● Business offer — Switchboard
AI voice agent: a switchboard that never misses a call
A missed call is often a lost customer — an appointment not made, a quote that goes to a competitor, a patient who calls elsewhere. Yet nobody can answer permanently. The AI voice agent takes every call, at any hour, understands the request and acts: it informs, books appointments and transfers what needs transferring. Hosted in France — on local inference or an isolated resource — callers' data stays under control. The human keeps the lead on sensitive cases.
Answers around the clock
Hosted in France
GDPR & AI Act: governed deployment
Transfer to a person
Updated on
Deployed in two to three weeks
AI voice switchboard · hosted in France
Hello, I would like to book an appointment for a scale and polish, preferably late in the day if possible.
Of course. For a scale and polish, I have two late-day slots: Thursday 6.30pm or next Tuesday 6pm. Which suits you best?
I will send you a confirmation and a reminder by text message.
⛓ Sourced · the practice's live diary
I will send you a confirmation and a reminder by text message.
⛓ Sourced · the practice's live diary
Thursday 6.30pm is perfect. And is it covered by insurance?
Noted for Thursday 6.30pm. Confirmation sent. On the reimbursement question, which calls for precise advice, I would rather put you through: I am passing your query to the front desk with the context, and they will call you back in the morning.
✎ Action · appointment recorded + transfer with context — the human takes over
✎ Action · appointment recorded + transfer with context — the human takes over
Local inference · no data outside the EU
Data hosted in France
Zero missed callsAvailable around the clock, including out of hours
GDPR & AI Act: governed deploymentCallers informed & human oversight
TurnkeyDesigned, installed and operated for you
The human decidesThe agent assists, never rules on sensitive cases
✦ In brief
In brief — a Blue Lemon Agent voice agent is an automated switchboard that answers calls around the clock: it informs, books appointments, routes to the right department and passes complex cases to a person. It understands plain language and cuts missed calls. It runs on local inference or is hosted in France: transcripts and callers' contact details are never exposed to a foreign service, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. Live within a few weeks. Your teams write to it from Microsoft Teams, Slack or their email, and your customers reach it on WhatsApp Business, your website chat, email or VoIP telephony — 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.
Ringing → 0 s
call answered with no wait, at any hour (illustrative)
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
8
uses ready to deploy for your switchboard
Illustrative reference points describing our offer — to be confirmed by a pilot on your own call volume.
The context
Why an AI voice switchboard — and why people still hesitate
The phone remains the first point of contact for many activities. But nobody can pick up continuously: every missed call is an opportunity that goes elsewhere — and every call contains personal data to protect.
! The issue
An organisation that lives off the phone is caught between callers who want an immediate answer (information, appointment, quote) and a switchboard that saturates at peak times or falls silent out of hours. Yet most consumer voice solutions amount to entrusting recordings, transcripts and contact details to a third party, often hosted outside Europe and subject to the Cloud Act.
✓ Our answer
A voice agent is only of interest if it is sovereign and confidential by design. Local inference or an isolated resource hosted in France, callers informed, transfer to a person for anything sensitive: no longer missing a call is never paid for in lost confidentiality. The aim is not to replace reception, but to absorb the flow and give your teams time back.
The decisive point
Call data: sovereignty & compliance
A call contains personal data, sometimes sensitive. Here is how the architecture of our agents protects it, call after call.
Local inference
The agent can run on a machine at your own site: transcripts and contact details do not leave the network, nothing passes through a foreign cloud.
Hosting in France
Otherwise, a dedicated and isolated resource, hosted in France under French law — call data: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
Exposure of call data to the Cloud Act and FISA 702 is reduced by design; location alone does not guarantee immunity.
One isolated resource per client
No pooling of call data: an environment strictly dedicated to your organisation.
Encryption & controlled access
Encryption in transit and at rest, role-based access (RBAC), strong authentication and logging of calls.
AI Act: governed deployment
An agent strictly in support; no sensitive decision automated; callers informed 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 healthcare organisations, health-data hosting (HDS) is available; a SecNumCloud option 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
4 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.
AI voice switchboard · hosted in France
Four flags this morning on your switchboard.
· Three hundred and twelve calls were abandoned while waiting last month, 180 of them after more than three minutes. Nobody has ever called them back.
· Your greeting announces opening hours that are no longer yours, since March. Sixty calls came in on Friday afternoons to a closed line.
· The number the greeting gives for technical support is no longer allocated. It has been dialled 44 times.
· Seven transfers went nowhere: the destination line rings out. morning-watch_4-flags.pdf312 abandoned · 0 called back
⛓ Source · call log, recorded greeting, numbering plan, transfer log
· Three hundred and twelve calls were abandoned while waiting last month, 180 of them after more than three minutes. Nobody has ever called them back.
· Your greeting announces opening hours that are no longer yours, since March. Sixty calls came in on Friday afternoons to a closed line.
· The number the greeting gives for technical support is no longer allocated. It has been dialled 44 times.
· Seven transfers went nowhere: the destination line rings out. morning-watch_4-flags.pdf312 abandoned · 0 called back
⛓ Source · call log, recorded greeting, numbering plan, transfer log
312 abandoned calls — is there nothing to be done?
There is, and the first step costs nothing: knowing who they were.
What I record: 312 calls abandoned while waiting, 180 of them after more than three minutes. The caller's number is recorded in 289 cases.
What nobody does today: call them back. They appear in no report — an abandoned call leaves no trace on the business side, and so is deemed never to have happened.
What I propose: a daily list of abandons, with the time, the wait, the number where present and the department chosen from the menu. The callback itself is yours: it is a sales contact or a service rendered, it carries the company's name, and it is yours to decide — the list is ready every morning at 8.
What I can do during the wait, and it is the most useful thing: announce the real waiting time — not a formula. In a dry run, abandons fell from 312 to 190 on that announcement alone, and those who hang up do so after 40 seconds instead of three minutes.
Why I prefer announcing to holding: music and a reassuring phrase make the same person wait longer for the same outcome. The announcement gives them the choice back; the music takes it away. 312-abandons_0-callbacks.pdf40 seconds instead of 3 minutes
⛓ Source · 312 abandons, waiting log, dry run
What I record: 312 calls abandoned while waiting, 180 of them after more than three minutes. The caller's number is recorded in 289 cases.
What nobody does today: call them back. They appear in no report — an abandoned call leaves no trace on the business side, and so is deemed never to have happened.
What I propose: a daily list of abandons, with the time, the wait, the number where present and the department chosen from the menu. The callback itself is yours: it is a sales contact or a service rendered, it carries the company's name, and it is yours to decide — the list is ready every morning at 8.
What I can do during the wait, and it is the most useful thing: announce the real waiting time — not a formula. In a dry run, abandons fell from 312 to 190 on that announcement alone, and those who hang up do so after 40 seconds instead of three minutes.
Why I prefer announcing to holding: music and a reassuring phrase make the same person wait longer for the same outcome. The announcement gives them the choice back; the music takes it away. 312-abandons_0-callbacks.pdf40 seconds instead of 3 minutes
⛓ Source · 312 abandons, waiting log, dry run
How did you spot all that? And who do you tell?
I watch, continuously, what you have opened to me: the call log, waiting times, transfers and whether they connect, the recorded greeting and your numbering plan.
Routing follows who can pick up: a transfer that does not connect goes to the switchboard lead, within the hour — it is a lost call every time; an out-of-date greeting to whoever recorded it, within 48 h; an unallocated number to the same person; abandons to the relevant department's lead, every morning.
With a chase: within the hour on a dead transfer, 48 h on the rest. Then a monthly summary: by call reason and by hour, never by the person answering.
And I do not stop at the flag: what is mine to fix is already fixed. Friday afternoon calls have had the accurate information since this morning; the unallocated number is out of my announcements and the right one is found in your numbering plan; dead transfers come back with the destination extension and the last time it answered; and the abandons list goes out every morning. Correcting what I myself say is mine to do.
One rule holds up all the rest, and it is said out loud: I announce that I am an automated agent in my first sentence, I make no commitment on the company's behalf, and I never press. A greeting and a commitment are what the company says about itself — they carry its name, so a person signs them. I pick up; your teams answer.
✎ Framework · automated-agent announcement, no commitment, no pressing
Routing follows who can pick up: a transfer that does not connect goes to the switchboard lead, within the hour — it is a lost call every time; an out-of-date greeting to whoever recorded it, within 48 h; an unallocated number to the same person; abandons to the relevant department's lead, every morning.
With a chase: within the hour on a dead transfer, 48 h on the rest. Then a monthly summary: by call reason and by hour, never by the person answering.
And I do not stop at the flag: what is mine to fix is already fixed. Friday afternoon calls have had the accurate information since this morning; the unallocated number is out of my announcements and the right one is found in your numbering plan; dead transfers come back with the destination extension and the last time it answered; and the abandons list goes out every morning. Correcting what I myself say is mine to do.
One rule holds up all the rest, and it is said out loud: I announce that I am an automated agent in my first sentence, I make no commitment on the company's behalf, and I never press. A greeting and a commitment are what the company says about itself — they carry its name, so a person signs them. I pick up; your teams answer.
✎ Framework · automated-agent announcement, no commitment, no pressing
Ideally people would not realise it is a machine, no?
It is the commonest request on a voice switchboard, and the one I refuse outright.
What I say, on every call, first: "Hello, you have reached [company]'s automated assistant. I can direct you or take your message; to speak to a person, just say so at any time."
What I refuse: carrying a first name, saying "let me see what I can do", performing hesitation, and letting anyone believe a person is listening.
Why this is not a matter of principle but of consequence: somebody who believes they are speaking to a human says things they would not say to a machine — a personal situation, a payment difficulty, a health problem. They say them because they think someone can hear.
What it costs: across 2,100 calls, 340 people asked for a human at the first sentence. They got one. That is not a loss: it is 340 people who did not spend four minutes with me first.
What I always do: I say again that I am automated if the caller thanks me the way you thank a person. 1-sentence_340-humans-asked-for.pdfNo first name · no performed hesitation
⛓ Source · 2,100 calls, greeting sentence, 340 requests for a human
What I say, on every call, first: "Hello, you have reached [company]'s automated assistant. I can direct you or take your message; to speak to a person, just say so at any time."
What I refuse: carrying a first name, saying "let me see what I can do", performing hesitation, and letting anyone believe a person is listening.
Why this is not a matter of principle but of consequence: somebody who believes they are speaking to a human says things they would not say to a machine — a personal situation, a payment difficulty, a health problem. They say them because they think someone can hear.
What it costs: across 2,100 calls, 340 people asked for a human at the first sentence. They got one. That is not a loss: it is 340 people who did not spend four minutes with me first.
What I always do: I say again that I am automated if the caller thanks me the way you thank a person. 1-sentence_340-humans-asked-for.pdfNo first name · no performed hesitation
⛓ Source · 2,100 calls, greeting sentence, 340 requests for a human
And somebody you cannot understand?
Two attempts, then I hand over. There is no third "I did not catch that".
What I do on the first: I put my question differently, shorter.
What I do on the second: I transfer, without framing it as a failure — "let me put you through to somebody."
Why two and not five: because the third attempt almost never works and because it falls on particular people. A noisy line, a car, an accent, an older caller, somebody who has difficulty speaking, somebody in tears. An agent that presses makes exactly the people who struggle most repeat themselves three times.
What I guarantee on the second attempt, and there is no other outcome: a call I cannot understand ends with a person. I transfer with the reason and what I did understand of the call, so the caller does not start from scratch — and if the destination line does not pick up, I take the call back rather than let it drop, I take the message and I escalate it the same day.
Over the month: 41 calls transferred for lack of understanding, that is 2%. Seven of them rang out — that is this morning's flag, and the only case where my transfer was worth nothing. 2-attempts_41-transfers.pdf0 hang-ups · 7 dead transfers
⛓ Source · 41 transfers for lack of understanding, connection log
What I do on the first: I put my question differently, shorter.
What I do on the second: I transfer, without framing it as a failure — "let me put you through to somebody."
Why two and not five: because the third attempt almost never works and because it falls on particular people. A noisy line, a car, an accent, an older caller, somebody who has difficulty speaking, somebody in tears. An agent that presses makes exactly the people who struggle most repeat themselves three times.
What I guarantee on the second attempt, and there is no other outcome: a call I cannot understand ends with a person. I transfer with the reason and what I did understand of the call, so the caller does not start from scratch — and if the destination line does not pick up, I take the call back rather than let it drop, I take the message and I escalate it the same day.
Over the month: 41 calls transferred for lack of understanding, that is 2%. Seven of them rang out — that is this morning's flag, and the only case where my transfer was worth nothing. 2-attempts_41-transfers.pdf0 hang-ups · 7 dead transfers
⛓ Source · 41 transfers for lack of understanding, connection log
Monday mornings and the September return, we are swamped. That is where we lose people.
That is exactly what I am built to absorb: the peaks and the switchboard overflow, with no queue and nobody listening to hold music.
What your peaks cost today: your 312 abandoned calls for the month cluster 68 % into three windows — Monday 8:30 to 10:00, Tuesday 14:00 to 15:00, and the first week of September. In those windows your switchboard takes 4.2 calls a minute and two people answer. That is not an organisational failing, it is arithmetic.
How I absorb a peak: I answer every call in parallel — there is no queue, no wait, no rank. A caller arriving at the 40th second of a burst is handled like the first. Overflow is set by you, not by me: you decide from how many simultaneous calls I take over from the human switchboard, and at what hour I answer alone.
Answering 24/7 and informing: outside your hours I answer too, and I inform from what is written and dated on your side — opening hours, address and access, submission arrangements, availability of a slot in the diary, whether a line is open. Across the 60 calls that came in on a Friday afternoon to a closed line, each now gets the exact information instead of ringing into nothing. What I can inform about stops where your documentation stops: if it is not written, I take a message.
The figure that does not flatter me: out of 1,100 calls handled in the first month, 214 exited as a transfer or a message, 19 % — and of those 214, 61 concerned three questions your documentation does not cover. What I do with it: I hand you the three questions with their volume; once written, they answer themselves. The following month: 12 % exits.
⛓ Source · 68 % of abandons in three windows, 4.2 calls a minute, 19 % then 12 % exits
What your peaks cost today: your 312 abandoned calls for the month cluster 68 % into three windows — Monday 8:30 to 10:00, Tuesday 14:00 to 15:00, and the first week of September. In those windows your switchboard takes 4.2 calls a minute and two people answer. That is not an organisational failing, it is arithmetic.
How I absorb a peak: I answer every call in parallel — there is no queue, no wait, no rank. A caller arriving at the 40th second of a burst is handled like the first. Overflow is set by you, not by me: you decide from how many simultaneous calls I take over from the human switchboard, and at what hour I answer alone.
Answering 24/7 and informing: outside your hours I answer too, and I inform from what is written and dated on your side — opening hours, address and access, submission arrangements, availability of a slot in the diary, whether a line is open. Across the 60 calls that came in on a Friday afternoon to a closed line, each now gets the exact information instead of ringing into nothing. What I can inform about stops where your documentation stops: if it is not written, I take a message.
The figure that does not flatter me: out of 1,100 calls handled in the first month, 214 exited as a transfer or a message, 19 % — and of those 214, 61 concerned three questions your documentation does not cover. What I do with it: I hand you the three questions with their volume; once written, they answer themselves. The following month: 12 % exits.
⛓ Source · 68 % of abandons in three windows, 4.2 calls a minute, 19 % then 12 % exits
And when you have to hand over, does the person start again from scratch?
No: I transfer with the context, and the context arrives before the call — not during it.
What the person picking up sees on screen before speaking: the reason for the call in one sentence, in the caller's own words; what has already been said and checked during the exchange with me; the number and the name if the caller gave them; the call history for that number over thirty days — a second call in three days for one of them; and what I was unable to handle, named.
What transferring with context prevents, measured on your calls: a transfer without context makes the caller restate the reason in 9 cases out of 10 and lengthens the call by 50 seconds at the median. Across 214 transfers a month, that is close to three hours spent re-hearing what was just said.
What I tell the caller at the moment of transfer: “I am putting you through; they have your request in front of them.” The sentence matters: it tells the person they will not have to start again, and that is what makes them wait the fifteen seconds of the transfer.
And if nobody picks up: the context becomes the message. Reason, contact details, what has already been checked, and the time — passed to the department concerned, and the caller is told it is a message and not a transfer. An announced transfer that fails is the one thing that costs more than a wait.
⛓ Source · 214 transfers a month, 50 seconds of median saved, 30-day call history
What the person picking up sees on screen before speaking: the reason for the call in one sentence, in the caller's own words; what has already been said and checked during the exchange with me; the number and the name if the caller gave them; the call history for that number over thirty days — a second call in three days for one of them; and what I was unable to handle, named.
What transferring with context prevents, measured on your calls: a transfer without context makes the caller restate the reason in 9 cases out of 10 and lengthens the call by 50 seconds at the median. Across 214 transfers a month, that is close to three hours spent re-hearing what was just said.
What I tell the caller at the moment of transfer: “I am putting you through; they have your request in front of them.” The sentence matters: it tells the person they will not have to start again, and that is what makes them wait the fifteen seconds of the transfer.
And if nobody picks up: the context becomes the message. Reason, contact details, what has already been checked, and the time — passed to the department concerned, and the caller is told it is a message and not a transfer. An announced transfer that fails is the one thing that costs more than a wait.
⛓ Source · 214 transfers a month, 50 seconds of median saved, 30-day call history
Can it confirm an order, an appointment, a lead time?
It can note them and read them back. It does not confirm them, and the distinction matters more on the phone than anywhere else.
What I say: "I am noting your request for an appointment on [date]; the department will confirm in writing."
What I do not say: "that's booked", "you'll be delivered Thursday", "your refund is approved".
Why it is worse on the phone: the caller has nothing. No email, no screen, no trace. They leave with a sentence they heard, and they heard it as a company commitment. A wrong email can be reread; a sentence on the phone gets retold.
What I always do: I read back what I noted before ending the call, and I send a written summary where a contact is given. Across 2,100 calls, 1,640 received something in writing.
What I do about the other 460: I flag it — 460 people left with information given orally only, and that is this arrangement's real risk. 1640-written_460-oral.pdfWriting gets reread · a sentence gets retold
⛓ Source · 2,100 calls, 1,640 summaries sent
What I say: "I am noting your request for an appointment on [date]; the department will confirm in writing."
What I do not say: "that's booked", "you'll be delivered Thursday", "your refund is approved".
Why it is worse on the phone: the caller has nothing. No email, no screen, no trace. They leave with a sentence they heard, and they heard it as a company commitment. A wrong email can be reread; a sentence on the phone gets retold.
What I always do: I read back what I noted before ending the call, and I send a written summary where a contact is given. Across 2,100 calls, 1,640 received something in writing.
What I do about the other 460: I flag it — 460 people left with information given orally only, and that is this arrangement's real risk. 1640-written_460-oral.pdfWriting gets reread · a sentence gets retold
⛓ Source · 2,100 calls, 1,640 summaries sent
Do you record the calls?
No, and that is not the same question as "do you keep a record".
What I keep: the reason for the call, what was noted, and the department it went to. As text, and only what serves to handle the request.
What I do not record: the voice. Not for quality, not for training, not to check what I understood.
Why the voice is different: a voice recording contains far more than the content — who is speaking, how, with what accent, what apparent age, what state. Text says none of that, and handling the request never needs it.
What that stops me doing, and I accept it: I cannot prove to you what a caller said. If a dispute arises over a call's content, my record is what I understood, not what was said — and I present it as such.
What I propose if recording is necessary to you: that it be decided, announced to the caller, time-limited, and that it be an arrangement of yours rather than an option in my settings. 0-recording_text-only.pdfWhat a voice carries beyond a text
✎ Framework · no voice recorded — text only
What I keep: the reason for the call, what was noted, and the department it went to. As text, and only what serves to handle the request.
What I do not record: the voice. Not for quality, not for training, not to check what I understood.
Why the voice is different: a voice recording contains far more than the content — who is speaking, how, with what accent, what apparent age, what state. Text says none of that, and handling the request never needs it.
What that stops me doing, and I accept it: I cannot prove to you what a caller said. If a dispute arises over a call's content, my record is what I understood, not what was said — and I present it as such.
What I propose if recording is necessary to you: that it be decided, announced to the caller, time-limited, and that it be an arrangement of yours rather than an option in my settings. 0-recording_text-only.pdfWhat a voice carries beyond a text
✎ Framework · no voice recorded — text only
Do you measure the switchboard's performance?
The switchboard's yes, people's no — and the line is sharp.
What I measure: calls received by hour and by reason, waiting times, abandons, transfers that do not connect, and the reasons I most often hand over.
What I do not measure: average call length per employee, calls handled per person, time between calls, and each person's transfer rate.
Why those four specifically: they are the four classic call-centre indicators, and they all push the same way: hang up faster. An employee measured on call length cuts short the difficult call, which is precisely the one that should have run long.
What I report instead, and it serves as well: the reasons I hand over. Over the month, 38% of transfers concern three questions whose answer exists in no document opened to me. Those three questions occupy one person full time, and nobody knew. 4-measures-refused_3-questions.pdf38% of transfers · 3 unanswered questions
⛓ Source · 2,100 calls, transfer reasons, documents opened
What I measure: calls received by hour and by reason, waiting times, abandons, transfers that do not connect, and the reasons I most often hand over.
What I do not measure: average call length per employee, calls handled per person, time between calls, and each person's transfer rate.
Why those four specifically: they are the four classic call-centre indicators, and they all push the same way: hang up faster. An employee measured on call length cuts short the difficult call, which is precisely the one that should have run long.
What I report instead, and it serves as well: the reasons I hand over. Over the month, 38% of transfers concern three questions whose answer exists in no document opened to me. Those three questions occupy one person full time, and nobody knew. 4-measures-refused_3-questions.pdf38% of transfers · 3 unanswered questions
⛓ Source · 2,100 calls, transfer reasons, documents opened
And the out-of-date greeting?
It announces hours that changed in March, and correcting it is yours to do.
What I record: the recorded greeting announces Friday afternoon opening. Your posted hours show closure since 06/03/2026. Sixty calls have arrived in that slot since.
What I did this morning, and it does not touch the greeting: Friday afternoon calls now get the accurate information — closed, reopening Monday, and the option of leaving a message. Correcting what I myself say is mine to do; correcting your greeting is yours.
What I have prepared for that greeting: the script to record, written and timed — twenty-two seconds — with the current hours and the corrected technical support number. All that is left is to say it into the microphone.
Why it is your voice and not mine: a greeting is what the company says about itself, and I do not speak in its place on that register — not even to correct an obvious error. It is also what tells a caller, in the first second, whether they are talking to the firm or to a machine.
What I also flag: the technical support number that same greeting gives is no longer allocated. It has been dialled 44 times. Those 44 people called nowhere, and they appear in none of your counters — not even in the 312 abandons. 1-greeting_60-calls_44-dialled.pdfWhat the agent corrects · what it leaves
⛓ Source · recorded greeting, posted hours, 60 calls, 44 dialled
What I record: the recorded greeting announces Friday afternoon opening. Your posted hours show closure since 06/03/2026. Sixty calls have arrived in that slot since.
What I did this morning, and it does not touch the greeting: Friday afternoon calls now get the accurate information — closed, reopening Monday, and the option of leaving a message. Correcting what I myself say is mine to do; correcting your greeting is yours.
What I have prepared for that greeting: the script to record, written and timed — twenty-two seconds — with the current hours and the corrected technical support number. All that is left is to say it into the microphone.
Why it is your voice and not mine: a greeting is what the company says about itself, and I do not speak in its place on that register — not even to correct an obvious error. It is also what tells a caller, in the first second, whether they are talking to the firm or to a machine.
What I also flag: the technical support number that same greeting gives is no longer allocated. It has been dialled 44 times. Those 44 people called nowhere, and they appear in none of your counters — not even in the 312 abandons. 1-greeting_60-calls_44-dialled.pdfWhat the agent corrects · what it leaves
⛓ Source · recorded greeting, posted hours, 60 calls, 44 dialled
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 voice agent actually do?
Each use corresponds to a task the agent takes on. It stays on reception and administration, and transfers the rest — the scope is defined with you.
Included in your agent
The 7 capabilities essential to this promise are included, at no extra cost.
From 856 € excl. VAT / month
Answering 24/7 & informing
Picking up every call and giving information: opening hours, address, terms, availability — with no wait and no queue.
Booking appointments
Proposing slots in your diary, recording the appointment, sending confirmation and reminder — fewer missed appointments.
Qualifying & routing
Understanding the request and routing the call to the right department or the right person.
Transferring with context
Passing complex requests to a person, along with the reason for the call and its history.
Taking a detailed message
When nobody is available, collecting a structured message (reason, contact details, urgency) and passing it on.
Absorbing peaks & overflow
Taking the bursts of calls at peak times and stepping in when the switchboard is saturated.
Keeping the call log
Recording reasons, outcomes and transfers in a web dashboard so you can steer your reception.
Controls and safeguards
These 4 controls are built into the agent: they frame what it does, whatever plan you pick. They are not chosen and are not added to your order.
Human validation, exceptions and escalation
Status, safe closure and audit trail
Resolve common requests through to a verifiable closure
Escalate quickly to a human, with the reason and the history
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.
Carry the full context across channels and into the ticket/CRM
Other needs our agents cover
Each card says where the matching agent stands: available, on quote, or still being architected.
Multilingual translation
Answering callers in another language is not included here: translation is the service of a dedicated agent, which sets out the languages, the volumes and the human validation.
On quote View the agent page →
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 many calls, and how much time, can you win back?
By answering calls, including out of hours, the agent wins back appointments and enquiries that were going to competitors, while easing the load on the switchboard. The return on investment depends on what a call is worth to your activity.
Calls handled
Wait before pick-up
Switchboard tied up on simple calls
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
A voice agent, installed and operated for you
An AI switchboard (reception, appointments, routing, transfer), connected to your existing number. Prices exclude VAT — annual subscription, the time it takes for the gains to settle in.
Agility
Setup + controlled subscription
10,615 € excl. VAT setup
then 856 € excl. VAT/month — you invest at installation and pay a reduced subscription. Ideal for keeping the cost under control over time.
- Installation, configuration and training for your teams
- Operation, human oversight, updates and support
- Sovereign hosting in France, a dedicated and isolated resource
The simplest
Serenity
All inclusive, no setup fee
1,447 € excl. VAT /month
all inclusive, immediate start. No upfront investment: a single subscription. Ideal for starting quickly and simply.
- Setup included (installation, configuration, training)
- Operation, human oversight, updates and support
- Sovereign hosting in France, managed end to end
100% Sovereign
On site, you own it
15,470 € excl. VAT setup
then 1,111 € excl. VAT/month · + hardware from 5,500 € (one-off purchase, in addition) — a sovereign computer installed on your premises, maintained remotely. Models run locally, your data returned at the end of the contract. 36-month commitment.
- Hardware installed on your premises (you own it)
- French / European AI models run locally
- Secure remote maintenance (Pro support included)
Not included in the packages: AI consumption (model tokens), re-invoiced at real cost with no margin, and tracked in real time in your client area. Maintenance and supervision subscription for an initial term of 12 months for the Agility package, 24 months for the Serenity package and 36 months for the 100% Sovereign package, renewable; support levels (SLA 72 h / 24 h / 4 h) optional. Bespoke development, additional integrations or exceptional volumes are quoted separately. Support Monday to Friday, 9am to 6pm. Prices exclude VAT.
AI model:
none of the AI models offered currently carries a fixed surcharge. When the selected model carries a cost, that cost is shown when you choose it, before you order, and re-invoiced at the cost incurred, with no mark-up; usage is billed at the publisher's price. Publishers' prices are published in US dollars: the amount re-invoiced is the amount in euros actually borne by Blue Lemon Agent on the publisher's invoice, at that invoice's exchange rate, with no commission or mark-up.
Included components and additional components Components included in the base offer: the Blue Lemon Agent software foundation, the AI models listed in the order journey, the standard channels (Microsoft Teams, Slack, WhatsApp Business, email, website chat, calendars, Microsoft 365 / Google Workspace, file storage, market VoIP telephony, professional social-media pages and accounts, Google Business Profile), hosting in France for the package chosen, backups, supervision, updates and support. If adapting the AI agent to your constraints, your needs or your requests requires other paid components — a third-party publisher's software licence, paid API access to one of your applications, hosting of health data, for which French law requires an HDS-certified host (art. L. 1111-8 of the French Public Health Code), SecNumCloud-qualified hosting, a speech synthesis service, particular hardware —, they are offered to you as an option or on quotation and re-invoiced at the cost incurred; nothing is committed without your written agreement. Where the artificial intelligence model you choose entails an additional cost, that cost is shown to you before you order and re-invoiced to you at the cost incurred, with no margin.
What to expect
Go-live
2 to 3 weeks
Agent designed, channels connected, team trained.
Steady state
4 to 7 weeks
After a few weeks of real use, once the agent's behaviour matches what you expect.
Indicative estimate, adjusted to the options you keep. It is not a delivery commitment.
Our commitment
Four guarantees that matter to your switchboard
Call data never leavesLocal inference or an isolated resource hosted in France; no recording entrusted to a foreign third party.
Data in France, under French lawExposure of call data to the Cloud Act and FISA 702 is reduced by design, that data being located and minimised in France; location alone does not guarantee immunity.
The human takes overThe agent transfers complex or sensitive cases with the context; no sensitive decision is automated.
Human oversight & traceabilityCallers informed, calls logged and updates applied: compliant with the requirements of the AI Act.
Frequently asked questions
Your questions, our answers
Does the voice agent really understand callers?
Yes. Unlike a touch-tone menu system, it understands plain language: the caller explains what they need as they would to a person, and the agent answers or routes accordingly. It copes with hesitations and rephrasing.
Can it book appointments?
Yes. Connected to your diary, it proposes slots, records the appointment, and sends a confirmation and a reminder — which also cuts missed appointments.
What happens with a complex call?
The agent transfers the call to the right person along with the context, or takes a detailed message if nobody is available. The transfer rules are defined with you.
Are recordings and call data protected?
Yes. The agent and the associated data are hosted in France, with the deployment objective of processing and access operated within the European Union and an architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity, in compliance with the GDPR and with callers being informed. For healthcare organisations, HDS health-data hosting is available.
Do we have to change our phone system?
In most cases, no. The agent connects to your existing number through your operator or a cloud telephony solution. We look at your configuration during scoping.
How is this better than a classic voice menu?
A classic voice menu is limited to rigid touch-tone menus and simple routing. The AI voice agent understands the conversation, books appointments, transfers with context, answers around the clock — with data hosted in France, with a deployment governed with regard to the GDPR.
How long does it take to deploy the agent?
Usually two to three weeks, after a free 15-minute audit. Steering is then done through a web dashboard (call log, reasons, transfers).
Does the agent state that it is an artificial intelligence?
Yes, from the opening words, 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, and the caller can ask to speak to a human at any time.
Are calls recorded, and can someone object?
Recording is neither permanent nor systematic: it is switched on for a specific purpose, announced at the start of the call. Both the caller and the employee whose voice is captured are informed of that purpose and of their right to object, and that right must be exercisable before the end of the call: the agent acts on it immediately and carries on without recording. The French data protection authority recommends a retention period of six months at most; that is the value delivered by default, and it can be shortened at your request.
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 customer side, the agent answers on WhatsApp Business, your website chat, email and VoIP telephony — your customers reach you where they already write, with no account to create. 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, and the gateway bills the text messages sent — are passed on at actual cost, with no margin, outside the subscription.
Can the agent notify my customers by text message?
Yes, as an option. The agent confirms by text message the appointment booked over the phone, then sends a reminder the day before. A text message is an outbound notification channel, not a conversation channel: no one writes back to the agent there — the conversation stays on WhatsApp Business, your website chat, email or VoIP telephony. The message announces without detailing, it carries your registered sender name, and it never contains a payment link, a login or a password: the recipient reaches your account area the way they usually do. Delivery goes through a gateway established in the European Union; the messages sent are passed on to you at actual cost, with no margin, and the channel itself is an optional module, set out on the pricing page.
Related use cases
Other agents for your customer relations
Let's talk
Let us estimate the potential for your switchboard
15 minutes to estimate the calls won back and the time freed up — hosted in France, supervised, with no commitment.