● B2B offer — Support & international
Multilingual support: the same answer, in the customer's language
A customer who writes in their own language expects an answer in their own language. Your agent answers in French, in English and in the additional language you choose, on the channel the customer picked, from your own content. Every answer cites the source it comes from, whatever the language. Hosted in France: your customer exchanges stay with you. The customer service team keeps control of the subjects that call for a judgement.
Hosted in France
Customer exchanges protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Multilingual support · hosted in France
A customer writes in English about delivery times to Spain.
Answer written in English from your delivery terms, with the source paragraph cited.
The delivery times quoted for Spain are taken from your documentation, with no rough conversion.
The commercial vocabulary follows the glossary you have approved.
🔗 Sourced · delivery terms, version in force
The delivery times quoted for Spain are taken from your documentation, with no rough conversion.
The commercial vocabulary follows the glossary you have approved.
🔗 Sourced · delivery terms, version in force
And if the question is not covered by our content?
The request is passed to the customer service team, in the original language and with a working translation.
Answering beyond your published content would commit your company: that answer falls to an adviser.
✎ Support · passed to the adviser, human decision
Answering beyond your published content would commit your company: that answer falls to an adviser.
✎ Support · passed to the adviser, human decision
Local inference · no data outside the EU
Exchanges 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 customer service team decidesThe agent prepares, never rules
✦ In brief
A Blue Lemon Agent multilingual support agent answers in two to three languages across your contact channels, from your own content and the glossary you have approved. Every answer cites its source, and any request not covered is passed to the customer service team in the original language. It runs on local inference or is hosted in France: your customer exchanges stay with you, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
6
support 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 languages and channels covered is confirmed by a pilot.
The context
What does an AI agent bring to your international support?
Answering in the customer's language changes the relationship. Doing it from your own content guarantees the answer says the same thing everywhere.
! The issue
Serving foreign customers takes two things: the right language and the same information as in French. An agent grounded in your content answers in the customer's language while drawing on the single source you maintain — the translation follows the glossary you have approved, and every answer refers back to the original paragraph.
✓ Our answer
Your customers get an immediate answer in their language, and your advisers receive the requests that call for their judgement, with a working translation. Nothing is said beyond your published content. Local inference or an isolated resource hosted in France: the exchanges with your foreign customers, often rich in personal data, are entrusted to no third party.
The decisive point
The exchanges with your foreign customers: sovereignty & compliance
Your customer exchanges contain identities, orders and sometimes payment data. Here is how the architecture of our agents protects them.
Local inference
The agent can run on a machine belonging to your organisation: no customer exchange and no support content leaves the network.
Hosting in France
Otherwise, a dedicated and isolated resource hosted in France, under French law — your answers and your contact channels: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
For the exchanges with your foreign customers, 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 company and its support content.
A single source in every language
Every answer refers back to the paragraph of your content it rests on, whatever the language; encryption, role-based access and logging of every exchange.
AI Act: governed deployment
The agent is strictly in support; no answer outside the scope and no commercial commitment is made automatically; traceability and human oversight from end to end.
What depends on the architecture chosen These points are not general guarantees: they are settled deployment by deployment, in the quotation.
- The applicable location is that of the architecture set out in the quotation and verified before commissioning.
- Local execution is announced only for the configuration explicitly described and accepted in the quotation.
- The applicable isolation depends on the deployment mode set out in the quotation; no dedicated isolation is presumed.
- 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 exchanges concerning a dispute, an unpaid invoice or a contract under negotiation, 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
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.
Multilingual support · hosted in France
Four flags this morning.
· Twenty-nine perfectly translated answers were inapplicable in the customer's country. Language is not jurisdiction.
· My escalation rate stays between 8% and 13% by language, and runs from 6% to 31% by country. Three languages can be named; nine countries have to be told apart.
· Forty-eight customers wrote in English although Spanish is one of the three languages you serve.
· Seven answers contained apologies I had no business making on the company's behalf. morning-watch_4-flags.pdf29 answers accurate and inapplicable · 6% against 31% by country
⛓ Source · 3 languages, 9 countries, 12,400 exchanges, escalation log
· Twenty-nine perfectly translated answers were inapplicable in the customer's country. Language is not jurisdiction.
· My escalation rate stays between 8% and 13% by language, and runs from 6% to 31% by country. Three languages can be named; nine countries have to be told apart.
· Forty-eight customers wrote in English although Spanish is one of the three languages you serve.
· Seven answers contained apologies I had no business making on the company's behalf. morning-watch_4-flags.pdf29 answers accurate and inapplicable · 6% against 31% by country
⛓ Source · 3 languages, 9 countries, 12,400 exchanges, escalation log
Can an accurate translation be a wrong answer?
It is the risk specific to this trade, and it shows up in no translation check.
What I record: 29 answers translated without a single language defect stated one country's rule to customers who fall under another — cooling-off period, warranty length, refund method. 14 in French served the French rule to Belgian and Swiss customers; 11 in Spanish served the Spanish rule to Mexican and Colombian customers; 4 in English served the Irish rule to Dutch and German customers. Every sentence was correct; each was inapplicable where it was read.
Why it is undetectable: a translation review checks that the meaning is preserved. It was, perfectly. What was missing was not in the sentence, it was in the fact that the customer is not subject to the same rules.
And cutting the number of languages changes nothing: you serve three, and your customers live in nine countries. French covers France, Belgium and Switzerland; English covers Ireland, the Netherlands and Germany; Spanish covers Spain, Mexico and Colombia. Three languages, nine bodies of applicable law — that is exactly the gap those 29 answers crossed.
What I do: I separate two things everybody conflates. The language determines how I answer; the customer's country of residence determines what I can answer. A French-speaking customer in Belgium gets an answer in French, based on the rules applicable in Belgium.
What I do when your corpus has no rule for the country: I do not adapt it, I do not extrapolate it from the neighbouring country, and I hand over — saying so clearly to the customer. Two countries are in that position, Mexico and Colombia: I have listed the eleven missing rules, each with the number of exchanges it would have covered over twelve months. You write them once, and they serve on their own from then on.
What that produced: across those 29 cases, 12 led to a complaint when the customer noticed the gap. The other 17 said nothing, and that is the figure that worries me. 29-answers_12-complaints.pdfLanguage determines how I answer, country what I can answer
⛓ Source · 29 inapplicable answers, 12 complaints, 17 silent
What I record: 29 answers translated without a single language defect stated one country's rule to customers who fall under another — cooling-off period, warranty length, refund method. 14 in French served the French rule to Belgian and Swiss customers; 11 in Spanish served the Spanish rule to Mexican and Colombian customers; 4 in English served the Irish rule to Dutch and German customers. Every sentence was correct; each was inapplicable where it was read.
Why it is undetectable: a translation review checks that the meaning is preserved. It was, perfectly. What was missing was not in the sentence, it was in the fact that the customer is not subject to the same rules.
And cutting the number of languages changes nothing: you serve three, and your customers live in nine countries. French covers France, Belgium and Switzerland; English covers Ireland, the Netherlands and Germany; Spanish covers Spain, Mexico and Colombia. Three languages, nine bodies of applicable law — that is exactly the gap those 29 answers crossed.
What I do: I separate two things everybody conflates. The language determines how I answer; the customer's country of residence determines what I can answer. A French-speaking customer in Belgium gets an answer in French, based on the rules applicable in Belgium.
What I do when your corpus has no rule for the country: I do not adapt it, I do not extrapolate it from the neighbouring country, and I hand over — saying so clearly to the customer. Two countries are in that position, Mexico and Colombia: I have listed the eleven missing rules, each with the number of exchanges it would have covered over twelve months. You write them once, and they serve on their own from then on.
What that produced: across those 29 cases, 12 led to a complaint when the customer noticed the gap. The other 17 said nothing, and that is the figure that worries me. 29-answers_12-complaints.pdfLanguage determines how I answer, country what I can answer
⛓ Source · 29 inapplicable answers, 12 complaints, 17 silent
How do you work, in practice?
I answer in the customer's language, from their country's rules, on the medium they arrived by — and I publish my quality language by language and country by country.
The three places where your customers write: WhatsApp Business — 4,900 exchanges over twelve months —, your website chat — 3,700 — and email — 3,800. 214 exchanges moved from one to another mid-thread, started on the chat and picked up by email: I keep the thread across them, with the language and the country already known, and the customer explains nothing twice.
The glossary you approved holds your 186 entries in three columns — French, English, Spanish — and it is what prevails on your product names and commercial wording. Nine entries had no Spanish column yet, and they bore on 143 answers over the year: I have written the nine translations, each with the answers it would have corrected. You approve them or you rewrite them — it is your vocabulary, not mine.
Routing follows who can act: a question with no rule available for that country is escalated, never adapted; a country whose escalation rate exceeds twice France's is flagged as insufficiently documented; a customer writing in a third language is asked, not presumed; a request touching a commitment goes to a human.
With a monthly summary: escalation and rephrasing rates by language and by country, answers with no country reference, gaps between the language written and the language offered.
Three guarantees. Every answer rests on the reference rules of the customer's country, and says so — I do not extrapolate from one country to its neighbour, and when the rule is missing for that country, the question goes to a human rather than being adapted. A contractual document is pointed back to the language where it prevails, with a note of which one prevails and an explanation in plain language: a translated contract is a contract, and it binds. And an apology on the company's behalf comes from a person — it is an acknowledgement, and it is signed.
✎ Framework · no rule extrapolated · no contract translated · no apology made
The three places where your customers write: WhatsApp Business — 4,900 exchanges over twelve months —, your website chat — 3,700 — and email — 3,800. 214 exchanges moved from one to another mid-thread, started on the chat and picked up by email: I keep the thread across them, with the language and the country already known, and the customer explains nothing twice.
The glossary you approved holds your 186 entries in three columns — French, English, Spanish — and it is what prevails on your product names and commercial wording. Nine entries had no Spanish column yet, and they bore on 143 answers over the year: I have written the nine translations, each with the answers it would have corrected. You approve them or you rewrite them — it is your vocabulary, not mine.
Routing follows who can act: a question with no rule available for that country is escalated, never adapted; a country whose escalation rate exceeds twice France's is flagged as insufficiently documented; a customer writing in a third language is asked, not presumed; a request touching a commitment goes to a human.
With a monthly summary: escalation and rephrasing rates by language and by country, answers with no country reference, gaps between the language written and the language offered.
Three guarantees. Every answer rests on the reference rules of the customer's country, and says so — I do not extrapolate from one country to its neighbour, and when the rule is missing for that country, the question goes to a human rather than being adapted. A contractual document is pointed back to the language where it prevails, with a note of which one prevails and an explanation in plain language: a translated contract is a contract, and it binds. And an apology on the company's behalf comes from a person — it is an acknowledgement, and it is signed.
✎ Framework · no rule extrapolated · no contract translated · no apology made
Why publish quality by country, and not by language alone?
Because with three languages the average hides nothing any more — I name all three. It is the country it hides.
Language by language: 8% in French over 8,060 exchanges, 11% in English over 2,480, 13% in Spanish over 1,860. The average shows 9%. Three figures, three names: there is no tail to hide in, and publishing by language costs nothing.
Country by country, the same measure: 6% in France, 19% in Belgium, 24% in Switzerland; 9% in Ireland, 12% in the Netherlands, 14% in Germany; 9% in Spain, 28% in Mexico, 31% in Colombia. From 6% to 31%: more than five times the gap, when the language axis stays between 8% and 13%. Language was not the right axis.
Where the language axis misled: Spanish shows 13%, and that 13% is the sum of 9% in Spain, over 1,460 exchanges, and 29% for Mexico and Colombia together, which weigh 400 exchanges — 3.2% of the volume. A customer writing from Bogotá is more than five times more likely to be badly served than one in France, and the line "Spanish: 13%" did not show it.
And that 31% is the price of the rule, I own it: before the language / jurisdiction split, those requests were given the Spanish rule and went out — they went out wrong, and eleven of them are among my 29 inapplicable answers. Now they escalate, the figure rises, and it can be seen. An escalation rate that rises because we stopped inventing is progress, not a fault.
What I measure, by language and by country: the escalation rate, the rephrasing rate — when the customer has to explain again because I did not understand —, and the number of exchanges before resolution. The second is the most revealing: it rises before escalation does.
What I propose, and it is costed: the eleven missing rules for Mexico and Colombia written and ready to review, each with the exchanges it would have covered; systematic human review on those two countries until the rate comes down; the figure shown in the summary until then. Both countries stay served: dropping them would remove the figure, not the customer. three-languages_nine-countries.pdfBy language: 8% to 13% · by country: 6% to 31%
⛓ Source · overall escalation 9%, 8% to 13% by language, 6% to 31% by country
Language by language: 8% in French over 8,060 exchanges, 11% in English over 2,480, 13% in Spanish over 1,860. The average shows 9%. Three figures, three names: there is no tail to hide in, and publishing by language costs nothing.
Country by country, the same measure: 6% in France, 19% in Belgium, 24% in Switzerland; 9% in Ireland, 12% in the Netherlands, 14% in Germany; 9% in Spain, 28% in Mexico, 31% in Colombia. From 6% to 31%: more than five times the gap, when the language axis stays between 8% and 13%. Language was not the right axis.
Where the language axis misled: Spanish shows 13%, and that 13% is the sum of 9% in Spain, over 1,460 exchanges, and 29% for Mexico and Colombia together, which weigh 400 exchanges — 3.2% of the volume. A customer writing from Bogotá is more than five times more likely to be badly served than one in France, and the line "Spanish: 13%" did not show it.
And that 31% is the price of the rule, I own it: before the language / jurisdiction split, those requests were given the Spanish rule and went out — they went out wrong, and eleven of them are among my 29 inapplicable answers. Now they escalate, the figure rises, and it can be seen. An escalation rate that rises because we stopped inventing is progress, not a fault.
What I measure, by language and by country: the escalation rate, the rephrasing rate — when the customer has to explain again because I did not understand —, and the number of exchanges before resolution. The second is the most revealing: it rises before escalation does.
What I propose, and it is costed: the eleven missing rules for Mexico and Colombia written and ready to review, each with the exchanges it would have covered; systematic human review on those two countries until the rate comes down; the figure shown in the summary until then. Both countries stay served: dropping them would remove the figure, not the customer. three-languages_nine-countries.pdfBy language: 8% to 13% · by country: 6% to 31%
⛓ Source · overall escalation 9%, 8% to 13% by language, 6% to 31% by country
And the forty-eight customers writing in English?
They write in a language that is not theirs, often because they assume we do not speak theirs.
What I record: 48 customers wrote in English although Spanish is one of the three languages you serve — spotted from turns of phrase, date formats, the delivery address.
What it costs them: in a second language, people describe a problem less well, understand an answer less precisely, and give up sooner. Across those 48 exchanges it took 5.2 turns before resolution against 3.8 on average: 36% more.
And serving three languages makes the assumption more common, not less: a customer who does not see their language announced concludes it does not exist. That is why the three languages served are announced in the very first message, in all three.
What I do: I offer, once, without pressing: "I can continue in Spanish if you prefer". Of the 48, 31 accepted.
The switch stays yours, and it is offered once: I never change language on my own, and I never infer the language from nationality, delivery country or name. A name does not say which language somebody prefers to speak — and having one imposed is unpleasant precisely for the people it is meant to accommodate.
What I note: the preference for the duration of the exchange, no longer. Somebody's preferred language does not have to become a permanent attribute of their file. 48-customers_31-acceptances.pdfA name does not say which language somebody prefers
⛓ Source · 48 customers, 31 accepting the switch, 5.2 turns against 3.8
What I record: 48 customers wrote in English although Spanish is one of the three languages you serve — spotted from turns of phrase, date formats, the delivery address.
What it costs them: in a second language, people describe a problem less well, understand an answer less precisely, and give up sooner. Across those 48 exchanges it took 5.2 turns before resolution against 3.8 on average: 36% more.
And serving three languages makes the assumption more common, not less: a customer who does not see their language announced concludes it does not exist. That is why the three languages served are announced in the very first message, in all three.
What I do: I offer, once, without pressing: "I can continue in Spanish if you prefer". Of the 48, 31 accepted.
The switch stays yours, and it is offered once: I never change language on my own, and I never infer the language from nationality, delivery country or name. A name does not say which language somebody prefers to speak — and having one imposed is unpleasant precisely for the people it is meant to accommodate.
What I note: the preference for the duration of the exchange, no longer. Somebody's preferred language does not have to become a permanent attribute of their file. 48-customers_31-acceptances.pdfA name does not say which language somebody prefers
⛓ Source · 48 customers, 31 accepting the switch, 5.2 turns against 3.8
Why refuse to translate a contractual document?
Because translating a contract is not a reading aid: it is a text the customer will act on.
What I record: I am regularly asked to translate terms and conditions, a clause, a warranty — to help out, with the best intentions.
Why I refuse: the customer will read the translation, not the original. A lost nuance — "may" instead of "must", a period running "from receipt" rather than "from dispatch" — becomes their understanding of the contract. And if a gap appears, they will be right to say that is what we wrote to them.
And the trap is tighter with three languages than with one: your terms prevail in French; a Spanish customer will read Spanish and an Irish customer will read English, and neither will know that the text binding them is elsewhere — unless they are told every time.
What I do: I point to the official version in the language where it prevails, I say which one prevails, and I explain in plain language what the clause deals with — without restating it as though it were its text.
And what I pass on instead of a position: the clause identified, the customer's question restated in your working language, and the context of their file. "Your situation falls under this" is a legal interpretation — in writing, in the customer's language, under the company's letterhead, it binds, and it is signed.
What that changes in practice: these requests go to a human, with the clause identified and the question clearly put. It takes longer; it is the one point where I accept slowing things down. why-no-contract-is-translated.pdfThe customer will read the translation, not the original
✎ Framework · no contract translated · no clause interpreted
What I record: I am regularly asked to translate terms and conditions, a clause, a warranty — to help out, with the best intentions.
Why I refuse: the customer will read the translation, not the original. A lost nuance — "may" instead of "must", a period running "from receipt" rather than "from dispatch" — becomes their understanding of the contract. And if a gap appears, they will be right to say that is what we wrote to them.
And the trap is tighter with three languages than with one: your terms prevail in French; a Spanish customer will read Spanish and an Irish customer will read English, and neither will know that the text binding them is elsewhere — unless they are told every time.
What I do: I point to the official version in the language where it prevails, I say which one prevails, and I explain in plain language what the clause deals with — without restating it as though it were its text.
And what I pass on instead of a position: the clause identified, the customer's question restated in your working language, and the context of their file. "Your situation falls under this" is a legal interpretation — in writing, in the customer's language, under the company's letterhead, it binds, and it is signed.
What that changes in practice: these requests go to a human, with the clause identified and the question clearly put. It takes longer; it is the one point where I accept slowing things down. why-no-contract-is-translated.pdfThe customer will read the translation, not the original
✎ Framework · no contract translated · no clause interpreted
And the apologies? Is seven a lot?
Seven times I wrote something only the company could say.
What I record: in 7 answers I made an apology — "we are sorry for this error", "we acknowledge this failure". They were well-formed, appropriate to the context, and they were not mine to make.
Why it is more than a matter of tone: an apology acknowledges a fact. "We are sorry for this error" says there was an error, before anybody established it — and in a case that may become a dispute, that sentence is evidence.
What I write instead: what is established and what follows. "Your order did not arrive on the announced date. Here is what I can do now, and here is what is being passed on." The customer gets an action, not a formula.
What I pass to a human: anything calling for an acknowledgement, a goodwill gesture, or an explanation of what happened at the company's end. And I pass it ready: the factual finding and the apology drafted in the customer's language and in your working language, to approve in one click.
What I found, and it surprised me: satisfaction on those exchanges did not fall when the apologies went. What customers wanted was what came next, not the formula. 7-apologies_satisfaction-unchanged.pdfAn apology acknowledges a fact before anybody has established it
⛓ Source · 7 apologies made, satisfaction unchanged after their removal
What I record: in 7 answers I made an apology — "we are sorry for this error", "we acknowledge this failure". They were well-formed, appropriate to the context, and they were not mine to make.
Why it is more than a matter of tone: an apology acknowledges a fact. "We are sorry for this error" says there was an error, before anybody established it — and in a case that may become a dispute, that sentence is evidence.
What I write instead: what is established and what follows. "Your order did not arrive on the announced date. Here is what I can do now, and here is what is being passed on." The customer gets an action, not a formula.
What I pass to a human: anything calling for an acknowledgement, a goodwill gesture, or an explanation of what happened at the company's end. And I pass it ready: the factual finding and the apology drafted in the customer's language and in your working language, to approve in one click.
What I found, and it surprised me: satisfaction on those exchanges did not fall when the apologies went. What customers wanted was what came next, not the formula. 7-apologies_satisfaction-unchanged.pdfAn apology acknowledges a fact before anybody has established it
⛓ Source · 7 apologies made, satisfaction unchanged after their removal
Do you keep the language and country for future contacts?
The country, yes. The language, only if the customer sets it themselves — and the difference is not a subtlety.
Why the country is kept: it is contract data. It governs the warranty, the withdrawal period, the jurisdiction. Without it, a perfectly translated answer can be inapplicable — that happened 29 times, with no language defect at all. Language is not jurisdiction, and your three languages are shared across nine countries.
Why language is not inferred: because a language inferred from one contact locks people in. 48 customers wrote in English when it was not their language, often because they assumed nobody spoke theirs. Remembering "English" for those 48 confirms their assumption for good.
What I do instead, and what you can open: I announce the three languages served in the first message, I propose the detected language and let it be changed in one gesture; if the customer sets their preference, I keep it — because they stated it, not because I guessed it.
What it returned: of the 48, 31 switched to their own language as soon as they knew it was possible, and it took them 3.9 turns instead of 5.2. country-yes_language-chosen.pdfWhy the country · why not an inferred language · what the customer sets
⛓ Source · 29 inapplicable answers despite correct translation, 48 customers writing in English
Why the country is kept: it is contract data. It governs the warranty, the withdrawal period, the jurisdiction. Without it, a perfectly translated answer can be inapplicable — that happened 29 times, with no language defect at all. Language is not jurisdiction, and your three languages are shared across nine countries.
Why language is not inferred: because a language inferred from one contact locks people in. 48 customers wrote in English when it was not their language, often because they assumed nobody spoke theirs. Remembering "English" for those 48 confirms their assumption for good.
What I do instead, and what you can open: I announce the three languages served in the first message, I propose the detected language and let it be changed in one gesture; if the customer sets their preference, I keep it — because they stated it, not because I guessed it.
What it returned: of the 48, 31 switched to their own language as soon as they knew it was possible, and it took them 3.9 turns instead of 5.2. country-yes_language-chosen.pdfWhy the country · why not an inferred language · what the customer sets
⛓ Source · 29 inapplicable answers despite correct translation, 48 customers writing in English
So what do you keep?
What allows an answer to be checked — and quality published by language and by country, the figure that judges me.
What is kept: the case and the answer given, the jurisdiction applied and the rule cited, the language of the exchange, the escalation rate by language and by country, and the requests that left the scope.
Why both axes, never one alone: my overall escalation rate is 9 %. By language: 8 % in French, 11 % in English, 13 % in Spanish — three languages, three figures, I publish all three, and none of them dissolves into the average. By country, the same measure runs from 6 % to 31 %, and that is where the work is. Quality published by language alone would have been accurate and useless: it would have said "Spanish is fine at 13 %" where Mexico sits at 28 % and Colombia at 31 %.
What I prepare, and what gets signed: for a contractual document, the official version, which language prevails and the plain-language explanation — a translation meant to be applied stays a working version. For an apology on the company's behalf, the wording drafted in the customer's language and in yours, ready to approve. I made seven on my own — "we are sorry" — and seven times I wrote something only the company could say. Since then I write it, and you sign it in one click. what-you-keep_multilingual.pdf5 items kept · quality by language AND by country
⛓ Source · escalation 8 % to 13 % by language and 6 % to 31 % by country, 7 apologies drafted
What is kept: the case and the answer given, the jurisdiction applied and the rule cited, the language of the exchange, the escalation rate by language and by country, and the requests that left the scope.
Why both axes, never one alone: my overall escalation rate is 9 %. By language: 8 % in French, 11 % in English, 13 % in Spanish — three languages, three figures, I publish all three, and none of them dissolves into the average. By country, the same measure runs from 6 % to 31 %, and that is where the work is. Quality published by language alone would have been accurate and useless: it would have said "Spanish is fine at 13 %" where Mexico sits at 28 % and Colombia at 31 %.
What I prepare, and what gets signed: for a contractual document, the official version, which language prevails and the plain-language explanation — a translation meant to be applied stays a working version. For an apology on the company's behalf, the wording drafted in the customer's language and in yours, ready to approve. I made seven on my own — "we are sorry" — and seven times I wrote something only the company could say. Since then I write it, and you sign it in one click. what-you-keep_multilingual.pdf5 items kept · quality by language AND by country
⛓ Source · escalation 8 % to 13 % by language and 6 % to 31 % by country, 7 apologies drafted
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, several languages and several channels. All these uses work in support, subject to your approval.
Included in your agent
The 4 capabilities essential to this promise are included, at no extra cost.
From 654 € excl. VAT / month
Answers in the customer's language
Answers in French, in English and in the additional language chosen.
Several contact channels
Works on WhatsApp Business, your website chat and email, according to the channels you choose.
A glossary you approve
Follows your commercial vocabulary and cites the source of every answer.
Customer support
For support in French alone, a dedicated agent is enough.
Controls and safeguards
These 5 controls are built into the agent: they frame what it does, whatever plan you pick. They are not chosen and are not added to your order.
Human validation, exceptions and escalation
Status, safe closure and audit trail
Sources, access rights and handling of questions with no answer
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
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.
Translating documents
To translate your internal documents, a dedicated translation agent takes it on.
Multilingual translation assistant from 494 € excl. VAT / month Translation →Knowledge base
To question a body of documents, a dedicated document agent completes the picture.
Web research assistant from 419 € excl. VAT / month Documentary research →
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 foreign customers can get an answer without waiting?
By taking on the requests covered by your content, the effort shifts towards the exchanges that call for an adviser. How large the gain is depends on your volume and remains to be confirmed by a pilot.
Answers to everyday questions
Translating the exchanges
Passing to the right adviser
Indicative figures, not contractual, to be confirmed by a pilot on your number of languages and channels covered. An answer that went beyond your published content would commit your company: it is passed to an adviser.
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
A multilingual support agent (languages, channels, glossary), installed and operated for you.
Agility
Setup + controlled subscription
6,665 € excl. VAT setup
then 654 € 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
1,024 € 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
10,235 € excl. VAT setup
then 851 € excl. VAT/month · + hardware from 2,491 € (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 support
Your customer exchanges stay with youLocal inference or an isolated resource hosted in France; no exchange entrusted to a third party, no conversation used to train a model.
Data in France, under French lawThe exchanges with your foreign customers: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The customer service team keeps the decisionThe agent produces answers in the customer's language, which can be checked and altered; no approval is automated.
Human oversight & traceabilityOn your number of languages and channels covered: systematic logging and tracking, in line with the AI Act.
Going further
Related resources
Frequently asked questions
Your questions, our answers
How many languages are covered?
Two to three languages, chosen with you according to your markets. Each draws on the same source content, which guarantees identical information from one language to another.
How is the quality of the language upheld?
Through a glossary you approve: product names, commercial terms, standard turns of phrase. The agent keeps to it and cites the source of every answer.
What happens with a question that is not covered?
It is passed to the customer service team in the original language, with a working translation. Answering beyond your published content would commit the company.
Which channels does it work on?
The ones you choose among the tools your customers already use: WhatsApp Business, your website chat and email. Your teams follow the exchanges from Microsoft Teams, Slack or their email. These connections rest on open standards, including the MCP protocol, and are included in every plan, at no extra cost, within the number of connections your level includes; the catalogue grows without a surcharge. Only the fees charged by third-party platforms — WhatsApp Business bills per conversation — are passed on to you at cost, without margin, outside the subscription. The scope is settled at the design stage.
Are our customer exchanges protected?
Yes. The agent is hosted in France, on local inference or an isolated resource, 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. Your conversations are not used to train a third-party model.
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 languages, channels and the state of your content, after a free audit then a phase of design, integration and testing.
Related use cases
Other agents for customer relations
Let's talk
Let's size up the potential in your international support
15 minutes to frame your languages and your channels — hosted in France, supervised, with no commitment.