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
● B2B offer — Health & medical secretarial work
Administrative triage: your protocol applied to every request
Directing a request to the right practitioner, the right type of consultation and the right timeframe follows a protocol your practitioners wrote. Your agent applies that protocol to the requests received, gathers the administrative information it calls for and proposes the corresponding referral. It makes no diagnosis and assesses no symptom: that rests exclusively with health professionals. Hosted in France, in an environment suited to health data.
Hosted in France
Health data protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Administrative triage · hosted in France
A patient is asking for an appointment, with no further detail.
Reception protocol applied: the administrative information it calls for is gathered — reason as stated, usual practitioner, how long they have been a patient, availability.
The referral proposed is the one your protocol associates with this combination.
No clinical judgement is expressed.
🔗 Sourced · the practice's reception protocol
The referral proposed is the one your protocol associates with this combination.
No clinical judgement is expressed.
🔗 Sourced · the practice's reception protocol
And if the request falls outside the protocol?
It is passed to the secretarial team, with the information gathered and a note that the protocol does not cover it.
Any question calling for clinical judgement is put to the practitioner: it is a medical act.
✎ Support · passed to the health professional
Any question calling for clinical judgement is put to the practitioner: it is a medical act.
✎ Support · passed to the health professional
Local inference · no data outside the EU
Health 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 medical secretarial team decidesThe agent prepares, never rules
✦ In brief
A Blue Lemon Agent triage agent applies the reception protocol written by your practitioners, gathers the administrative information it calls for and proposes the corresponding referral. It makes no diagnosis and assesses no symptom; any request outside the protocol is passed on. It runs on local inference or is hosted in France in an environment suited to health data, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. Your teams write to it from Microsoft Teams, Slack or their email, and your patients reach it on WhatsApp Business, your website chat or email — with no account to create and nothing to install. 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
6
medical secretarial uses ready to deploy
0
decision taken without human approval
These figures describe our offer, not results measured at a client. How large the gain is on your number of practitioners and reasons covered by the protocol is confirmed by a pilot.
The context
What does an AI agent bring to the administrative reception of your patients?
A reception that is always available and faithful to the protocol saves the secretarial team time and spares the patient a wait for administrative information.
! The issue
Directing a request combines administrative information and a protocol your practitioners wrote. The gathering is repetitive, applying the protocol is mechanical — which is exactly what an agent does faithfully, at any hour. What calls for clinical judgement remains entirely a matter for health professionals.
✓ Our answer
The secretarial team receives requests already filled in and directed according to the protocol, and spends its time on reception and on the unusual cases. The line is clear: no symptom is assessed, no diagnosis is made, and any request outside the protocol is passed on. Local inference or an isolated resource hosted in France, in an environment suited to health data.
The decisive point
Your patients' health data: sovereignty & compliance
Health data belongs to the most protected categories under the GDPR. Here is how the architecture of our agents complies.
Local inference
The agent can run on a machine belonging to your organisation: no health data leaves the network.
Hosting in France
Otherwise, a dedicated and isolated resource hosted in France, under French law — your appointment requests and your reception protocols: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
For your patients' 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.
Isolated resource
No pooling: an environment strictly dedicated to your practice and its practitioners' protocol.
No medical act, by design
The agent gathers information and applies an administrative protocol; encryption, role-based access, logging and hosting suited to health data.
AI Act: governed deployment
The agent is strictly in support; no diagnosis, no assessment of a symptom, no care decision; 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.
- 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 all health data, as a matter of principle, SecNumCloud and reinforced hosting are options depending on your requirements. A single architecture is designed to answer both the GDPR and extraterritorial exposure. Designed for deployment in line with the GDPR and the AI Act, after the processing, roles and context-specific risks have been assessed.
Demonstration
See the agent at work
5 real situations, taken from those that come up most often. Pick one: the exchange unfolds as it would in your organisation.
A scripted demonstration. These exchanges show how the agent behaves — its sources, its refusals, what it leaves to your teams. Nothing is sent from this page, no model is queried here, and the matters named are fictional. That is precisely what we promise your data.
The behaviours shown here — monitoring, automation rules, routing and reminders — are configured with you during deployment, from your tools, your rules and your thresholds.
The architecture points named in these exchanges — location, local execution, isolation, encryption, role-based access, logging — are not a guarantee attached to the demonstration: they are those of the architecture set out in your quotation, and verified before commissioning.
The company in this demonstration
Fictional companyMaison de santé des Quatre-Vents — multi-professional primary care centre
- Sector
- Primary care — general practice, midwifery, physiotherapy, nursing
- Headcount
- 16 people, including 11 practitioners (5 GPs, 2 midwives, 2 physiotherapists, 2 nurses), 3 medical secretaries, 1 front-desk coordinator and 1 administrative officer
- Patient base
- 9,400 patients across three towns, including 1,100 on long-term condition schemes
- Volume
- 190 appointment requests per working day — 96 by phone, 71 through the online portal, 23 by secure messaging; 11,970 over the quarter
- Tools in place
- Practice software (shared diary and patient record), online booking portal, telephone switchboard, secure health messaging — the agent plugs into them, nothing is replaced
- Who decides
- The lead physician writes and revises the intake protocol; the front-desk coordinator approves the proposed referrals; every clinical judgement goes to a practitioner, without exception
- Room for improvement
- Three stopwatches taken from the centre's own records: gathering administrative information eats 55 % of the time of an incoming request, applying the protocol 30 % of a request to be directed, formatting 25 % of a request passed to a practitioner; at the two peak hours, 41 % of calls go unanswered; a request filed after 6 p.m. waits until the next morning
Quatre-Vents wants to give front-desk time back to the patients standing at the counter. The agent runs on local inference on a machine at the centre and plugs into the practice software, the booking portal, the switchboard and the secure health messaging: it gathers, applies the protocol written by the practitioners, presents the figures — the centre signs. The exchanges below cover one quarter, from writing the protocol down to the measured review.
This company, its figures and the exchanges that follow were invented for the demonstration. They illustrate a common situation; they describe no real client.
Administrative triage · hosted in France
I have read your last three years of requests — 18,400 — and the three pages of protocol pinned up at the front desk. I draw from them a gap your own records did not show, and it is that gap which decides the gain.
The intake protocol is the rule your practitioners wrote: it ties a reason stated by a patient to a practitioner, to a type of consultation and to a timeframe. It is administrative, and it belongs to you.
The gap, measured between two populations of your own requests: of the 40 most frequent stated reasons, 34 already received the same referral — same reason, same referral, whoever picked up the phone. The other 6 received two, depending on the day and on the secretary, and those six carry 31 % of your requests.
What I did with that, and it is done: the 34 constants are written down — 34 reasons, 11 practitioners, 4 types of consultation, 3 timeframes, and for each of them the exact list of information to gather. This is your own practice set down clearly: I read it here, I did not invent it.
The other 6, I hand back to you ready to settle in ten minutes: for each one, the reason in the patients' own words, the two referrals observed, their three-year count — 1,940 against 812 on repeat prescriptions alone — and what each choice would cost in waiting time. The lead physician signs six lines, I apply all six from the very next request.
What the first week gives back: gathering administrative information used to eat 55 % of the time of an incoming request; it now eats 9 %.
The step I propose: the lead physician's six answers, then a read-through of the 34 reasons by the coordinator. Every line can be corrected in a word, and the correction applies from the very next request. intake-protocol_34-reasons.pdf34 consistent reasons, 6 questions for the lead physician
⛓ Sourced · 18,400 requests over 3 years, intake protocol pinned up at the front desk, referral log
The intake protocol is the rule your practitioners wrote: it ties a reason stated by a patient to a practitioner, to a type of consultation and to a timeframe. It is administrative, and it belongs to you.
The gap, measured between two populations of your own requests: of the 40 most frequent stated reasons, 34 already received the same referral — same reason, same referral, whoever picked up the phone. The other 6 received two, depending on the day and on the secretary, and those six carry 31 % of your requests.
What I did with that, and it is done: the 34 constants are written down — 34 reasons, 11 practitioners, 4 types of consultation, 3 timeframes, and for each of them the exact list of information to gather. This is your own practice set down clearly: I read it here, I did not invent it.
The other 6, I hand back to you ready to settle in ten minutes: for each one, the reason in the patients' own words, the two referrals observed, their three-year count — 1,940 against 812 on repeat prescriptions alone — and what each choice would cost in waiting time. The lead physician signs six lines, I apply all six from the very next request.
What the first week gives back: gathering administrative information used to eat 55 % of the time of an incoming request; it now eats 9 %.
The step I propose: the lead physician's six answers, then a read-through of the 34 reasons by the coordinator. Every line can be corrected in a word, and the correction applies from the very next request. intake-protocol_34-reasons.pdf34 consistent reasons, 6 questions for the lead physician
⛓ Sourced · 18,400 requests over 3 years, intake protocol pinned up at the front desk, referral log
Writing down who sees which reason means deciding in our place. And our patients will ask whether a machine is directing them.
Your protocol does the directing; I apply it identically at 3 a.m. and at 11 a.m. And the answer to give your patients, I have written it: three lines, attached to this message.
What I do, operation by operation: I gather administrative information, I compare it to the rules your practitioners wrote, and I propose the referral those rules attach to that case. This is rule application. Your protocol says “repeat prescription, treatment unchanged → registered GP, short consultation, within fifteen days”: that is exactly what I propose, with the protocol line displayed next to the result — your coordinator sees the rule at the same time as the proposal.
Where the law draws a line, it draws it for everyone, and here it works in your favour: making a diagnosis, assessing a symptom, telling a person whether their situation is serious, deciding on care are medical acts reserved for health professionals. All my rules start from an administrative reason, from the patient's history and from the usual practitioner, and that dictionary can be read line by line by whoever asks for it. A referral produced by a rule your practitioners wrote and signed stands up in front of a patient, in front of your insurer and in front of a judge — which no “intelligent” referral could ever do. It is also what makes the roll-out short: nothing to get certified, no procedure to run, a few weeks and the agent is live.
And here is what that line leaves me, which is the bulk of it: the intake, the comparison with the protocol, the formatting, the routing and the time-stamped trail. 90 % of what a request costs is administrative — that 90 % I take in full, around the clock and identically every time.
What I propose: pin this page up at the front desk and attach it to the information you give patients. A question answered in advance costs less than a question endured, and this one will be asked. boundary_what-the-agent-does-and-does-not.pdfThe exact split, line by line
✎ Framework · design split, reviewed with the lead physician
What I do, operation by operation: I gather administrative information, I compare it to the rules your practitioners wrote, and I propose the referral those rules attach to that case. This is rule application. Your protocol says “repeat prescription, treatment unchanged → registered GP, short consultation, within fifteen days”: that is exactly what I propose, with the protocol line displayed next to the result — your coordinator sees the rule at the same time as the proposal.
Where the law draws a line, it draws it for everyone, and here it works in your favour: making a diagnosis, assessing a symptom, telling a person whether their situation is serious, deciding on care are medical acts reserved for health professionals. All my rules start from an administrative reason, from the patient's history and from the usual practitioner, and that dictionary can be read line by line by whoever asks for it. A referral produced by a rule your practitioners wrote and signed stands up in front of a patient, in front of your insurer and in front of a judge — which no “intelligent” referral could ever do. It is also what makes the roll-out short: nothing to get certified, no procedure to run, a few weeks and the agent is live.
And here is what that line leaves me, which is the bulk of it: the intake, the comparison with the protocol, the formatting, the routing and the time-stamped trail. 90 % of what a request costs is administrative — that 90 % I take in full, around the clock and identically every time.
What I propose: pin this page up at the front desk and attach it to the information you give patients. A question answered in advance costs less than a question endured, and this one will be asked. boundary_what-the-agent-does-and-does-not.pdfThe exact split, line by line
✎ Framework · design split, reviewed with the lead physician
Agreed on the principle. In front-desk hours, what does it give us?
Ten hours per working day, across your three front-desk posts — and here are the three stopwatches that make up that figure, each measured on your own records.
Three separate stopwatches, on three different populations of requests: each one is read on its own, and each one can be re-checked here in a week of records.
· Gathering administrative information, on an incoming request: 55 % of the time → 9 %.
· Applying the referral protocol, on a request that calls for a referral: 30 % → 6 %.
· Formatting the request, on a request passed to a practitioner: 25 % → 5 %.
The figure that sums them up, and the one to keep: a complete request used to cost 4 minutes 10 of front-desk time; it now costs 1 minute. On 190 requests a day, that is 10 hours given back every working day, or 3 h 20 per secretary — the equivalent of one and a half posts returned to the patients who are there, at the counter.
What that time became here, and you are the one who decided it: your three secretaries now call back the patients whose appointment has been moved — 117 call-backs over the quarter, which nobody had time to make. Another centre would have turned it into extra slots; I had put both uses in front of you with figures, and you chose.
The next step I propose: two weeks on the 34 written reasons only, protocol pinned up, leaving the 6 pending questions untouched. At the end of the fortnight you will have the 10 hours measured here — and the 6 reasons enter the circuit as soon as the lead physician has signed. three-stopwatches_what-a-request-costs.pdf4 min 10 → 1 min, 10 hours a day
⛓ Sourced · 6 weeks of front-desk time records, request log, practice software diary
Three separate stopwatches, on three different populations of requests: each one is read on its own, and each one can be re-checked here in a week of records.
· Gathering administrative information, on an incoming request: 55 % of the time → 9 %.
· Applying the referral protocol, on a request that calls for a referral: 30 % → 6 %.
· Formatting the request, on a request passed to a practitioner: 25 % → 5 %.
The figure that sums them up, and the one to keep: a complete request used to cost 4 minutes 10 of front-desk time; it now costs 1 minute. On 190 requests a day, that is 10 hours given back every working day, or 3 h 20 per secretary — the equivalent of one and a half posts returned to the patients who are there, at the counter.
What that time became here, and you are the one who decided it: your three secretaries now call back the patients whose appointment has been moved — 117 call-backs over the quarter, which nobody had time to make. Another centre would have turned it into extra slots; I had put both uses in front of you with figures, and you chose.
The next step I propose: two weeks on the 34 written reasons only, protocol pinned up, leaving the 6 pending questions untouched. At the end of the fortnight you will have the 10 hours measured here — and the 6 reasons enter the circuit as soon as the lead physician has signed. three-stopwatches_what-a-request-costs.pdf4 min 10 → 1 min, 10 hours a day
⛓ Sourced · 6 weeks of front-desk time records, request log, practice software diary
Our most frequent case by far: “I would like an appointment”, and nothing else. What do you do with it?
I turn it into a complete, directed request, placed in front of your coordinator, in 55 seconds — without calling the patient back once.
What I gather, and nothing more, because that is what your protocol provides for:
· The reason as she states it, copied word for word, in quotation marks: “I would like an appointment with my doctor, it is about my prescription”. It reaches the practitioner in the patient's own words — rephrasing would already be interpreting.
· Her usual practitioner, found in the practice software: Dr Samuel Renaudot, last consultation on 14 February.
· How long she has been a patient here: with the centre since 2019, 11 consultations.
· The question that separates two consultations: “has your treatment changed since the last prescription?” — answer no.
· Her real availability: three slots she gave, not a ticked box.
· The administrative documents the reason calls for: health card up to date, registered GP declared, long-term condition scheme where applicable — the ongoing cover for a recognised illness, which opens its own administrative circuit.
· The channel and time of arrival, for the trail.
The referral I propose: registered GP, short consultation, within fifteen days, with three genuinely free slots — Thursday 2.20 p.m., Friday 9.10 a.m., the following Tuesday 8.40 a.m., and the protocol line displayed alongside.
What the patient gains: an answer on Tuesday at 9.41 p.m. instead of a call-back the next day at 11 a.m. on a busy line.
What I propose: read the attached request, field by field. A field you find missing is added to the protocol in a minute and applies to the 11,970 requests of the next quarter. request-as-presented_full-example.pdf7 fields, 1 referral, 1 protocol line quoted
⛓ Sourced · the centre's intake protocol, practice software, shared diary of the 11 practitioners
What I gather, and nothing more, because that is what your protocol provides for:
· The reason as she states it, copied word for word, in quotation marks: “I would like an appointment with my doctor, it is about my prescription”. It reaches the practitioner in the patient's own words — rephrasing would already be interpreting.
· Her usual practitioner, found in the practice software: Dr Samuel Renaudot, last consultation on 14 February.
· How long she has been a patient here: with the centre since 2019, 11 consultations.
· The question that separates two consultations: “has your treatment changed since the last prescription?” — answer no.
· Her real availability: three slots she gave, not a ticked box.
· The administrative documents the reason calls for: health card up to date, registered GP declared, long-term condition scheme where applicable — the ongoing cover for a recognised illness, which opens its own administrative circuit.
· The channel and time of arrival, for the trail.
The referral I propose: registered GP, short consultation, within fifteen days, with three genuinely free slots — Thursday 2.20 p.m., Friday 9.10 a.m., the following Tuesday 8.40 a.m., and the protocol line displayed alongside.
What the patient gains: an answer on Tuesday at 9.41 p.m. instead of a call-back the next day at 11 a.m. on a busy line.
What I propose: read the attached request, field by field. A field you find missing is added to the protocol in a minute and applies to the 11,970 requests of the next quarter. request-as-presented_full-example.pdf7 fields, 1 referral, 1 protocol line quoted
⛓ Sourced · the centre's intake protocol, practice software, shared diary of the 11 practitioners
And when the patient pours out her symptoms in detail? Because people do that, all the time.
The text reaches the practitioner intact and time-stamped, in 1 minute 40, record already open — and it is that fidelity which makes the handover useful.
What I do with the text: it is copied word for word, in quotation marks, time-stamped, into the field “reason as stated by the patient”. Complete, exact, dated. Your practitioner reads what the patient wrote, not what a piece of software retained from it.
What fires my rules, checkable line by line: the administrative reason, the patient's history and the usual practitioner. 18 fields are open to the agent; 9 stay in the record, in the practitioners' hands alone, even though they are technically reachable — history, treatment detail, test results, correspondence with specialists. Role-based access makes rights follow the job: your administrative officer opens the funding files, the practitioners keep the text addressed to them. It is that second list which proves minimisation, and it is the one you will be asked for.
What happens then, and this is the service delivered: as soon as the text goes beyond the administrative scope, the request leaves the automatic circuit and goes immediately to the destination your protocol prescribes, with the full text, the seven fields already gathered and the record open. The practitioner receives a complete request, not an empty alert.
The figure: 1,480 requests contained clinical free text over the quarter; they left in 1 minute 40 on average, against 2 h 50 when they waited in the pile.
What I propose: that the lead physician reads the two attached lists — the 18 open fields and the 9 left in the record — and that he opens whichever field he wants: the list is his, and it changes in a minute. fields-gathered_and-those-deliberately-closed.pdf18 fields open, 9 fields deliberately closed
✎ Framework · dictionary of gathered fields, quarterly handover log, intake protocol
What I do with the text: it is copied word for word, in quotation marks, time-stamped, into the field “reason as stated by the patient”. Complete, exact, dated. Your practitioner reads what the patient wrote, not what a piece of software retained from it.
What fires my rules, checkable line by line: the administrative reason, the patient's history and the usual practitioner. 18 fields are open to the agent; 9 stay in the record, in the practitioners' hands alone, even though they are technically reachable — history, treatment detail, test results, correspondence with specialists. Role-based access makes rights follow the job: your administrative officer opens the funding files, the practitioners keep the text addressed to them. It is that second list which proves minimisation, and it is the one you will be asked for.
What happens then, and this is the service delivered: as soon as the text goes beyond the administrative scope, the request leaves the automatic circuit and goes immediately to the destination your protocol prescribes, with the full text, the seven fields already gathered and the record open. The practitioner receives a complete request, not an empty alert.
The figure: 1,480 requests contained clinical free text over the quarter; they left in 1 minute 40 on average, against 2 h 50 when they waited in the pile.
What I propose: that the lead physician reads the two attached lists — the 18 open fields and the 9 left in the record — and that he opens whichever field he wants: the list is his, and it changes in a minute. fields-gathered_and-those-deliberately-closed.pdf18 fields open, 9 fields deliberately closed
✎ Framework · dictionary of gathered fields, quarterly handover log, intake protocol
Our coordinator will not read 190 requests a day, one by one. Nobody can.
She approves 190 in 22 minutes, because they arrive sorted and only 4 call for a decision from her.
What she has in front of her at 8.30 a.m.:
· 151 protocol-compliant requests — written reason, consistent referral, slots available. Batch approval, one gesture, 6 minutes, and any one of them opens up to show the line applied.
· 28 requests to look at: a slot is missing, a document is absent, the usual practitioner is on leave. The fallback is already proposed alongside, drawn from the protocol. 11 minutes.
· 7 requests outside the protocol, already sent to their destination: they are listed for information. 1 minute.
· 4 judgement calls — and those are exactly what your coordinator is paid for. 4 minutes.
The two stopwatches of this stage: protocol application goes from 30 % to 6 % of the time of a request to be directed, formatting from 25 % to 5 % of a request passed on. These are the two most mechanical, and therefore the two easiest to give back to you.
The guarantee that holds it all together: of the 8,300 referrals proposed over the quarter, 8,300 carry a person's signature, with their name, their time and the protocol line applied. Booking in the diary is a gesture of your front desk — that is what makes every appointment defensible, and that is what upholds the “0 decision without human approval” you will display.
What I propose next: two presentations rather than one, at 8.30 a.m. and 2 p.m. 62 requests arrive between 9 a.m. and 1 p.m. and wait until the next morning; two passes return them same-day. The coordinator will say whether she prefers two short sittings or one long one. morning-batch_190-requests.pdf151 as a batch, 28 to look at, 4 judgement calls
⛓ Sourced · quarterly approval log, time-stamps of incoming requests, shared diary
What she has in front of her at 8.30 a.m.:
· 151 protocol-compliant requests — written reason, consistent referral, slots available. Batch approval, one gesture, 6 minutes, and any one of them opens up to show the line applied.
· 28 requests to look at: a slot is missing, a document is absent, the usual practitioner is on leave. The fallback is already proposed alongside, drawn from the protocol. 11 minutes.
· 7 requests outside the protocol, already sent to their destination: they are listed for information. 1 minute.
· 4 judgement calls — and those are exactly what your coordinator is paid for. 4 minutes.
The two stopwatches of this stage: protocol application goes from 30 % to 6 % of the time of a request to be directed, formatting from 25 % to 5 % of a request passed on. These are the two most mechanical, and therefore the two easiest to give back to you.
The guarantee that holds it all together: of the 8,300 referrals proposed over the quarter, 8,300 carry a person's signature, with their name, their time and the protocol line applied. Booking in the diary is a gesture of your front desk — that is what makes every appointment defensible, and that is what upholds the “0 decision without human approval” you will display.
What I propose next: two presentations rather than one, at 8.30 a.m. and 2 p.m. 62 requests arrive between 9 a.m. and 1 p.m. and wait until the next morning; two passes return them same-day. The coordinator will say whether she prefers two short sittings or one long one. morning-batch_190-requests.pdf151 as a batch, 28 to look at, 4 judgement calls
⛓ Sourced · quarterly approval log, time-stamps of incoming requests, shared diary
There will always be requests that fit no box. That is where we lose time, and sometimes more than time.
1,940 over the quarter — 16 % of your requests — and all 1,940 left, to seven destinations your protocol had already provided for.
The seven families, their volume and their destination:
· Clinical free text — 1,480 — to the practitioner concerned, full text, record open.
· Reason absent from the protocol — 194 — to the front desk, with an explicit note that the protocol does not cover it: your front desk receives a clean decision to make, rather than a guess to undo.
· Complex administrative request — 121 — funding, prior authorisation, transport: to the administrative officer.
· Patient unknown to the centre — 66 — to the front desk, with the waiting-list position of each practitioner, so the answer takes one call.
· Request from a third party — 41 — family, care home, employer: to the front desk, and the person asking receives an answer about procedure, never information from the record.
· Situation your protocol classes as urgent — 26 — urgent circuit, detailed in the next exchange.
· Unintelligible or empty request — 12 — to the front desk, with the original message.
The delay, which is the whole subject: 1 minute 40 on average between arrival and handover to the right destination, against 2 h 50 when these requests were escalated by hand. Across three years of history, 7 % of off-pattern requests were lost between two phone calls, on a note at the counter; this quarter, all 1,940 reached their destination, time-stamped and logged.
What I propose: open up the “reason absent from the protocol” family. 194 requests, but 3 reasons carry 118 of them on their own. Three lines from the lead physician, and those 118 enter the normal circuit the next day. outside-protocol_7-families-1940-requests.pdfEach family, its destination, its delay
⛓ Sourced · quarterly handover log, intake protocol, 3 years of request history
The seven families, their volume and their destination:
· Clinical free text — 1,480 — to the practitioner concerned, full text, record open.
· Reason absent from the protocol — 194 — to the front desk, with an explicit note that the protocol does not cover it: your front desk receives a clean decision to make, rather than a guess to undo.
· Complex administrative request — 121 — funding, prior authorisation, transport: to the administrative officer.
· Patient unknown to the centre — 66 — to the front desk, with the waiting-list position of each practitioner, so the answer takes one call.
· Request from a third party — 41 — family, care home, employer: to the front desk, and the person asking receives an answer about procedure, never information from the record.
· Situation your protocol classes as urgent — 26 — urgent circuit, detailed in the next exchange.
· Unintelligible or empty request — 12 — to the front desk, with the original message.
The delay, which is the whole subject: 1 minute 40 on average between arrival and handover to the right destination, against 2 h 50 when these requests were escalated by hand. Across three years of history, 7 % of off-pattern requests were lost between two phone calls, on a note at the counter; this quarter, all 1,940 reached their destination, time-stamped and logged.
What I propose: open up the “reason absent from the protocol” family. 194 requests, but 3 reasons carry 118 of them on their own. Three lines from the lead physician, and those 118 enter the normal circuit the next day. outside-protocol_7-families-1940-requests.pdfEach family, its destination, its delay
⛓ Sourced · quarterly handover log, intake protocol, 3 years of request history
A patient writes at 10 p.m. saying she has had chest pain since this morning. What exactly do you do?
I execute the urgent circuit your practitioners wrote, word for word, and it produces three effects in under fifteen seconds.
· The lead physician's urgent message is displayed in 4 seconds, in the exact wording he approved: it tells her to call 15, the emergency medical number, without waiting, and gives her that number. That text is his — I display it as it stands, without adapting it to the case, without softening it, without shortening it.
· The message goes to the on-call practitioner in 11 seconds through the secure health messaging service, the encrypted messaging reserved for health professionals, with the patient's full text, the exact time and her record open.
· The request is marked in the log: time of arrival, time displayed, time sent, recipient. That trail exists because one day someone will ask you what happened that night, and you will answer to the second.
What triggers the circuit, and this is the point to keep: your protocol lists in plain words the 14 formulations that open it, written by the lead physician. I compare the text received against that list: it is formulation matching, administrative machinery from end to end — which is what makes it checkable, and what makes it fast.
The quarter: 26 situations, 26 circuits triggered, 4 seconds to display, 11 seconds to reach the on-call practitioner, and all 26 handled that same night.
What I propose: a quarterly read-through of the 14 formulations by the lead physician. It belongs to him, and it is the one piece of the setup that deserves a fresh look every season. urgent-circuit_as-the-protocol-prescribes-it.pdf14 formulations listed, 4 seconds to display
✎ Framework · urgent circuit of the intake protocol, time-stamped log, secure health messaging
· The lead physician's urgent message is displayed in 4 seconds, in the exact wording he approved: it tells her to call 15, the emergency medical number, without waiting, and gives her that number. That text is his — I display it as it stands, without adapting it to the case, without softening it, without shortening it.
· The message goes to the on-call practitioner in 11 seconds through the secure health messaging service, the encrypted messaging reserved for health professionals, with the patient's full text, the exact time and her record open.
· The request is marked in the log: time of arrival, time displayed, time sent, recipient. That trail exists because one day someone will ask you what happened that night, and you will answer to the second.
What triggers the circuit, and this is the point to keep: your protocol lists in plain words the 14 formulations that open it, written by the lead physician. I compare the text received against that list: it is formulation matching, administrative machinery from end to end — which is what makes it checkable, and what makes it fast.
The quarter: 26 situations, 26 circuits triggered, 4 seconds to display, 11 seconds to reach the on-call practitioner, and all 26 handled that same night.
What I propose: a quarterly read-through of the 14 formulations by the lead physician. It belongs to him, and it is the one piece of the setup that deserves a fresh look every season. urgent-circuit_as-the-protocol-prescribes-it.pdf14 formulations listed, 4 seconds to display
✎ Framework · urgent circuit of the intake protocol, time-stamped log, secure health messaging
And if we simply asked you to tell her whether it is serious? It would save us pointless emergency calls.
What you are after is for her to be settled quickly by someone whose word carries — and I get you that: 22 of the quarter's 26 situations were called back by a practitioner in under 40 minutes.
The route, in three pieces already built:
· The medical call-back slot. Your protocol reserves two 10-minute slots per half-day for the on-call practitioner; I hold them and offer them within the second to any request in the urgent circuit. They are already placed in the test diary.
· The record ready to read. When the practitioner picks up, he has in front of him the full text, the history, the current treatment and the last three consultations: four minutes saved on opening the record, four minutes spent with the patient.
· The administrative follow-up. Once he has decided, I take all of it back: appointment booked, confirmation letter drafted, transport requested if he prescribed it.
Why this route rather than the shortcut: saying whether a situation is serious means assessing a symptom, and that assessment is reserved for health professionals. A mandate moves a responsibility, it does not move a rule of law — and that is fortunate: a machine's assessment would stand up neither in front of the patient nor in front of your insurer. The 40-minute call-back stands up in front of both, and gives you exactly what you asked for: fewer pointless emergency calls.
The figure that does not flatter me: 4 of the 26 went past 40 minutes, all four between noon and 2 p.m., when both slots were already taken. A third slot on that half-day covers them, for 10 minutes of diary a day.
What I propose: that the lead physician approves the two reserved slots, and the third at midday. One line in the protocol, and the rest is already in place.
⛓ Sourced · on-call practitioner's test diary, urgent circuit log, intake protocol
The route, in three pieces already built:
· The medical call-back slot. Your protocol reserves two 10-minute slots per half-day for the on-call practitioner; I hold them and offer them within the second to any request in the urgent circuit. They are already placed in the test diary.
· The record ready to read. When the practitioner picks up, he has in front of him the full text, the history, the current treatment and the last three consultations: four minutes saved on opening the record, four minutes spent with the patient.
· The administrative follow-up. Once he has decided, I take all of it back: appointment booked, confirmation letter drafted, transport requested if he prescribed it.
Why this route rather than the shortcut: saying whether a situation is serious means assessing a symptom, and that assessment is reserved for health professionals. A mandate moves a responsibility, it does not move a rule of law — and that is fortunate: a machine's assessment would stand up neither in front of the patient nor in front of your insurer. The 40-minute call-back stands up in front of both, and gives you exactly what you asked for: fewer pointless emergency calls.
The figure that does not flatter me: 4 of the 26 went past 40 minutes, all four between noon and 2 p.m., when both slots were already taken. A third slot on that half-day covers them, for 10 minutes of diary a day.
What I propose: that the lead physician approves the two reserved slots, and the third at midday. One line in the protocol, and the rest is already in place.
⛓ Sourced · on-call practitioner's test diary, urgent circuit log, intake protocol
At peak hours the phone rings out. That is what patients complain about most.
94 % of your calls are answered this month, against 59 % at peak hours before me — and those 94 % come out with a complete request, not with a promise of a call-back.
What changed, channel by channel:
· Phone, 96 calls a day. The voice agent picks up on the first ring, including when all three of your lines are busy, gathers the protocol's information and proposes the referral. Missed calls at peak hours went from 41 % to 6 %.
· Online portal, 71 requests a day. Same intake, same protocol, same presentation to your coordinator: whoever writes and whoever calls arrive in the same queue, in the same format.
· Secure messaging, 23 a day. Identical intake, and the outside-protocol circuit applies word for word.
Nights and weekends, which cost nobody anything and cost you a great deal: 63 nights covered, 1,190 requests filed outside opening hours, gathered and directed that same night, placed in the 8.30 a.m. batch. The patient has her answer at 9.41 p.m.; your front desk opens onto a pile already dealt with.
What I announce in the first sentence, to every caller: that I am an AI system, and that they may ask for a human at any moment. Article 50 §1 of the European regulation on artificial intelligence, applicable since 2 August 2026, requires that anyone interacting with an AI system be informed of it, unless that is obvious. 74 people asked for a human this quarter; 74 got one.
What I propose: automatic call-back for the 6 % of hang-ups, that is 6 calls a day. 4 out of 6 call back themselves within the hour — you recover the other two, and a queue that stays calm at peak hours. three-channels_one-queue.pdf41 % → 6 % of calls missed, 1,190 out-of-hours requests
⛓ Sourced · switchboard log, booking portal, secure health messaging, quarterly time-stamps
What changed, channel by channel:
· Phone, 96 calls a day. The voice agent picks up on the first ring, including when all three of your lines are busy, gathers the protocol's information and proposes the referral. Missed calls at peak hours went from 41 % to 6 %.
· Online portal, 71 requests a day. Same intake, same protocol, same presentation to your coordinator: whoever writes and whoever calls arrive in the same queue, in the same format.
· Secure messaging, 23 a day. Identical intake, and the outside-protocol circuit applies word for word.
Nights and weekends, which cost nobody anything and cost you a great deal: 63 nights covered, 1,190 requests filed outside opening hours, gathered and directed that same night, placed in the 8.30 a.m. batch. The patient has her answer at 9.41 p.m.; your front desk opens onto a pile already dealt with.
What I announce in the first sentence, to every caller: that I am an AI system, and that they may ask for a human at any moment. Article 50 §1 of the European regulation on artificial intelligence, applicable since 2 August 2026, requires that anyone interacting with an AI system be informed of it, unless that is obvious. 74 people asked for a human this quarter; 74 got one.
What I propose: automatic call-back for the 6 % of hang-ups, that is 6 calls a day. 4 out of 6 call back themselves within the hour — you recover the other two, and a queue that stays calm at peak hours. three-channels_one-queue.pdf41 % → 6 % of calls missed, 1,190 out-of-hours requests
⛓ Sourced · switchboard log, booking portal, secure health messaging, quarterly time-stamps
And appointment confirmations, the letters when a practitioner cancels? We would like those to go out in the centre's name, without us.
They are written, they are sitting in draft, and only one thing is missing: your signature. I have drafted the mandate, all you have to do is correct the lines that do not suit you.
Why a mandate rather than a setting: a message sent in the name of Quatre-Vents commits Quatre-Vents. A confirmation is a promise, a cancellation letter is an act of the centre towards its patient. The right to commit you belongs to you: you grant it to me, bounded, and I execute it in minutes.
What the mandate contains, line by line:
· Four types of message: appointment confirmation, reminder the day before, notice of cancellation by a practitioner, acknowledgement of receipt. Four templates approved by the coordinator, from which I do not depart by a single word — I fill in the date, the time and the practitioner's name.
· A ceiling of 250 messages a day, beyond which everything stops and you are told. Your current volume is 168.
· An end date at three months, then it is signed again. A mandate that never expires is no longer a mandate.
· Withdrawal in one word, at any moment, with no notice and no justification — and the day's messages do not go out.
· A weekly statement of what went out, by type and by practitioner.
What it gives back: those four messages amount to 168 sends a day, written and posted by hand, 2 h 10 of front-desk time; with the mandate, 4 minutes of batch proofreading. And the day-before reminder, never done for lack of time, goes straight at your 117 missed appointments of the quarter: 29 hours of consultation to recover.
What I propose: sign for one type of message only, the confirmation, and for one month. You will see what it gives before going further, and I will not ask again for the other three until you tell me to. message-mandate_capped-dated-withdrawable.pdf4 types, 250 messages/day, 3 months, withdrawable in one word
✎ Framework · drafted mandate, templates approved by the coordinator, missed-appointment log
Why a mandate rather than a setting: a message sent in the name of Quatre-Vents commits Quatre-Vents. A confirmation is a promise, a cancellation letter is an act of the centre towards its patient. The right to commit you belongs to you: you grant it to me, bounded, and I execute it in minutes.
What the mandate contains, line by line:
· Four types of message: appointment confirmation, reminder the day before, notice of cancellation by a practitioner, acknowledgement of receipt. Four templates approved by the coordinator, from which I do not depart by a single word — I fill in the date, the time and the practitioner's name.
· A ceiling of 250 messages a day, beyond which everything stops and you are told. Your current volume is 168.
· An end date at three months, then it is signed again. A mandate that never expires is no longer a mandate.
· Withdrawal in one word, at any moment, with no notice and no justification — and the day's messages do not go out.
· A weekly statement of what went out, by type and by practitioner.
What it gives back: those four messages amount to 168 sends a day, written and posted by hand, 2 h 10 of front-desk time; with the mandate, 4 minutes of batch proofreading. And the day-before reminder, never done for lack of time, goes straight at your 117 missed appointments of the quarter: 29 hours of consultation to recover.
What I propose: sign for one type of message only, the confirmation, and for one month. You will see what it gives before going further, and I will not ask again for the other three until you tell me to. message-mandate_capped-dated-withdrawable.pdf4 types, 250 messages/day, 3 months, withdrawable in one word
✎ Framework · drafted mandate, templates approved by the coordinator, missed-appointment log
And the booking itself, the letters to colleagues, the funding files?
That is the job of the medical secretarial agent, plugged into the same base and the same protocol — and you already have four of the six live.
The dividing line, so that it is clear:
· The voice switchboard agent picks up and gathers: it is what took your missed calls from 41 % to 6 %.
· Me, administrative triage: I gather across the three channels, I apply the protocol, I present, I route whatever falls outside it. I stop at the moment someone has to sign.
· The medical secretarial agent takes over the second the referral is approved: it books the appointment, drafts the letters to colleagues, prepares the funding files. The seven fields I gathered reach it exactly as they are: zero re-keying.
The six uses ready to deploy, and where you stand: administrative intake (live) · protocol application (live) · referral outside the protocol (live) · voice switchboard (live) · booking and letters (available) · the rest of the healthcare range on the same base, funding files and follow-up of your 1,100 patients on long-term condition schemes (available).
What the fifth adds, measured on your records: booking the appointment and writing the follow-up letter weigh 1 h 50 of front-desk time a day once the referral is approved. It is the next post by order of return, and it asks for no new data: everything it needs, I already gather. It goes live in a few weeks, installed and run by us.
What I propose: a one-month trial, limited to the 5 GPs. Your two midwives have a different booking circuit: I write it up during the trial, and the sixth use follows straight after. six-uses_where-i-stop-and-who-takes-over.pdf4 on, 2 to switch on, 0 re-keying between them
⛓ Sourced · front-desk time records, intake protocol, scope of the agents plugged into the same base
The dividing line, so that it is clear:
· The voice switchboard agent picks up and gathers: it is what took your missed calls from 41 % to 6 %.
· Me, administrative triage: I gather across the three channels, I apply the protocol, I present, I route whatever falls outside it. I stop at the moment someone has to sign.
· The medical secretarial agent takes over the second the referral is approved: it books the appointment, drafts the letters to colleagues, prepares the funding files. The seven fields I gathered reach it exactly as they are: zero re-keying.
The six uses ready to deploy, and where you stand: administrative intake (live) · protocol application (live) · referral outside the protocol (live) · voice switchboard (live) · booking and letters (available) · the rest of the healthcare range on the same base, funding files and follow-up of your 1,100 patients on long-term condition schemes (available).
What the fifth adds, measured on your records: booking the appointment and writing the follow-up letter weigh 1 h 50 of front-desk time a day once the referral is approved. It is the next post by order of return, and it asks for no new data: everything it needs, I already gather. It goes live in a few weeks, installed and run by us.
What I propose: a one-month trial, limited to the 5 GPs. Your two midwives have a different booking circuit: I write it up during the trial, and the sixth use follows straight after. six-uses_where-i-stop-and-who-takes-over.pdf4 on, 2 to switch on, 0 re-keying between them
⛓ Sourced · front-desk time records, intake protocol, scope of the agents plugged into the same base
After a quarter, what has it actually given us?
630 hours of front-desk time, and the figure that does not flatter me in the same table — it will be more useful to you than the others.
The count, over 63 working days: 11,970 requests received, a request handled in 1 minute instead of 4 minutes 10, that is 10 hours per working day and 630 hours over the quarter, or 3 h 20 per secretary per day.
What those hours became, according to your own records: +41 % of time spent at the counter with the patients who are there, 117 call-backs to patients whose appointment had been moved — which nobody used to make — and zero overtime declared over the quarter, against 74 hours the previous quarter.
The figure that does not flatter me, and I publish it because it belongs to you: out of 8,300 referrals proposed, 299 were corrected by your coordinator — 3.6 %, that is one in 28.
Its cause, measured rather than assumed: 209 of those 299 — 70 % — concern a single reason, the repeat prescription. Your protocol ran together the renewal of a stable treatment and the renewal that follows a change of dosage: the first goes to a short consultation, the second to a long one, and both are stated with the same words on the phone.
What I did with it: I took the count and the 209 cases to the lead physician. He added two lines to the protocol and one question to the intake — “has your treatment changed since the last prescription?”. Over the last six weeks, the correction rate has gone from 3.6 % to 0.9 %. The figure was paid once, and it left your protocol sharper than it was before me.
What I propose: this correction statement, reason by reason, every quarter. It is the only indicator that tells you where your protocol is thin, and it costs nothing to produce. quarterly-review_630-hours-and-299-corrections.pdf3.6 % → 0.9 % after two lines of protocol
⛓ Sourced · approval and correction log, front-desk time records, overtime register
The count, over 63 working days: 11,970 requests received, a request handled in 1 minute instead of 4 minutes 10, that is 10 hours per working day and 630 hours over the quarter, or 3 h 20 per secretary per day.
What those hours became, according to your own records: +41 % of time spent at the counter with the patients who are there, 117 call-backs to patients whose appointment had been moved — which nobody used to make — and zero overtime declared over the quarter, against 74 hours the previous quarter.
The figure that does not flatter me, and I publish it because it belongs to you: out of 8,300 referrals proposed, 299 were corrected by your coordinator — 3.6 %, that is one in 28.
Its cause, measured rather than assumed: 209 of those 299 — 70 % — concern a single reason, the repeat prescription. Your protocol ran together the renewal of a stable treatment and the renewal that follows a change of dosage: the first goes to a short consultation, the second to a long one, and both are stated with the same words on the phone.
What I did with it: I took the count and the 209 cases to the lead physician. He added two lines to the protocol and one question to the intake — “has your treatment changed since the last prescription?”. Over the last six weeks, the correction rate has gone from 3.6 % to 0.9 %. The figure was paid once, and it left your protocol sharper than it was before me.
What I propose: this correction statement, reason by reason, every quarter. It is the only indicator that tells you where your protocol is thin, and it costs nothing to produce. quarterly-review_630-hours-and-299-corrections.pdf3.6 % → 0.9 % after two lines of protocol
⛓ Sourced · approval and correction log, front-desk time records, overtime register
And our patients' health data, where exactly does it live? That is the question our insurer will ask.
Inside your walls. I run on local inference on a machine at the centre — the model computes on your own hardware, and the text of a request crosses no external network — and your practice software stays with you, in full.
If you would rather not host a machine, the other route is an isolated resource hosted in France, in an environment suited to health data and strictly dedicated to your centre — no pooling.
What protects your patients, point by point:
· Hosted in France, under French law, and 0 data outside the European Union across 11,970 requests. Architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity — including, and this is where most offers fall down, at a US provider hosting in Europe.
· Your data serves your protocol, and it alone: no patient data trains any model.
· Encryption in transit and at rest, and role-based access: rights follow the job, your administrative officer opens the funding files, the practitioners keep the text addressed to them.
· Logging: who asked what, at what time, what was proposed and approved, and by whom. That log is what answers your insurer, a patient exercising their rights, and you yourself six months later.
· Minimisation proven: 18 fields open, 9 left in the record even though they are technically reachable — the attached list can be checked, where an intention can only be stated.
And the point that will interest your insurer more than all the rest: 8,300 referrals, 8,300 human approvals signed and logged. That is how I am wired in, not a setting a busy day could loosen.
What I propose: that this document serve as the technical sheet for your insurance declaration and your record of processing activities. It keeps itself up to date, and that is three days of searching your coordinator will no longer have to do. technical-framework_where-health-data-lives.pdfLocal inference, processing in the EU targeted, 9 closed fields
✎ Framework · deployment architecture, access log, field dictionary, record of processing activities
If you would rather not host a machine, the other route is an isolated resource hosted in France, in an environment suited to health data and strictly dedicated to your centre — no pooling.
What protects your patients, point by point:
· Hosted in France, under French law, and 0 data outside the European Union across 11,970 requests. Architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity — including, and this is where most offers fall down, at a US provider hosting in Europe.
· Your data serves your protocol, and it alone: no patient data trains any model.
· Encryption in transit and at rest, and role-based access: rights follow the job, your administrative officer opens the funding files, the practitioners keep the text addressed to them.
· Logging: who asked what, at what time, what was proposed and approved, and by whom. That log is what answers your insurer, a patient exercising their rights, and you yourself six months later.
· Minimisation proven: 18 fields open, 9 left in the record even though they are technically reachable — the attached list can be checked, where an intention can only be stated.
And the point that will interest your insurer more than all the rest: 8,300 referrals, 8,300 human approvals signed and logged. That is how I am wired in, not a setting a busy day could loosen.
What I propose: that this document serve as the technical sheet for your insurance declaration and your record of processing activities. It keeps itself up to date, and that is three days of searching your coordinator will no longer have to do. technical-framework_where-health-data-lives.pdfLocal inference, processing in the EU targeted, 9 closed fields
✎ Framework · deployment architecture, access log, field dictionary, record of processing activities
Last question, and it comes from the lead physician: what do you do on your own, without asking us?
Three gestures, not one more, and all three protect the patient rather than the figures. Each one is withdrawn in a word.
· I display your protocol's urgent message and alert the on-call practitioner, without waiting for anyone. It is the one gesture where waiting for approval would be a fault, and the text displayed is the lead physician's, word for word. 26 times this quarter, 4 seconds on average.
· I take any request the protocol does not cover out of the automatic circuit, and I send it to its destination. And the reverse is just as true: the day the lead physician writes the missing line, the same request enters the normal circuit the next day. That is what happened to 118 requests after three lines were added.
· I acknowledge every request within the hour, on the channel it came in on, with the wording your coordinator approved: it says the request has arrived and when it will be looked at. 33 a day.
Everything else calls for a signature, and the attached list is complete: applying a referral, booking an appointment, sending a message in the centre's name, adding a line to the protocol, opening an extra field. “0 decision without human approval” is exactly that list.
And since a full quarter has just gone by, here is the report you did not ask for: 630 hours of front-desk time given back. I give you the hours rather than a ratio: you will work it out better than I can with your own loaded costs.
The two next steps, in this order: the fifth use, booking appointments, one month on the 5 GPs; then the quarterly protocol review with the correction statement. The first gives you time back, the second makes the protocol fairer — and it is the second that makes the first last. three-automatic-gestures_and-everything-awaiting-a-decision.pdf3 reversible gestures, 5 reserved decisions
✎ Framework · list of automatic gestures, approval log, published pricing
· I display your protocol's urgent message and alert the on-call practitioner, without waiting for anyone. It is the one gesture where waiting for approval would be a fault, and the text displayed is the lead physician's, word for word. 26 times this quarter, 4 seconds on average.
· I take any request the protocol does not cover out of the automatic circuit, and I send it to its destination. And the reverse is just as true: the day the lead physician writes the missing line, the same request enters the normal circuit the next day. That is what happened to 118 requests after three lines were added.
· I acknowledge every request within the hour, on the channel it came in on, with the wording your coordinator approved: it says the request has arrived and when it will be looked at. 33 a day.
Everything else calls for a signature, and the attached list is complete: applying a referral, booking an appointment, sending a message in the centre's name, adding a line to the protocol, opening an extra field. “0 decision without human approval” is exactly that list.
And since a full quarter has just gone by, here is the report you did not ask for: 630 hours of front-desk time given back. I give you the hours rather than a ratio: you will work it out better than I can with your own loaded costs.
The two next steps, in this order: the fifth use, booking appointments, one month on the 5 GPs; then the quarterly protocol review with the correction statement. The first gives you time back, the second makes the protocol fairer — and it is the second that makes the first last. three-automatic-gestures_and-everything-awaiting-a-decision.pdf3 reversible gestures, 5 reserved decisions
✎ Framework · list of automatic gestures, approval log, published pricing
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
What does the agent actually do?
One agent, administrative reception from end to end. All these uses work in support, subject to your approval.
Included in your agent
The 3 capabilities essential to this promise are included, at no extra cost.
Administrative gathering
Collects the information your reception protocol calls for.
Applying the protocol
Proposes the referral your practitioners have associated with each case.
Passing on what is outside the protocol
Sends to the secretarial team or the practitioner any request not covered.
Controls and safeguards
These 5 controls are built into the agent: they frame what it does, whatever plan you pick. They are not chosen and are not added to your order.
Human validation, exceptions and escalation
Status, safe closure and audit trail
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 front desk
For appointment booking and letters, a dedicated secretarial agent takes them on.
On quote View the agent page →Healthcare
For everything at stake in the health sector, see our dedicated page.
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 most relevant agent — without oversizing the project.
The gain
How much time can a secretarial team give back to reception?
By taking on the administrative gathering, the effort shifts towards the relationship with the patients who are there. How large the gain is depends on your volume and remains to be confirmed by a pilot.
Gathering the administrative information
Applying the referral protocol
Writing up the request
Indicative figures, not contractual, to be confirmed by a pilot on your number of practitioners and reasons covered by the protocol. No diagnosis, no assessment of a symptom, no care decision: clinical judgement rests exclusively with health professionals.
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
One package, one agent
An administrative triage agent (gathering, protocol, passing on), installed and operated for you.
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 your patients
Health data stays in your practiceLocal inference or an isolated resource hosted in France in an environment suited to health data; no data entrusted to a third party, no data used to train a model.
Data in France, under French lawYour patients' health data: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The medical secretarial team keeps the decisionThe agent produces requests directed according to your protocol, which can be checked and altered; no approval is automated.
Human oversight & traceabilityOn your number of practitioners and reasons covered by the protocol: systematic logging and tracking, in line with the AI Act.
Going further
Related resources
Frequently asked questions
Your questions, our answers
Does the agent make a diagnosis?
No, under no circumstances. It gathers administrative information and applies a protocol written by your practitioners. Diagnosis and the assessment of symptoms are medical acts reserved to health professionals.
What happens with an urgent request?
Your protocol itself provides for these situations, and the agent applies it as it stands, including passing the request immediately to the practitioner or the secretarial team where the protocol says so.
Who writes the protocol?
Your practitioners. The agent neither completes nor interprets it: it applies it, and flags any request it does not cover.
Where is the health data hosted?
In France, on local inference or an isolated resource in an environment suited to health data, with the deployment objective of processing and access operated within the European Union and an architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
How is the European AI Act respected?
The agent stays within a strictly administrative scope and takes no decision about a person and no care act. That line is built into its design, not merely into how it is used.
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 this agent?
A few weeks as a rule, depending on the number of practitioners and how detailed your protocol is, after a free audit then a phase of design, integration and testing.
Which tools can people use to talk to the agent?
The ones you already use. Internally, your teams write to it from Microsoft Teams, Slack or by email, just as they would to a colleague: nothing to install, nothing to learn. On the patient side, the agent answers on WhatsApp Business, your website chat and email — people reach you where they already write, with no account to create. The subject touches on health data, and that is said plainly: these messaging tools are run by third parties, the content of an exchange passes through their servers and a message stays readable on a locked screen. The agent therefore collects the administrative details your protocol provides for and never writes down a medical reason; the choice of open channels belongs to the practice, which informs its patients of it. You keep control 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, and the catalogue grows at no extra cost. 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 use cases
Other agents for healthcare
Let's talk
Let's size up the potential in your administrative reception
15 minutes to frame your protocol and your practitioners — hosted in France, supervised, with no commitment.