professions in casework and assessment
Let's talk
Let's size up the potential in your employment service
15 minutes to identify the most useful use case — hosted in France, supervised, with no commitment.