An orchestration agent, scoped with you before it is priced. This agent coordinates several specialised agents. Its orchestration follows your actual workflows — which is why it is scoped with you rather than bought off the shelf. We establish the scope together, then the quotation commits it.
The journeys described below form the scope that this review refines and the quotation commits. The specialised agents it coordinates can be ordered today.
Request a quote
● Public sector — Medical administrative office
The medical secretaries' AI agent: lay out the reports, unclog the office
Laying out reports, handling letters and appointment notices, pre-coding procedures take up a considerable share of medical administrative assistants' time — at the expense of welcoming and supporting the patient. Your AI agent absorbs that repetitive work. Hosted in France — on local inference or an isolated resource, the HDS requirement settled at scoping — health data stays under control. The public officer keeps the decision: the AI agent assists, the public officer decides.
Hosted in France
Health data protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Medical administrative assistant · hosted in France, HDS requirement at scoping
Lay out Mrs Durand's consultation report from the dictation, and prepare the letter to her GP.
Report laid out from the dictation, structured on the department's template (reason, examination, conclusion). Letter to the GP ready, with the heading and contact details taken from the patient record.
Two points to check: an ambiguous medical term in the dictation and the dosage of one treatment — I have highlighted them. For review and approval.
⛓ Source · dictation + patient record (secure health messaging)
Two points to check: an ambiguous medical term in the dictation and the dosage of one treatment — I have highlighted them. For review and approval.
⛓ Source · dictation + patient record (secure health messaging)
Also schedule the follow-up appointment in three months and the notice.
Follow-up appointment proposed in the first free slot at 3 months, respecting the department's diary rules. Notice prepared with the date, place, preparation instructions and an automatic reminder. The pre-coding of the consultation's procedures has also been started, for approval.
Nothing is sent or coded without your agreement.
✎ Action · appointment and notice ready — the public officer approves
Nothing is sent or coded without your agreement.
✎ Action · appointment and notice ready — the public officer approves
Local inference · no data outside the EU
Data hosted in France
Sovereign by designLocal inference or hosting in France
GDPR & AI Act: governed deploymentTraceability & human oversight
TurnkeyDesigned, installed and operated for you
The public officer decidesThe AI agent assists, never approves alone
✦ In brief
In a public health establishment, a Blue Lemon Agent agent unclogs the medical office — laying out reports and letters, managing appointments and notices, pre-coding procedures — and gives time back to the medical administrative assistants. It runs on local inference or is hosted in France on a dedicated, isolated resource: health data never travels outside the European Union, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. The time won back is returned to welcoming and supporting the patient, in the continuity and equality of the public service. Live within a few weeks. The AI agent assists, the public officer decides. 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
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 medical offices — and why they hesitate
Patients expect shorter waits for appointments and reports sent on time. But the time of medical administrative assistants is mechanically absorbed by typing, letters and coding — and the data involved is among the most sensitive there is: health data.
! The issue
The medical office is caught between patients and clinicians who expect responsiveness and continuity, and an administrative load that keeps growing (reports, letters, notices, coding). Yet most consumer AI solutions amount to entrusting health data, patient records and medical correspondence to a third party, often hosted outside Europe and subject to the Cloud Act.
✓ Our answer
AI is only of interest to a public health service if it is sovereign and confidential by design. Local inference or an isolated resource hosted in France, the HDS requirement settled at scoping, systematic human oversight, the medical act and the final coding reserved to the professional: the time saved on administration is never paid for in lost confidentiality. The aim is not to replace the medical secretary, but to give them back time for the patient.
The decisive point
The confidentiality of health data: sovereignty & compliance
A medical office handles patients' most sensitive data. Here is how the architecture of our agents protects it, record by record.
Local inference
The agent can run on a machine at the establishment: no patient data leaves the network, nothing passes through a public cloud.
Hosting in France, HDS requirement at scoping
Otherwise, a dedicated and isolated resource, hosted in France under French law — your data: processing and access within the European Union targeted by the architecture. In France, hosting health data on behalf of others is subject to HDS certification of the host: that requirement is settled with you at scoping, before any go-live.
Reduced extraterritorial exposure
For health data, the architecture aims to reduce exposure to the Cloud Act and FISA 702; being located in France or in the European Union does not, on its own, guarantee immunity.
One isolated resource per establishment
No pooling of health data: an environment strictly dedicated to your establishment.
Encryption & controlled access
Encryption in transit and at rest, role-based access, strong authentication and logging of access.
AI Act: governed deployment
The agent is strictly in support; no coding is approved 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 health data, the HDS requirement is settled with you at scoping, with a SecNumCloud option 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 bodyVaubourg Hospital — public health establishment (fictional public body)
- Activity
- 320-bed public hospital — outpatient clinics in 5 specialties: cardiology, endocrinology, respiratory medicine, rheumatology, orthopaedic surgery
- Headcount
- 1,140 staff, including the 11 the AI agent serves: 9 medical administrative assistants in outpatient clinics and 2 nurse managers
- Service users
- 38,000 service users across the area, 41 % of them over 70, and 9 languages spoken at the front desk
- Volume of the office
- 41,000 outpatient consultations a year, 310 dictated clinic letters a week, 1,850 calls to the office a week
- Tools in place
- Electronic patient record, clinic diary, secure health messaging and digital dictation — the agent plugs into them, no change of software
- Who decides
- The doctor approves the clinic letter and the coding of procedures; the medical administrative assistant approves letters and appointment letters; the nurse manager settles the diary rules
- Room for improvement
- A clinic letter waits 6.4 days to reach the family doctor; 22 % of calls go unanswered at peak times; the waiting time for an endocrinology appointment is 94 days
Vaubourg Hospital is not looking to shrink its medical office: it is looking to give its nine medical administrative assistants back the time that typing, letters and coding take from them, and put it into welcoming and supporting service users. The AI agent runs on local inference on a hospital machine, on dedicated hosting in France, with the HDS requirement settled at scoping, and plugs into the patient record, the clinic diary and secure health messaging: it prepares, the public servant approves. The exchanges below cover one quarter, from the first dictation handled to the costed 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.
Medical administrative assistant · hosted in France, HDS requirement at scoping
I have read the 8,100 approved clinic letters of your last six months, and drawn five departmental templates from them — one per specialty. You had none written down; there are five this morning.
A medical administrative assistant is one of your nine outpatient office staff: they lay out clinic letters, prepare correspondence, keep the diary and receive service users. A clinic letter is the document the doctor dictates after seeing a service user, and which the family doctor is waiting for in order to carry on the follow-up.
The gap I measured, and it is the one that decides the gain: across those 8,100 letters, 68 % of the corrections your staff made were about LAYOUT — headings, order of sections, letterhead, units, page setting — and not about medical content. Your assistants are not proofreading medicine: they are re-laying out, and that is exactly the work that can be written down once and for all.
The second gap, the one the service user feels: a clinic letter waits 6.4 days to reach the family doctor, and 5.1 of those 6.4 days are waiting before typing, not typing itself. Your delay is not a speed problem, it is a queue problem — and a queue empties from the front.
What each template carries, drawn from your own approved documents: the order of sections the specialty actually uses — reason for referral, history, examination, investigations, conclusion, plan —, the approved opening and closing formulas, the expected units, and the items the specialty always writes down.
· Cardiology: the conclusion placed before the examination in 84 % of your approved letters — the reverse of what your official framework asks for. It is your practice that is right, not the framework.
· Endocrinology: weight, height and change since the previous consultation present in 9 letters out of 10. I pull them from the patient record: they stop being forgotten on busy days.
· Respiratory medicine: a section on “current inhaled treatment” that appears in no framework and that all three doctors write, all three of them.
What this gives you from the next dictation onwards: the clinic letter arrives ALREADY in the right shape. Your staff stop re-laying out in order to proofread — and proofreading is where their profession shows.
What I propose: that the five lead doctors and your two nurse managers read the five templates — one hour in total, once — and I apply them from the next morning's dictation. Every template changes with a word, and that one hour is what brings down the 5.1 days of waiting before typing. departmental-templates_5-specialties.pdf5 templates, what each specialty always writes
⛓ Sourced · 8,100 approved clinic letters over 6 months, successive drafts before approval, dispatch log to family doctors
A medical administrative assistant is one of your nine outpatient office staff: they lay out clinic letters, prepare correspondence, keep the diary and receive service users. A clinic letter is the document the doctor dictates after seeing a service user, and which the family doctor is waiting for in order to carry on the follow-up.
The gap I measured, and it is the one that decides the gain: across those 8,100 letters, 68 % of the corrections your staff made were about LAYOUT — headings, order of sections, letterhead, units, page setting — and not about medical content. Your assistants are not proofreading medicine: they are re-laying out, and that is exactly the work that can be written down once and for all.
The second gap, the one the service user feels: a clinic letter waits 6.4 days to reach the family doctor, and 5.1 of those 6.4 days are waiting before typing, not typing itself. Your delay is not a speed problem, it is a queue problem — and a queue empties from the front.
What each template carries, drawn from your own approved documents: the order of sections the specialty actually uses — reason for referral, history, examination, investigations, conclusion, plan —, the approved opening and closing formulas, the expected units, and the items the specialty always writes down.
· Cardiology: the conclusion placed before the examination in 84 % of your approved letters — the reverse of what your official framework asks for. It is your practice that is right, not the framework.
· Endocrinology: weight, height and change since the previous consultation present in 9 letters out of 10. I pull them from the patient record: they stop being forgotten on busy days.
· Respiratory medicine: a section on “current inhaled treatment” that appears in no framework and that all three doctors write, all three of them.
What this gives you from the next dictation onwards: the clinic letter arrives ALREADY in the right shape. Your staff stop re-laying out in order to proofread — and proofreading is where their profession shows.
What I propose: that the five lead doctors and your two nurse managers read the five templates — one hour in total, once — and I apply them from the next morning's dictation. Every template changes with a word, and that one hour is what brings down the 5.1 days of waiting before typing. departmental-templates_5-specialties.pdf5 templates, what each specialty always writes
⛓ Sourced · 8,100 approved clinic letters over 6 months, successive drafts before approval, dispatch log to family doctors
Our doctors dictate fast and use in-house abbreviations. One misreading on a dose, and we are the ones who carry it.
Your 240 in-house abbreviations are written down and applied, and every clinic letter arrives with its checking points already marked: 2.1 per letter, of which 0.4 on a dose or a value, each with its audio extract.
What I did before writing the first line: I collected from your five lead doctors the lexicon of their 240 in-house abbreviations, the ones no dictionary knows. They lived in your staff's heads and left with every retirement — they are written down, they stay.
What I copy verbatim, character by character: doses, laboratory values, dates and laterality — left or right. These go from dictation to document with no rephrasing. A “40 mg” heard for “14 mg” cannot be caught at proofreading, because the reader reads what they expect: that is exactly what the highlight catches.
What every highlight carries, and it is what makes it usable: the audio extract, positioned to the second. Your assistant listens for three seconds instead of hunting through ten minutes of dictation. Checking a doubtful point goes from 4 minutes to 20 seconds.
The time this shifts, on the item you had costed: laying out a clinic letter took 60 % of the processing time — 22 minutes. It now takes 18 %, or 6 minutes 36, proofreading and approval included. Across 310 letters a week, that is 80 hours given back every week to your nine staff, in other words a full day for each of them.
What I propose: that the lexicon become a departmental document, reviewed once a year by the lead doctors. Over the first six weeks, 34 new abbreviations added themselves to it — precisely the ones a new member of staff takes six months to learn, and they will have them on day one. clinic-letter_laid-out-and-highlighted.pdfThe proposed document, its 2 highlights and their audio extract
⛓ Sourced · departmental lexicon (240 abbreviations), digital dictations, highlighted points across 4,030 clinic letters
What I did before writing the first line: I collected from your five lead doctors the lexicon of their 240 in-house abbreviations, the ones no dictionary knows. They lived in your staff's heads and left with every retirement — they are written down, they stay.
What I copy verbatim, character by character: doses, laboratory values, dates and laterality — left or right. These go from dictation to document with no rephrasing. A “40 mg” heard for “14 mg” cannot be caught at proofreading, because the reader reads what they expect: that is exactly what the highlight catches.
What every highlight carries, and it is what makes it usable: the audio extract, positioned to the second. Your assistant listens for three seconds instead of hunting through ten minutes of dictation. Checking a doubtful point goes from 4 minutes to 20 seconds.
The time this shifts, on the item you had costed: laying out a clinic letter took 60 % of the processing time — 22 minutes. It now takes 18 %, or 6 minutes 36, proofreading and approval included. Across 310 letters a week, that is 80 hours given back every week to your nine staff, in other words a full day for each of them.
What I propose: that the lexicon become a departmental document, reviewed once a year by the lead doctors. Over the first six weeks, 34 new abbreviations added themselves to it — precisely the ones a new member of staff takes six months to learn, and they will have them on day one. clinic-letter_laid-out-and-highlighted.pdfThe proposed document, its 2 highlights and their audio extract
⛓ Sourced · departmental lexicon (240 abbreviations), digital dictations, highlighted points across 4,030 clinic letters
And the letter to the family doctor? Today it has to be retyped afterwards, and it goes out when someone has the time.
It is ready at the same time as the clinic letter, with the department's letterhead and the correspondent's details taken from the patient record. It is waiting for a signature and nothing else — and over the last twelve weeks, 1,610 went out, not one without the doctor's named sign-off.
Secure health messaging is the encrypted channel reserved for health professionals: it replaces post and ordinary email for sending medical information. The letter goes out through it, and through it alone.
What the letter already contains: the form of address specific to the correspondent, the reason for referral, the conclusion and plan taken from the approved clinic letter, and the investigations requested with their deadlines.
What I add and nobody had time to add: where the family doctor has referred this service user before, a reminder of the previous consultation and its date — it is the first thing the correspondent looks for, and until now they looked for it in their own records.
The time this shifts: preparing a letter took 20 % of the processing time — 10 minutes. It now takes 8 %, or 4 minutes. 31 hours given back every week on that item alone.
And the delay, which is what the service user and the correspondent see: 6.4 days → 0.6 days. The doctor's signature is the condition of dispatch, and it is what gives the letter its standing with the correspondent and with any third party. You get both: signed, and out the same day.
What it changes for the service user, and it is the argument that carries before your supervisory board: the family doctor has the clinic letter before the service user rings to ask where things stand. Calls of the “my doctor hasn't received anything” kind went from 62 to 9 a week: 53 fewer calls on your switchboard, and 53 service users who never had to worry.
What I propose next: letters to colleagues, same template and same sign-off route. They account for 40 % of your dispatches, nobody had counted them, and they still run at the 6.4 days I have just brought down to 0.6 on the family doctor. letter-to-family-doctor_ready-to-sign.pdf1,610 letters in 12 weeks, 6.4 days → 0.6 days
⛓ Sourced · electronic patient record, secure health messaging log, incoming calls classified by reason
Secure health messaging is the encrypted channel reserved for health professionals: it replaces post and ordinary email for sending medical information. The letter goes out through it, and through it alone.
What the letter already contains: the form of address specific to the correspondent, the reason for referral, the conclusion and plan taken from the approved clinic letter, and the investigations requested with their deadlines.
What I add and nobody had time to add: where the family doctor has referred this service user before, a reminder of the previous consultation and its date — it is the first thing the correspondent looks for, and until now they looked for it in their own records.
The time this shifts: preparing a letter took 20 % of the processing time — 10 minutes. It now takes 8 %, or 4 minutes. 31 hours given back every week on that item alone.
And the delay, which is what the service user and the correspondent see: 6.4 days → 0.6 days. The doctor's signature is the condition of dispatch, and it is what gives the letter its standing with the correspondent and with any third party. You get both: signed, and out the same day.
What it changes for the service user, and it is the argument that carries before your supervisory board: the family doctor has the clinic letter before the service user rings to ask where things stand. Calls of the “my doctor hasn't received anything” kind went from 62 to 9 a week: 53 fewer calls on your switchboard, and 53 service users who never had to worry.
What I propose next: letters to colleagues, same template and same sign-off route. They account for 40 % of your dispatches, nobody had counted them, and they still run at the 6.4 days I have just brought down to 0.6 on the family doctor. letter-to-family-doctor_ready-to-sign.pdf1,610 letters in 12 weeks, 6.4 days → 0.6 days
⛓ Sourced · electronic patient record, secure health messaging log, incoming calls classified by reason
The waiting time for an endocrinology appointment is 94 days. We hear about it at every supervisory board, and we will get no extra doctor and no extra room.
You are short of slots far less than you think: 14.8 % of the endocrinology ones are lost, and I can tell you which, when, and from what cause.
Across the last 18 months of the diary:
· 9.1 % missed appointments with no cancellation — the service user does not come, and nobody could take the place.
· 4.3 % of slots released too late to be refilled — the cancellation arrives the day before, the waiting list is not called, the slot falls.
· 1.4 % of slots blocked by mistake and never reopened, including a weekly slot closed since a doctor left in March 2025: 62 consultations lost for a box nobody reopened.
What that weighs: 9.2 slots lost a week out of 62. Your 94-day queue contains a queue of empty slots, and that is the good news in this record: it can be recovered without buying anything.
The second point, and it explains the first: your departments' diary rules were written down nowhere. Your staff applied them from memory, and they varied from one person to the next. I have reconstructed them from 18 months of diaries actually kept: seven rules per specialty — duration by reason, slots reserved for first appointments, protected fallback slots, safety interval before an investigation, the doctor's theatre days.
And I do not stop at writing them: I have run them across your 18 months to tell you what each one is worth. “First appointments on Tuesday mornings”: +2.4 slots a week. “Ten-day safety interval before imaging”: 11 wasted consultations avoided for 0.8 slot a week — it returns eleven useful consultations, and you see that costed before you sign rather than after.
What I propose: that your nurse manager reads the seven endocrinology rules and corrects what needs correcting — they are hers, I have only read them in her diary and costed them. She signs one, it is in force that evening, and it is that signature that makes it defensible to the service user who challenges it. diary-rules_5-specialties.pdf7 rules per specialty, each costed over 18 months
⛓ Sourced · 18 months of the endocrinology diary, cancellation log, waiting list
Across the last 18 months of the diary:
· 9.1 % missed appointments with no cancellation — the service user does not come, and nobody could take the place.
· 4.3 % of slots released too late to be refilled — the cancellation arrives the day before, the waiting list is not called, the slot falls.
· 1.4 % of slots blocked by mistake and never reopened, including a weekly slot closed since a doctor left in March 2025: 62 consultations lost for a box nobody reopened.
What that weighs: 9.2 slots lost a week out of 62. Your 94-day queue contains a queue of empty slots, and that is the good news in this record: it can be recovered without buying anything.
The second point, and it explains the first: your departments' diary rules were written down nowhere. Your staff applied them from memory, and they varied from one person to the next. I have reconstructed them from 18 months of diaries actually kept: seven rules per specialty — duration by reason, slots reserved for first appointments, protected fallback slots, safety interval before an investigation, the doctor's theatre days.
And I do not stop at writing them: I have run them across your 18 months to tell you what each one is worth. “First appointments on Tuesday mornings”: +2.4 slots a week. “Ten-day safety interval before imaging”: 11 wasted consultations avoided for 0.8 slot a week — it returns eleven useful consultations, and you see that costed before you sign rather than after.
What I propose: that your nurse manager reads the seven endocrinology rules and corrects what needs correcting — they are hers, I have only read them in her diary and costed them. She signs one, it is in force that evening, and it is that signature that makes it defensible to the service user who challenges it. diary-rules_5-specialties.pdf7 rules per specialty, each costed over 18 months
⛓ Sourced · 18 months of the endocrinology diary, cancellation log, waiting list
We've tried reminders. People don't reply, and the missed-appointment rate never moved a point.
A reminder that offers nothing gets nothing. This one offers to hand the slot back in one gesture — and it is that gesture that took missed appointments from 14.8 % to 6.2 % in twelve weeks.
What changes, point by point:
· Two reminders, at 7 days and 2 days, not one the day before. The day-before reminder arrives too late for the slot to be refilled: that is the one that got nothing.
· A way to hand the slot back in one click or one word, without having to ring the office. 214 slots handed back in 12 weeks, against 31 before.
· The waiting list called within the hour after a slot is released. 176 of the 214 slots refilled, 122 of them the same day.
· The reminder rewritten in easy-read where the service user's record indicates it — one idea per sentence, no acronym, date and place at the top. Among those service users, non-attendance was 21 %; it is now 9 %.
And the appointment letter, prepared in the same movement as the booking: date, place and department, preparation instructions specific to the reason — fasting or not, prescription and current treatments to bring, ID and health card, how long to allow —, and a map to the right building. 16 % of your service users turned up at the wrong building: the figure was sitting in the front desk's day book, and nobody had added it up.
The result, the one the supervisory board is waiting for: 5 more endocrinology consultations a week, and the queue down from 94 to 61 days in twelve weeks. No extra slot, no extra doctor, with the diary you already had.
What I propose next: rheumatology, where the wait is 71 days and non-attendance runs at 12.4 %. The same work is worth about 3 consultations a week there, the seven rules are already written and costed, and all that is missing is your manager's signature. appointment-letters-and-reminders_12-weeks.pdf214 slots handed back, 176 refilled, 122 the same day
⛓ Sourced · reminder and cancellation log over 12 weeks, waiting list, front desk day book
What changes, point by point:
· Two reminders, at 7 days and 2 days, not one the day before. The day-before reminder arrives too late for the slot to be refilled: that is the one that got nothing.
· A way to hand the slot back in one click or one word, without having to ring the office. 214 slots handed back in 12 weeks, against 31 before.
· The waiting list called within the hour after a slot is released. 176 of the 214 slots refilled, 122 of them the same day.
· The reminder rewritten in easy-read where the service user's record indicates it — one idea per sentence, no acronym, date and place at the top. Among those service users, non-attendance was 21 %; it is now 9 %.
And the appointment letter, prepared in the same movement as the booking: date, place and department, preparation instructions specific to the reason — fasting or not, prescription and current treatments to bring, ID and health card, how long to allow —, and a map to the right building. 16 % of your service users turned up at the wrong building: the figure was sitting in the front desk's day book, and nobody had added it up.
The result, the one the supervisory board is waiting for: 5 more endocrinology consultations a week, and the queue down from 94 to 61 days in twelve weeks. No extra slot, no extra doctor, with the diary you already had.
What I propose next: rheumatology, where the wait is 71 days and non-attendance runs at 12.4 %. The same work is worth about 3 consultations a week there, the seven rules are already written and costed, and all that is missing is your manager's signature. appointment-letters-and-reminders_12-weeks.pdf214 slots handed back, 176 refilled, 122 the same day
⛓ Sourced · reminder and cancellation log over 12 weeks, waiting list, front desk day book
And urgent cases? We don't want a diary that runs itself and decides for us who goes first.
The diary is kept by your nurse manager, and every week I hand her the record she has never had time to produce: the 9.2 lost slots, one by one, with their cause and the rule that produced them.
Fallback slots are the ones your nurse manager keeps free each day for a situation that cannot wait the normal delay. They are protected: they stay out of every ordinary booking proposal, even when the queue is long and filling them would improve my figures. That is exactly why they exist — a diary that optimises always fills them, and that is precisely the failure your protection prevents, week after week.
The three things I do without being asked, and they stop there:
· I call the waiting list within the hour when a slot is released, in order of registration, with no exception. The order is written down and it depends on no judgement.
· I warn the nurse manager as soon as a queue crosses the threshold she has set, with the week's lost slots alongside.
· I reopen an offered slot if the service user does not confirm within 48 hours — and the reverse holds: if they confirm afterwards and the slot is still free, they get it back without having to call.
And here is what I prepare for her, fully written and costed, so that all she has to do is settle it:
· Appointment moves. When a doctor books a theatre day, I prepare the rebooking of the 14 service users affected, each on a slot that respects the constraints already recorded in their file, and the manager approves the list on one screen. 14 apology calls avoided, and 14 service users who learn their new date the same day instead of the following week.
· The rules to adjust, each with what it cost and returned in slots over the past month. She signs, it is in force that evening.
What stays entirely with your department, and it is what gives an appointment its value: where a service user stands in a queue is a public-service decision. It is set by a written rule of your manager and your doctors — I apply it to the letter, I measure it, and each month I tell you what it costs and what it returns.
One exception, and it comes from you: when a doctor marks a request “to be seen within eight days”, I place it on a fallback slot and I tell the nurse manager straight away. That is her rule, written in the diary for two years; I apply it, and each month I show her what it consumes.
What I propose: a thirty-minute diary review each month with the nurse manager — lost slots, thresholds crossed, costed rules to sign. The report is ready before the meeting. Over the quarter: 9,400 appointments handled, and every move started from a screen approved by a public servant.
✎ Framework · diary rules settled by the nurse manager, log of proposals and approvals
Fallback slots are the ones your nurse manager keeps free each day for a situation that cannot wait the normal delay. They are protected: they stay out of every ordinary booking proposal, even when the queue is long and filling them would improve my figures. That is exactly why they exist — a diary that optimises always fills them, and that is precisely the failure your protection prevents, week after week.
The three things I do without being asked, and they stop there:
· I call the waiting list within the hour when a slot is released, in order of registration, with no exception. The order is written down and it depends on no judgement.
· I warn the nurse manager as soon as a queue crosses the threshold she has set, with the week's lost slots alongside.
· I reopen an offered slot if the service user does not confirm within 48 hours — and the reverse holds: if they confirm afterwards and the slot is still free, they get it back without having to call.
And here is what I prepare for her, fully written and costed, so that all she has to do is settle it:
· Appointment moves. When a doctor books a theatre day, I prepare the rebooking of the 14 service users affected, each on a slot that respects the constraints already recorded in their file, and the manager approves the list on one screen. 14 apology calls avoided, and 14 service users who learn their new date the same day instead of the following week.
· The rules to adjust, each with what it cost and returned in slots over the past month. She signs, it is in force that evening.
What stays entirely with your department, and it is what gives an appointment its value: where a service user stands in a queue is a public-service decision. It is set by a written rule of your manager and your doctors — I apply it to the letter, I measure it, and each month I tell you what it costs and what it returns.
One exception, and it comes from you: when a doctor marks a request “to be seen within eight days”, I place it on a fallback slot and I tell the nurse manager straight away. That is her rule, written in the diary for two years; I apply it, and each month I show her what it consumes.
What I propose: a thirty-minute diary review each month with the nurse manager — lost slots, thresholds crossed, costed rules to sign. The report is ready before the meeting. Over the quarter: 9,400 appointments handled, and every move started from a screen approved by a public servant.
✎ Framework · diary rules settled by the nurse manager, log of proposals and approvals
Almost one file in two arrives incomplete on the day. We lose consultations over a forgotten prescription, and the service user waits another three months.
The completeness check now happens 10 days out rather than on the day, and that is all that changes: 10 days out, the service user can still act. Your 1,190 incomplete files have become 210.
What I read, and how: the documents arrive as photos, scans or post. I read them by character recognition — OCR recovers the text inside an image, including a photo taken on a phone and at an angle — and I check the presence, the legibility AND the date of each one, not just the presence. A prescription from 2023 is present and useless: a presence-only check counted it as good.
The record, appointment by appointment: 2,480 files checked over the quarter, 1,190 incomplete 10 days out — 48 %. The three gaps that make up the bulk of the figure: the prescription of current treatments (41 % of gaps), a recent imaging report (28 %), the referral letter from the family doctor (16 %).
What I do with it: each service user concerned receives, 10 days out, the exact list of what they are missing, in easy-read, with how to send it — rather than a message asking for “the missing documents”. A reminder that does not name the document gets nothing: that is what yours did, and it is why they never moved the figure.
The result: 210 files still incomplete on the day instead of 1,190 — 8.5 % instead of 48 %. And 34 consultations that would have been postponed took place, which counts twice: the slot is used, and the service user does not wait another three months.
What I prepare for the doctor, and it makes the decision immediate: for each of the 210 files still incomplete, they receive the day before the missing document, what it brings to that particular consultation, and what comparable consultations of the quarter gave without it. They settle it at a glance: see the service user anyway — which is what happens in the great majority of cases — or postpone, and I then rebook the appointment within the hour, on the first useful slot.
What I propose: extend the check to orthopaedic surgery clinics, where incompleteness at 10 days out reaches 61 % and where an incomplete file pushes back a theatre slot, not a consultation. The gain there is of a different order altogether, and the check is ready to start on Monday. completeness-check_2480-files.pdf48 % → 8.5 %, missing document by missing document
⛓ Sourced · documents filed to the patient record, log of postponed consultations, reminders sent
What I read, and how: the documents arrive as photos, scans or post. I read them by character recognition — OCR recovers the text inside an image, including a photo taken on a phone and at an angle — and I check the presence, the legibility AND the date of each one, not just the presence. A prescription from 2023 is present and useless: a presence-only check counted it as good.
The record, appointment by appointment: 2,480 files checked over the quarter, 1,190 incomplete 10 days out — 48 %. The three gaps that make up the bulk of the figure: the prescription of current treatments (41 % of gaps), a recent imaging report (28 %), the referral letter from the family doctor (16 %).
What I do with it: each service user concerned receives, 10 days out, the exact list of what they are missing, in easy-read, with how to send it — rather than a message asking for “the missing documents”. A reminder that does not name the document gets nothing: that is what yours did, and it is why they never moved the figure.
The result: 210 files still incomplete on the day instead of 1,190 — 8.5 % instead of 48 %. And 34 consultations that would have been postponed took place, which counts twice: the slot is used, and the service user does not wait another three months.
What I prepare for the doctor, and it makes the decision immediate: for each of the 210 files still incomplete, they receive the day before the missing document, what it brings to that particular consultation, and what comparable consultations of the quarter gave without it. They settle it at a glance: see the service user anyway — which is what happens in the great majority of cases — or postpone, and I then rebook the appointment within the hour, on the first useful slot.
What I propose: extend the check to orthopaedic surgery clinics, where incompleteness at 10 days out reaches 61 % and where an incomplete file pushes back a theatre slot, not a consultation. The gain there is of a different order altogether, and the check is ready to start on Monday. completeness-check_2480-files.pdf48 % → 8.5 %, missing document by missing document
⛓ Sourced · documents filed to the patient record, log of postponed consultations, reminders sent
And coding the procedures? That is what takes us the most time after typing, and it is where we get it wrong — in both directions.
3,780 pre-codings are on the table for the quarter, each justified by the sentence of the clinic letter that carries it. 9.4 % were refined by the doctor: I publish that figure because it is what measures my quality, and because every refinement feeds back into the next proposal — over the last six weeks it stands at 6.1 %.
Coding a procedure means translating what was done during the consultation into official references: it determines the establishment's activity income and feeds public activity statistics.
Where each proposal starts from: the clinic letter APPROVED by the doctor, never the raw letter. Coding from an unapproved text means coding from a text that can still change. And every proposed procedure carries the exact sentence that justifies it, with its line: the doctor reads the justification before the code, not the other way round.
The three figures of the quarter, and the third is the unexpected one:
· 4,030 consultations, 3,780 pre-codings proposed. The remaining 250 carried a new or combined situation: for each one I put up the two candidate codes, the sentence that separates them and the precise question to settle. All 250 were settled in 50 minutes across the whole quarter.
· 412 procedures actually carried out and never coded found in the clinic letters, of which the doctors retained 389. These are not extra procedures: they are procedures that had taken place and were recorded nowhere.
· 27 proposals I flagged myself as risking over-coding, before being asked. A pre-coding that corrected in one direction only would be an income tool; this one is an accuracy tool — and that is what puts you beyond the first objection an audit would raise.
And here is the gesture that is yours, and it is worth more than an automatism: the sign-off commits the establishment's billing, and it is what makes that billing defensible in an audit. So I propose to make it instant through a mandate, written, capped, dated and revocable in a word:
· Scope: outpatient procedures in the five specialties, excluding any new or amended procedure, which stays signed off one by one — and I hand it to you justified, with the sentence of the clinic letter to hand, in 40 seconds.
· Cap: batch sign-off of 40 procedures at most, one batch per clinic session.
· Term: until 31/12/2026, with no tacit renewal.
· Reversibility: every signed-off procedure stays amendable, dated and signed, and revoking the mandate undoes nothing already signed off.
What the mandate buys, costed: signing off the 3,780 pre-codings goes from 4 minutes per consultation to 25 seconds per procedure signed off in a batch — 190 hours given back over the quarter, and the doctor still sees every line with its justification alongside.
What I propose next: the same record, specialty by specialty, at your monthly coding review. A specialty whose refinement rate climbs back above 9.4 % signals a new procedure or a changing practice — two things that are settled in an hour when you see them coming. procedure-pre-coding_and-sign-off-mandate.pdf3,780 justified proposals, 389 procedures recovered, 27 over-codings removed
⛓ Sourced · approved clinic letters, log of coding changes by doctors, quarterly activity record
Coding a procedure means translating what was done during the consultation into official references: it determines the establishment's activity income and feeds public activity statistics.
Where each proposal starts from: the clinic letter APPROVED by the doctor, never the raw letter. Coding from an unapproved text means coding from a text that can still change. And every proposed procedure carries the exact sentence that justifies it, with its line: the doctor reads the justification before the code, not the other way round.
The three figures of the quarter, and the third is the unexpected one:
· 4,030 consultations, 3,780 pre-codings proposed. The remaining 250 carried a new or combined situation: for each one I put up the two candidate codes, the sentence that separates them and the precise question to settle. All 250 were settled in 50 minutes across the whole quarter.
· 412 procedures actually carried out and never coded found in the clinic letters, of which the doctors retained 389. These are not extra procedures: they are procedures that had taken place and were recorded nowhere.
· 27 proposals I flagged myself as risking over-coding, before being asked. A pre-coding that corrected in one direction only would be an income tool; this one is an accuracy tool — and that is what puts you beyond the first objection an audit would raise.
And here is the gesture that is yours, and it is worth more than an automatism: the sign-off commits the establishment's billing, and it is what makes that billing defensible in an audit. So I propose to make it instant through a mandate, written, capped, dated and revocable in a word:
· Scope: outpatient procedures in the five specialties, excluding any new or amended procedure, which stays signed off one by one — and I hand it to you justified, with the sentence of the clinic letter to hand, in 40 seconds.
· Cap: batch sign-off of 40 procedures at most, one batch per clinic session.
· Term: until 31/12/2026, with no tacit renewal.
· Reversibility: every signed-off procedure stays amendable, dated and signed, and revoking the mandate undoes nothing already signed off.
What the mandate buys, costed: signing off the 3,780 pre-codings goes from 4 minutes per consultation to 25 seconds per procedure signed off in a batch — 190 hours given back over the quarter, and the doctor still sees every line with its justification alongside.
What I propose next: the same record, specialty by specialty, at your monthly coding review. A specialty whose refinement rate climbs back above 9.4 % signals a new procedure or a changing practice — two things that are settled in an hour when you see them coming. procedure-pre-coding_and-sign-off-mandate.pdf3,780 justified proposals, 389 procedures recovered, 27 over-codings removed
⛓ Sourced · approved clinic letters, log of coding changes by doctors, quarterly activity record
Our assistants also receive test results. Can you tell the service user directly when a result is abnormal? That is what would gain us the most.
Telling a service user a result, no — and it is the only “no” in this demonstration. Here is what I do instead, which gives the service user the same service and does it markedly faster: 2.3 days become 41 minutes.
What blocks it, named precisely: announcing and interpreting a test result is a medical act, reserved to health professionals; performing it without holding the title is a criminal offence. This is not a design precaution, it is the law — and the route I have built around it gives the service user more than you were asking me for.
The lawful route, and it delivers more than you asked:
· I read every result as it arrives in secure health messaging, seven days a week, including the Saturday morning when your office is closed.
· I spot values outside the limits YOUR doctors have set — they are their limits, written by them, specialty by specialty; I compare each value against its own and name the gap, figure against figure.
· I alert the doctor on call within minutes, with the file already open: the last consultation, current treatment, previous results for comparison, and the service user's number ready to dial.
· And the letter to the family doctor is written in the same movement, so that the doctor's call and the correspondent's information go out together instead of three days apart.
The figure, and it is the argument: a critical result was seen by a doctor in 2.3 days on average — the time it took to climb the office's pile. It is now seen in 41 minutes. Over the quarter, 6 service users were called back the same day where they would have been called two or three days later.
It is not me who tells them: it is the doctor, and sooner. That is exactly what the service user wanted — they were not asking to be told by a machine, they were asking to be told quickly.
What I propose: that the limits be reviewed once a year by the five lead doctors, and that any alert not followed by a call be traced. Over the quarter: 118 alerts, 118 taken up by a doctor, 118 service users told by a physician — and a delay that no longer depends on the hour someone opens the pile. results-route_from-2-3-days-to-41-minutes.pdf118 alerts, 118 doctors engaged, 118 service users told by a physician
✎ Framework · limits set by the lead doctors, log of alerts and follow-up, secure health messaging
What blocks it, named precisely: announcing and interpreting a test result is a medical act, reserved to health professionals; performing it without holding the title is a criminal offence. This is not a design precaution, it is the law — and the route I have built around it gives the service user more than you were asking me for.
The lawful route, and it delivers more than you asked:
· I read every result as it arrives in secure health messaging, seven days a week, including the Saturday morning when your office is closed.
· I spot values outside the limits YOUR doctors have set — they are their limits, written by them, specialty by specialty; I compare each value against its own and name the gap, figure against figure.
· I alert the doctor on call within minutes, with the file already open: the last consultation, current treatment, previous results for comparison, and the service user's number ready to dial.
· And the letter to the family doctor is written in the same movement, so that the doctor's call and the correspondent's information go out together instead of three days apart.
The figure, and it is the argument: a critical result was seen by a doctor in 2.3 days on average — the time it took to climb the office's pile. It is now seen in 41 minutes. Over the quarter, 6 service users were called back the same day where they would have been called two or three days later.
It is not me who tells them: it is the doctor, and sooner. That is exactly what the service user wanted — they were not asking to be told by a machine, they were asking to be told quickly.
What I propose: that the limits be reviewed once a year by the five lead doctors, and that any alert not followed by a call be traced. Over the quarter: 118 alerts, 118 taken up by a doctor, 118 service users told by a physician — and a delay that no longer depends on the hour someone opens the pile. results-route_from-2-3-days-to-41-minutes.pdf118 alerts, 118 doctors engaged, 118 service users told by a physician
✎ Framework · limits set by the lead doctors, log of alerts and follow-up, secure health messaging
22 % of our calls go unanswered at peak times. Service users call back, and the problem feeds itself.
1,850 calls a week, all answered on the first ring, sorted into three families. Unanswered calls at peak times went from 22 % to 4 %.
The first thing the agent says, before anything else: that it is an artificial intelligence, and that the caller can ask for a public servant at any moment. That is not a configuration option: the European regulation on artificial intelligence has required, since 2 August 2026, that any person interacting with an AI system be informed. The announcement is made in one short sentence, before any question, and it opens the call rather than closing it: 3.1 % ask for a human, 96.9 % carry on.
The week's sorting:
· 1,180 recurring questions handled end to end — date and time of an appointment, what to bring, how to prepare for a test, the status of a request, opening hours and access. 5 minutes → 3 minutes, that is 10 % → 6 % of processing time, and above all zero waiting, at 8.30 as at 17.00.
· 470 calls routed to the right department with the reason already noted: whoever picks up knows why they are being called, and the service user does not tell their story twice.
· 200 calls passed to a public servant, record already open and with everything that has already been said.
What ALWAYS goes to a public servant, without exception and without the caller having to ask: a request for a result or a medical explanation · a complaint · any situation where the caller expresses distress, pain or anxiety · a request to cancel an operation appointment · anyone who asks a second time for a human. An automatic answer to someone who is unwell does more damage than a wait — it is the rule that weighed most in writing the 1,180 answers.
The time given back to your staff: 39 hours a week on recurring questions alone. And the essential is not there: it is the interruption that disappears. An assistant laying out a clinic letter was interrupted 23 times a day; they are now interrupted 4 times. A layout resumed after an interruption costs more than the layout itself, and that is what made your 22 minutes impossible to compress.
What I propose: that the 1,180 recurring answers be read once by an assistant-and-manager pair, then fixed. You change one with a word when a rule changes, and it applies to the very next call — that is what guarantees that all 1,180 service users of the week hear exactly the same answer. office-switchboard_1850-calls-sorted.pdf1,180 handled, 470 routed, 200 for a public servant
⛓ Sourced · 1,850 calls of the week classified by reason, transfer log, record of interruptions at the desk
The first thing the agent says, before anything else: that it is an artificial intelligence, and that the caller can ask for a public servant at any moment. That is not a configuration option: the European regulation on artificial intelligence has required, since 2 August 2026, that any person interacting with an AI system be informed. The announcement is made in one short sentence, before any question, and it opens the call rather than closing it: 3.1 % ask for a human, 96.9 % carry on.
The week's sorting:
· 1,180 recurring questions handled end to end — date and time of an appointment, what to bring, how to prepare for a test, the status of a request, opening hours and access. 5 minutes → 3 minutes, that is 10 % → 6 % of processing time, and above all zero waiting, at 8.30 as at 17.00.
· 470 calls routed to the right department with the reason already noted: whoever picks up knows why they are being called, and the service user does not tell their story twice.
· 200 calls passed to a public servant, record already open and with everything that has already been said.
What ALWAYS goes to a public servant, without exception and without the caller having to ask: a request for a result or a medical explanation · a complaint · any situation where the caller expresses distress, pain or anxiety · a request to cancel an operation appointment · anyone who asks a second time for a human. An automatic answer to someone who is unwell does more damage than a wait — it is the rule that weighed most in writing the 1,180 answers.
The time given back to your staff: 39 hours a week on recurring questions alone. And the essential is not there: it is the interruption that disappears. An assistant laying out a clinic letter was interrupted 23 times a day; they are now interrupted 4 times. A layout resumed after an interruption costs more than the layout itself, and that is what made your 22 minutes impossible to compress.
What I propose: that the 1,180 recurring answers be read once by an assistant-and-manager pair, then fixed. You change one with a word when a rule changes, and it applies to the very next call — that is what guarantees that all 1,180 service users of the week hear exactly the same answer. office-switchboard_1850-calls-sorted.pdf1,180 handled, 470 routed, 200 for a public servant
⛓ Sourced · 1,850 calls of the week classified by reason, transfer log, record of interruptions at the desk
Our service users don't understand our appointment letters. Some arrive fasting for nothing, others eat before a blood test.
The department's 340 appointment letters are rewritten in easy-read, tested with 24 service users, then signed off one by one by your ward manager. Explanation calls fell by 24 %.
Easy-read is a way of writing designed to be understood by everyone: one idea per sentence, everyday words, no unexplained acronym, and the most important information first.
What that changes on an appointment letter:
· “You must be fasting” becomes “Do not eat or drink anything after midnight. You may drink water.” It is the second sentence that was missing, and it is the one that produced your calls — and your postponed blood tests.
· Acronyms are spelled out at first use. Your letters carried 11 on average.
· The vital information moves to the top: date, time, building, what to bring. The rest follows.
What I copy verbatim, and it is what makes easy-read usable in healthcare: a dose, a date, a time, a place, a safety instruction. I rewrite the sentence around it; the instruction itself passes through intact. That rule is what allowed all 340 appointment letters to be reworked in three weeks — without it, every line would have had to be double-read.
The test, because an untested easy-read letter is merely a rewritten letter: 24 volunteer service users, including 9 over 75 and 4 non-French speakers. Every sentence misunderstood by 2 people or more was rewritten. 7 sentences out of 41 fell at that test — and that is the most useful figure of the lot: 7 sentences the department held to be clear were not.
Results measured over twelve weeks: explanation calls −24 % · service users arriving at the wrong building 16 % → 3 % · consultations cancelled on the day for preparation not followed: 19 → 4.
And the sign-off, because a rewritten letter that goes out without one is no longer defensible: every easy-read version is signed off by your ward manager before the first send, and the original version stays attached to the record. All 340 were signed off in three weeks, alongside the clinics, so the 7 sentences that fell at the test fell twice over: with the service users first, with the manager next. And the online request form meets the public-sector accessibility standard — screen reader, contrast, keyboard navigation. An appointment letter a partially sighted service user cannot read is a breach of equal treatment, not a layout detail.
What I propose: put the department's 62 standard letters through the same treatment, starting with the 6 that trigger 70 % of explanation calls. Those 6 are already written and tested: they are waiting for your reading, and on their own they are worth half of the gain still to be had. easy-read-appointment-letters_tested-and-signed-off.pdf7 sentences rewritten after testing, −24 % explanation calls
⛓ Sourced · 340 departmental appointment letters, comprehension test with 24 service users, calls classified by reason over 12 weeks
Easy-read is a way of writing designed to be understood by everyone: one idea per sentence, everyday words, no unexplained acronym, and the most important information first.
What that changes on an appointment letter:
· “You must be fasting” becomes “Do not eat or drink anything after midnight. You may drink water.” It is the second sentence that was missing, and it is the one that produced your calls — and your postponed blood tests.
· Acronyms are spelled out at first use. Your letters carried 11 on average.
· The vital information moves to the top: date, time, building, what to bring. The rest follows.
What I copy verbatim, and it is what makes easy-read usable in healthcare: a dose, a date, a time, a place, a safety instruction. I rewrite the sentence around it; the instruction itself passes through intact. That rule is what allowed all 340 appointment letters to be reworked in three weeks — without it, every line would have had to be double-read.
The test, because an untested easy-read letter is merely a rewritten letter: 24 volunteer service users, including 9 over 75 and 4 non-French speakers. Every sentence misunderstood by 2 people or more was rewritten. 7 sentences out of 41 fell at that test — and that is the most useful figure of the lot: 7 sentences the department held to be clear were not.
Results measured over twelve weeks: explanation calls −24 % · service users arriving at the wrong building 16 % → 3 % · consultations cancelled on the day for preparation not followed: 19 → 4.
And the sign-off, because a rewritten letter that goes out without one is no longer defensible: every easy-read version is signed off by your ward manager before the first send, and the original version stays attached to the record. All 340 were signed off in three weeks, alongside the clinics, so the 7 sentences that fell at the test fell twice over: with the service users first, with the manager next. And the online request form meets the public-sector accessibility standard — screen reader, contrast, keyboard navigation. An appointment letter a partially sighted service user cannot read is a breach of equal treatment, not a layout detail.
What I propose: put the department's 62 standard letters through the same treatment, starting with the 6 that trigger 70 % of explanation calls. Those 6 are already written and tested: they are waiting for your reading, and on their own they are worth half of the gain still to be had. easy-read-appointment-letters_tested-and-signed-off.pdf7 sentences rewritten after testing, −24 % explanation calls
⛓ Sourced · 340 departmental appointment letters, comprehension test with 24 service users, calls classified by reason over 12 weeks
And if a service user refuses to speak to a machine? Or if one of our staff worries about being measured by all this?
Two fair questions, and they have two clear, costed answers.
The service user first: they get a public servant immediately, without having to insist or justify themselves. 3.1 % of callers ask, and they get one in 40 seconds on average. Nobody should have to negotiate with a machine to reach a human being in a public service — and the call does not start over: the public servant receives what has already been said.
And equal treatment, which is the real question underneath: the order is order of arrival, held to the line, and the only exceptions are written departmental rules — the fallback slots, requests marked “to be seen within eight days” by a doctor. Each month I tell you how many service users each one moved ahead, and by how many days: rank is set by a rule somebody signed, never by a judgement I would pass on the person.
Your staff now. I produce the indicators your management decides to track — and I start by telling you, with figures, the one I advise against. Calls per member of staff: the day it becomes a tracked figure, it becomes a target; the target shortens the difficult calls — precisely the ones that needed to last — and the indicator stops measuring what it was thought to measure. Of your 1,850 calls in the week, the 200 passed to a public servant are the longest of all: they are the ones a volume target would cut first, and you would have lost the quality along with the instrument meant to watch over it.
That said, it is a choice, and it is entirely yours. If you want them, I produce them, and I bring at the same time the three conditions to be met: that the measurement be proportionate to its purpose, that staff be informed beforehand, and that the staff representative bodies be consulted before it is put in place. The file is ready and all it lacks is your signature — that signature is what will make the measurement defensible, and that is why it is yours to give.
What I measure in the meantime, and nobody disputes: the service, not the people. Calls answered, appointment waiting times, dispatch times for clinic letters, complete files. Four public-service indicators, and all four are improving.
What I propose: present those four indicators at the next staff meeting, together with the record of what the AI agent prepares and of who signs each decision. A team that has read that record uses it better, and defends it themselves when asked.
✎ Framework · log of requests for a public servant, departmental ordering rules, measurement scope settled with management
The service user first: they get a public servant immediately, without having to insist or justify themselves. 3.1 % of callers ask, and they get one in 40 seconds on average. Nobody should have to negotiate with a machine to reach a human being in a public service — and the call does not start over: the public servant receives what has already been said.
And equal treatment, which is the real question underneath: the order is order of arrival, held to the line, and the only exceptions are written departmental rules — the fallback slots, requests marked “to be seen within eight days” by a doctor. Each month I tell you how many service users each one moved ahead, and by how many days: rank is set by a rule somebody signed, never by a judgement I would pass on the person.
Your staff now. I produce the indicators your management decides to track — and I start by telling you, with figures, the one I advise against. Calls per member of staff: the day it becomes a tracked figure, it becomes a target; the target shortens the difficult calls — precisely the ones that needed to last — and the indicator stops measuring what it was thought to measure. Of your 1,850 calls in the week, the 200 passed to a public servant are the longest of all: they are the ones a volume target would cut first, and you would have lost the quality along with the instrument meant to watch over it.
That said, it is a choice, and it is entirely yours. If you want them, I produce them, and I bring at the same time the three conditions to be met: that the measurement be proportionate to its purpose, that staff be informed beforehand, and that the staff representative bodies be consulted before it is put in place. The file is ready and all it lacks is your signature — that signature is what will make the measurement defensible, and that is why it is yours to give.
What I measure in the meantime, and nobody disputes: the service, not the people. Calls answered, appointment waiting times, dispatch times for clinic letters, complete files. Four public-service indicators, and all four are improving.
What I propose: present those four indicators at the next staff meeting, together with the record of what the AI agent prepares and of who signs each decision. A team that has read that record uses it better, and defends it themselves when asked.
✎ Framework · log of requests for a public servant, departmental ordering rules, measurement scope settled with management
After a quarter, what has it given the department back?
1,950 hours over the quarter, that is 150 hours a week. And I give the calculation line by line, because it is always the total that gets argued with, never the calculation.
· Laying out clinic letters: 310 a week, 22 minutes → 6 minutes 36. Your scoping benchmark: 60 % → 18 %. 80 hours a week.
· Preparing letters to family doctors: 310 a week, 10 minutes → 4 minutes. That is 20 % → 8 %. 31 hours a week.
· Recurring service-user questions: 1,180 a week, 5 minutes → 3 minutes. That is 10 % → 6 %. 39 hours a week.
150 hours a week across a team of nine is a little under half of their repetitive administrative time — exactly the order of magnitude your scoping announced.
And I say straight away what that figure is: public-servant time given back to the service. No post cut, no post created — your nine assistants are nine. Your activity records say where those hours went: +41 % of time at the front desk, the resumption of walking older service users to the right building, and two half-days a week on the complex files that had been waiting for months.
What the service user saw, and this is what goes to the supervisory board:
· Clinic letter to the family doctor: 6.4 days → 0.6 days.
· Endocrinology waiting time: 94 → 61 days.
· Unanswered calls at peak times: 22 % → 4 %.
· Files incomplete on the day of the consultation: 48 % → 8.5 %.
· Critical result seen by a doctor: 2.3 days → 41 minutes.
And the line that speaks to your accounts without ever speaking of yield: 389 procedures actually carried out and until now uncoded were approved by the doctors, and 27 proposals that would have over-coded were removed before reaching sign-off. What improves here is the accuracy of coding in both directions — that is what holds up in an audit, and it is what makes your activity statistics faithful to what you actually do.
What I propose: that these five delays go into the annual activity report, alongside the activity figures. It is the surest way for the gain to stay visible the day it has become normal — and it will, within six months. quarterly-review_1950-hours-given-back.pdf3 costed items, 5 delays met, the calculation line by line
⛓ Sourced · office activity records, dispatch and diary logs, quarterly coding record
· Laying out clinic letters: 310 a week, 22 minutes → 6 minutes 36. Your scoping benchmark: 60 % → 18 %. 80 hours a week.
· Preparing letters to family doctors: 310 a week, 10 minutes → 4 minutes. That is 20 % → 8 %. 31 hours a week.
· Recurring service-user questions: 1,180 a week, 5 minutes → 3 minutes. That is 10 % → 6 %. 39 hours a week.
150 hours a week across a team of nine is a little under half of their repetitive administrative time — exactly the order of magnitude your scoping announced.
And I say straight away what that figure is: public-servant time given back to the service. No post cut, no post created — your nine assistants are nine. Your activity records say where those hours went: +41 % of time at the front desk, the resumption of walking older service users to the right building, and two half-days a week on the complex files that had been waiting for months.
What the service user saw, and this is what goes to the supervisory board:
· Clinic letter to the family doctor: 6.4 days → 0.6 days.
· Endocrinology waiting time: 94 → 61 days.
· Unanswered calls at peak times: 22 % → 4 %.
· Files incomplete on the day of the consultation: 48 % → 8.5 %.
· Critical result seen by a doctor: 2.3 days → 41 minutes.
And the line that speaks to your accounts without ever speaking of yield: 389 procedures actually carried out and until now uncoded were approved by the doctors, and 27 proposals that would have over-coded were removed before reaching sign-off. What improves here is the accuracy of coding in both directions — that is what holds up in an audit, and it is what makes your activity statistics faithful to what you actually do.
What I propose: that these five delays go into the annual activity report, alongside the activity figures. It is the surest way for the gain to stay visible the day it has become normal — and it will, within six months. quarterly-review_1950-hours-given-back.pdf3 costed items, 5 delays met, the calculation line by line
⛓ Sourced · office activity records, dispatch and diary logs, quarterly coding record
And what held out against you? We want that too, otherwise we won't believe you on the rest.
250 of the 4,030 clinic letters laid out were returned by doctors for substantive correction — 6.2 %. The figure is mine, it is published by me, I have found its three causes, and it stands today at 1.8 %.
The three causes, measured and closed:
· 178 of the 250 concerned respiratory medicine. Two reasons combined: the department uses 61 in-house abbreviations absent from the initial lexicon, and the dictation microphone in one consulting room clipped at the end of sentences. Nobody had connected that microphone to the typing errors: I brought it out by cross-referencing the error rate by consulting room with the hour of dictation, and the fault always fell at the same point in the sentence. The 61 abbreviations are in the lexicon, collected in one hour; the microphone has been replaced.
· 48 concerned letters dictated at the end of the day, longer and faster: an 11.4 % return rate against 4.1 % in the morning. Those dictations are now highlighted more generously — two extra checks are worth more than one point left unmarked.
· The last 24 are legitimate substantive corrections: the doctor sharpened their thinking on rereading. Those are the proof that proofreading works, and they are meant to stay.
The measured result over the last six weeks: 6.2 % → 1.8 %, and the remaining 1.8 % are, within a few cases, those legitimate rereadings.
So here is the floor I aim at, and why it is the right one: the level of doctors' rereadings — I am there. A letter reread by the doctor who dictated it always produces refinements, and that is living proof that your approval chain is turning: it is exactly what you will show at an accreditation visit.
What I propose: track this rate by specialty every month and look at it in the departmental review. A rise in one specialty signals a change of doctor, new equipment or a new type of clinic — three things that are settled in an hour when you see them the same month, as with the respiratory microphone. clinic-letter-quality_250-returns.pdf6.2 % → 1.8 %, the three causes and their correction
⛓ Sourced · doctors' returns for correction, error rate by consulting room and by hour of dictation, departmental lexicon
The three causes, measured and closed:
· 178 of the 250 concerned respiratory medicine. Two reasons combined: the department uses 61 in-house abbreviations absent from the initial lexicon, and the dictation microphone in one consulting room clipped at the end of sentences. Nobody had connected that microphone to the typing errors: I brought it out by cross-referencing the error rate by consulting room with the hour of dictation, and the fault always fell at the same point in the sentence. The 61 abbreviations are in the lexicon, collected in one hour; the microphone has been replaced.
· 48 concerned letters dictated at the end of the day, longer and faster: an 11.4 % return rate against 4.1 % in the morning. Those dictations are now highlighted more generously — two extra checks are worth more than one point left unmarked.
· The last 24 are legitimate substantive corrections: the doctor sharpened their thinking on rereading. Those are the proof that proofreading works, and they are meant to stay.
The measured result over the last six weeks: 6.2 % → 1.8 %, and the remaining 1.8 % are, within a few cases, those legitimate rereadings.
So here is the floor I aim at, and why it is the right one: the level of doctors' rereadings — I am there. A letter reread by the doctor who dictated it always produces refinements, and that is living proof that your approval chain is turning: it is exactly what you will show at an accreditation visit.
What I propose: track this rate by specialty every month and look at it in the departmental review. A rise in one specialty signals a change of doctor, new equipment or a new type of clinic — three things that are settled in an hour when you see them the same month, as with the respiratory microphone. clinic-letter-quality_250-returns.pdf6.2 % → 1.8 %, the three causes and their correction
⛓ Sourced · doctors' returns for correction, error rate by consulting room and by hour of dictation, departmental lexicon
Last question, and it is the real one: where does our service users' health data live?
It stays inside the establishment. I run on local inference on a machine at the hospital, and the patient record is copied nowhere.
Local inference means the model computes on your machine: the text of a dictation or a clinic letter is processed without crossing any outside network. If the establishment would rather not host a machine, the other route is a dedicated, isolated resource hosted in France, under French law — hosting health data on behalf of others requires, in France, a host holding HDS certification; that requirement is settled with you at scoping, before any go-live. In both cases, one resource per establishment: Vaubourg's health data stays Vaubourg's.
What that changes, file by file:
· No health data outside the European Union. Zero. Architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity, including where a US provider hosts in Europe — those texts target the company, not the location of its machines. It is the argument that sets you apart from every consumer tool on the market.
· Your dictations, letters and files train no model. What I read of Vaubourg serves Vaubourg.
· Encryption in transit and at rest, strong authentication, and role-based access — rights follow the job: a medical administrative assistant opens the clinic letters of their own division.
· Full access logging: who opened which record, and when. That log is what lets you answer, within minutes, a service user exercising their right of access, and your data protection officer.
· Medical correspondence goes through secure health messaging, and through it alone.
Governance, one line per decision: the doctor approves the clinic letter and the coding · the medical administrative assistant approves letters and appointment letters · the nurse manager settles the diary rules · management opens access rights. Over the quarter: 4,030 clinic letters, 1,610 letters, 9,400 appointments and 3,780 pre-codings, every one approved by a named public servant. That is how I am connected, and it is what makes each of those decisions reasoned, traced and open to challenge — which is what a public service must be able to demonstrate.
And here is the argument to keep for the end of your presentation: all of this can be proved in writing. I keep up to date the technical sheet your data protection officer and your supervisory authority ask for — hosting, subcontractors, retention periods, who accesses what, access log. It used to be requested once a year and take three days to rebuild; it comes out in two minutes.
What I propose: a one-hour compliance review each quarter with your data protection officer. The report is ready before the meeting and runs to four pages: it is all written down already. technical-framework_where-health-data-lives.pdfLocal inference or HDS, 0 transfer outside the EU in the target architecture, who signs what
✎ Framework · hosting architecture in France, HDS requirement settled at scoping, role matrix, access and approval logs
Local inference means the model computes on your machine: the text of a dictation or a clinic letter is processed without crossing any outside network. If the establishment would rather not host a machine, the other route is a dedicated, isolated resource hosted in France, under French law — hosting health data on behalf of others requires, in France, a host holding HDS certification; that requirement is settled with you at scoping, before any go-live. In both cases, one resource per establishment: Vaubourg's health data stays Vaubourg's.
What that changes, file by file:
· No health data outside the European Union. Zero. Architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity, including where a US provider hosts in Europe — those texts target the company, not the location of its machines. It is the argument that sets you apart from every consumer tool on the market.
· Your dictations, letters and files train no model. What I read of Vaubourg serves Vaubourg.
· Encryption in transit and at rest, strong authentication, and role-based access — rights follow the job: a medical administrative assistant opens the clinic letters of their own division.
· Full access logging: who opened which record, and when. That log is what lets you answer, within minutes, a service user exercising their right of access, and your data protection officer.
· Medical correspondence goes through secure health messaging, and through it alone.
Governance, one line per decision: the doctor approves the clinic letter and the coding · the medical administrative assistant approves letters and appointment letters · the nurse manager settles the diary rules · management opens access rights. Over the quarter: 4,030 clinic letters, 1,610 letters, 9,400 appointments and 3,780 pre-codings, every one approved by a named public servant. That is how I am connected, and it is what makes each of those decisions reasoned, traced and open to challenge — which is what a public service must be able to demonstrate.
And here is the argument to keep for the end of your presentation: all of this can be proved in writing. I keep up to date the technical sheet your data protection officer and your supervisory authority ask for — hosting, subcontractors, retention periods, who accesses what, access log. It used to be requested once a year and take three days to rebuild; it comes out in two minutes.
What I propose: a one-hour compliance review each quarter with your data protection officer. The report is ready before the meeting and runs to four pages: it is all written down already. technical-framework_where-health-data-lives.pdfLocal inference or HDS, 0 transfer outside the EU in the target architecture, who signs what
✎ Framework · hosting architecture in France, HDS requirement settled at scoping, role matrix, access and approval logs
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 medical administrative office
Each use corresponds to an agent we deploy. All of them work in support, subject to approval by the public officer.
Included in your agent
The 7 capabilities essential to this promise are included, at no extra cost.
Laying out the reports
Reports structured from the dictation or from notes, on the department's templates — proposed, for review and approval.
Letters & medical correspondence
Letters to the GP and to colleagues, headings and contact details taken from the record — ready to sign.
Appointments & notices
Organising appointments, reminders and notices, respecting the department's diary rules.
Pre-coding procedures
Preparing the pre-coding of procedures from the report, for the professional to approve.
Reading the file's documents
Extraction and completeness checking of the administrative documents in the patient record (OCR, evidence).
Administrative information for patients
Answer recurring questions: documents to bring, preparing for a test, the status of a request.
Plain language & easy-read draft
Preparation of a plain-language version and of a draft easy-read (FALC) transcription, submitted for human validation.
Controls and safeguards
These 6 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
Strictly separate administration, information and care decisions
Protect health data and restrict access
Human validation, quality monitoring and explicit limits
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 the patient record and appointment diary without double entry or breaks in the care pathway
Other needs our agents cover
Each card says where the matching agent stands: available, on quote, or still being architected.
Medical administrative assistant (letters, coding)
Reinforced confidentiality.
On quote View the agent page →Hospital reception and admissions (admissions office)
Strictly in support (administrative). Health data: the HDS requirement settled at scoping.
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 front desk win back?
By automating the laying out of reports and the preparation of letters and notices, a medical office can aim to halve its administrative time on repetitive tasks — reinvested in welcoming and supporting the patient.
Laying out a consultation report
Preparing a letter to the GP
Answering a recurring question from a member of the public
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 medical administrative agent (reports, letters, appointments, pre-coding), 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.
This agent is priced with you, not online. We are adjusting its scope at the moment, and online subscription stays closed while we do. Tell us what you need: we will come back to you with a price. Request a quote
Our commitment
Four guarantees that matter to a public health service
Health data never leaves the establishmentLocal inference or an isolated resource hosted in France, the HDS requirement settled at scoping; no patient data entrusted to a foreign third party.
Data in France, under French lawNative location and minimisation for health data; architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The public officer keeps the decisionThe AI agent produces reports, letters and pre-coding that can be checked; no approval and no medical act is automated.
Human oversight & traceabilityMonitoring, updates and logging of access: compliant with the requirements of the AI Act.
Frequently asked questions
Your questions, our answers
Is health data protected?
Yes: hosting in France, an isolated resource and recorded access, or local inference on a machine at the establishment. No data travels outside the European Union, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. In France, hosting health data on behalf of others is subject to HDS certification of the host: that requirement is settled with you at scoping, before any go-live.
Does the agent make a diagnosis?
No: it handles administrative tasks only — laying out reports, letters, appointments, pre-coding. The medical act and the final coding stay with the professional. The AI agent assists, the public officer decides.
Does it integrate with the establishment's software?
Yes, through the interfaces of the patient record and the secure health messaging. The agent connects to your existing tools and complements them, with no migration and no change of software imposed.
Can AI really lay out a reliable report?
Yes, to a large extent. The agent structures the report from the dictation or from notes on the department's templates, prepares the associated letters and highlights the points to check (an ambiguous term, a dosage). The secretary moves from typing to reviewing and approving — quicker, and worth more.
How is human oversight guaranteed?
Nothing is sent, coded or transmitted without the public officer's approval. The agent produces material that can be checked, logs its actions and stays strictly in support — in line with the AI Act's requirements on human oversight.
Do you need a large hospital to take this on?
No. The offer suits a practice or a health centre as well as a hospital department or a public establishment, and is rolled out gradually, one priority use at a time.
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, then a phase of design, integration and testing before going live and training the officers.
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 — are passed on at actual cost, with no margin, outside the subscription.
Related roles
Other professions in public health and front-of-house
Let's talk
Let's size up the potential in your school
15 minutes to identify the most useful use — hosted in France, the HDS requirement settled at scoping, supervised, with no commitment.