● B2B offer — Insurance & claims handling
Claims: the file built, the cover matched
A claim handled well starts with a complete file: the notification, the documents, the policy, the cover that applies. Your agent takes the notification according to your protocol, gathers the documents required, matches the policyholder's contract and prepares the file with the cover concerned. Hosted in France: your policyholders' personal data stays with you. The handler decides on acceptance and on settlement.
Hosted in France
Policyholder data protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Claims handling · hosted in France
Build the file for this water damage notification.
Notification taken according to your protocol: date, circumstances as stated, property concerned.
The policyholder's contract is matched, with the cover that may apply and its reference.
Three required documents are missing: the request is prepared.
🔗 Sourced · the notification, the contract and the handling protocol
The policyholder's contract is matched, with the cover that may apply and its reference.
Three required documents are missing: the request is prepared.
🔗 Sourced · the notification, the contract and the handling protocol
Is the claim accepted?
I set out the applicable cover, its reference in the contract and the elements of the file that bear on it.
Accepting a claim and setting the settlement concern money and the policyholder's rights: those decisions belong to the handler.
✎ Support · cover set out, the handler's decision
Accepting a claim and setting the settlement concern money and the policyholder's rights: those decisions belong to the handler.
✎ Support · cover set out, the handler's decision
Local inference · no data outside the EU
Files 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 claims handler decidesThe agent prepares, never rules
✦ In brief
A Blue Lemon Agent claims handling agent takes the notification according to your protocol, gathers the documents required, matches the policyholder's contract and sets out the applicable cover with its reference. It accepts no claim and sets no settlement. It runs on local inference or is hosted in France: your policyholders' data stays with you, 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 policyholders 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
claims handling 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 volume of notifications and the variety of your cover is confirmed by a pilot.
The context
What does an AI agent bring to your claims handling?
A complete file from day one shortens the whole process and lets the policyholder get an answer quickly.
! The issue
Building a claim file brings together a notification, documents and the policyholder's contract. That gathering work takes up the start of the process, while the decision turns on the cover. The agent takes on the gathering, asks for the missing documents and sets out the applicable cover with its reference in the contract.
✓ Our answer
The handler has a complete file and cover already matched, with the contractual reference for each. Accepting a claim and setting a settlement concern money and the policyholder's rights: those decisions stay human. Local inference or an isolated resource hosted in France: your policyholders' personal data is entrusted to no third party.
The decisive point
Policyholders' personal data: sovereignty & compliance
Claim notifications contain personal data, sometimes health data, and engage contractual rights. Here is how they are protected.
Local inference
The agent can run on a machine belonging to your organisation: no notification and no policyholder data leaves the network.
Hosting in France
Otherwise, a dedicated and isolated resource hosted in France, under French law — your claim notifications and your cover: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
For policyholders' personal 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 organisation and its handling protocol.
Every cover refers back to the contract
Any cover set out cites its reference in the policyholder's contract; encryption, role-based access and logging of every file built.
AI Act: governed deployment
The agent is strictly in support; no claim is accepted and no settlement is set 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 personal injury claims, disputed files and suspected fraud, 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 companyTernois Assurances — regional mutual insurer
- Sector
- Property and casualty insurance — home, artisans' commercial multi-risk, motor
- Headcount
- 96 staff, including 14 first-level claims handlers, 3 senior claims writers and 1 lawyer
- Policyholders served
- 78,000 policyholders — households and artisans — across three départements
- Volume
- 11,400 notifications a year, of which 4,200 escape of water and 1,900 commercial multi-risk; 62 standard contracts and 340 distinct items of cover
- Tools in place
- Claims handling system, document management system, online notification portal and telephone switchboard — the agent plugs into them, nothing is replaced
- Who decides
- The claims handler accepts the claim and sets the settlement; the head of department arbitrates sensitive files; the lawyer takes disputes and the medical adviser takes bodily injury claims
- Room for improvement
- A file takes 9.4 days to become complete; 41 % of notifications arrive with at least one document missing; an incomplete file takes 3 chasers on average, and each one adds 6 days
Ternois Assurances is not looking to settle differently, but to start work earlier: a file that is complete on day one shortens everything that follows. The agent runs on local inference on a machine at the mutual and plugs into the claims system, the document management system, the notification portal and the switchboard: it builds the file, the handler decides. The exchanges below cover one quarter, from the first night of notifications to the department's 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.
Claims handling · hosted in France
The 31 notifications that came in overnight have been taken and turned into files. Your handlers walk in to 31 open files, not 31 forms to re-key.
First level means the part of the work that comes before the decision: taking the notification, gathering the documents, matching the contract. The decision to accept comes afterwards, and it is yours.
The gap I measured in your own files, and it is the gap that decides the gain: across your 11,400 notifications last year, files whose circumstances were complete on day one settled in 12 days; the others in 34. It is not the complexity of the loss that separates the two groups — it is the date on which the file became complete.
What each of the 31 files holds this morning:
· The date of loss and the date of discovery, kept apart — the date of loss is when the damage happened, the date of discovery is when the policyholder noticed it; on an escape of water they are rarely the same day.
· The circumstances as the policyholder wrote them, word for word, without a comma moved.
· The property and premises concerned, taken from the contract where they appear in it.
· The contract matched, in the version applicable on the date of loss.
· The documents already supplied, filed, and the ones missing, listed.
The time this moves: taking and writing up a notification used to take 55 % of first-level handling time; it now takes 9 %. The remaining 9 % is your review, and that is the part that is worth something.
What I suggest: that your handlers open the day with the 6 files I have flagged — the ones where a missing document blocks everything else. The other 25 can wait until the afternoon at no cost. overnight-notifications_31-files-open.pdfWhat is in each file at 8 a.m.
⛓ Sourced · notification portal, claims handling system, 11,400 files from the previous year
First level means the part of the work that comes before the decision: taking the notification, gathering the documents, matching the contract. The decision to accept comes afterwards, and it is yours.
The gap I measured in your own files, and it is the gap that decides the gain: across your 11,400 notifications last year, files whose circumstances were complete on day one settled in 12 days; the others in 34. It is not the complexity of the loss that separates the two groups — it is the date on which the file became complete.
What each of the 31 files holds this morning:
· The date of loss and the date of discovery, kept apart — the date of loss is when the damage happened, the date of discovery is when the policyholder noticed it; on an escape of water they are rarely the same day.
· The circumstances as the policyholder wrote them, word for word, without a comma moved.
· The property and premises concerned, taken from the contract where they appear in it.
· The contract matched, in the version applicable on the date of loss.
· The documents already supplied, filed, and the ones missing, listed.
The time this moves: taking and writing up a notification used to take 55 % of first-level handling time; it now takes 9 %. The remaining 9 % is your review, and that is the part that is worth something.
What I suggest: that your handlers open the day with the 6 files I have flagged — the ones where a missing document blocks everything else. The other 25 can wait until the afternoon at no cost. overnight-notifications_31-files-open.pdfWhat is in each file at 8 a.m.
⛓ Sourced · notification portal, claims handling system, 11,400 files from the previous year
Our handlers often rewrite the circumstances to make them readable. Do you do the same?
I quote word for word, and I set my own reading apart: both are in the file, in two separate blocks. That is what makes the file stand up, and it is measurable.
Why, mechanically: rewritten circumstances do not just change the style, they change the date of loss and sometimes the cover. A policyholder who writes « I could hear water running for two or three days, I saw the stain on Monday » becomes, once tidied up, « leak found on Monday the 12th ». The sentence is clearer; it has made the most useful fact in the file disappear.
What that cost you: I read back 240 closed files from last year. 18 carried a rewrite that moved the date of loss, and 6 of those fell under a different item of cover than the one applied. Not one of them is carelessness: it is the mechanical result of tidying up.
What I deliver instead, on every file:
· Block 1 — the policyholder's words, verbatim, with the channel and the time stamp of the notification.
· Block 2 — my reading: dates retained, likely nature of the loss, cover to be examined, each line pointing back to the exact sentence that carries it.
· Block 3 — what remains ambiguous, and the question that would settle it. Of the 31 files from last night, 4 have one, and I have already drafted it.
The benefit for you is not about writing, it is about recovery: when you pursue a third party or their insurer, it is the time-stamped verbatim that stands up, not a summary. Your 3 senior writers used to reconstruct it by hand at the end of a file — it is now written from the first minute.
What I suggest next: that the 4 questions go out today rather than at first examination. A question asked on day 1 comes back on day 3 in your figures; asked on day 9, it comes back on day 16. circumstances_what-is-quoted-what-is-inferred.pdf18 rewrites found across 240 files
⛓ Sourced · time-stamped portal notifications, 240 closed files read back, applicable contracts
Why, mechanically: rewritten circumstances do not just change the style, they change the date of loss and sometimes the cover. A policyholder who writes « I could hear water running for two or three days, I saw the stain on Monday » becomes, once tidied up, « leak found on Monday the 12th ». The sentence is clearer; it has made the most useful fact in the file disappear.
What that cost you: I read back 240 closed files from last year. 18 carried a rewrite that moved the date of loss, and 6 of those fell under a different item of cover than the one applied. Not one of them is carelessness: it is the mechanical result of tidying up.
What I deliver instead, on every file:
· Block 1 — the policyholder's words, verbatim, with the channel and the time stamp of the notification.
· Block 2 — my reading: dates retained, likely nature of the loss, cover to be examined, each line pointing back to the exact sentence that carries it.
· Block 3 — what remains ambiguous, and the question that would settle it. Of the 31 files from last night, 4 have one, and I have already drafted it.
The benefit for you is not about writing, it is about recovery: when you pursue a third party or their insurer, it is the time-stamped verbatim that stands up, not a summary. Your 3 senior writers used to reconstruct it by hand at the end of a file — it is now written from the first minute.
What I suggest next: that the 4 questions go out today rather than at first examination. A question asked on day 1 comes back on day 3 in your figures; asked on day 9, it comes back on day 16. circumstances_what-is-quoted-what-is-inferred.pdf18 rewrites found across 240 files
⛓ Sourced · time-stamped portal notifications, 240 closed files read back, applicable contracts
And the ones that come in by phone or on paper? That is half of what we receive.
All three channels land in the same file, in the same format, on the same day. Paper was the one costing you the most, and it is the one that moves the most.
· Online portal: the file opens instantly. Nothing to do.
· Telephone: incoming calls are answered by a dedicated voice agent, which takes the notification according to your protocol and hands over to a handler as soon as the policyholder asks. It states that it is a machine in its first sentence, on every call — and that is not a setting: since 2 August 2026, article 50(1) of the European AI Regulation requires anyone interacting with an AI system to be informed.
· Post and paper documents: reading what comes in is handled by a dedicated document processing agent, which extracts the data and passes it to me. Your post used to take 2.3 days to become a file; it becomes one on the day it is opened.
What I never guess: a value that cannot be read on a scan. It stays empty and flagged, never filled in from a neighbouring file — a guessed value becomes indistinguishable from a read one three weeks later.
The figure that matters for your department: your 14 handlers were spending the equivalent of a half-time post every week re-keying what arrived by post and by phone. That half-time is back, and it went into working the older files.
What I suggest: opening the same route to the 1,900 commercial multi-risk notifications, where your artisans notify almost everything by phone. You set the date, I am ready in a week.
⛓ Sourced · switchboard logs, scanned incoming post, claims department timesheets
· Online portal: the file opens instantly. Nothing to do.
· Telephone: incoming calls are answered by a dedicated voice agent, which takes the notification according to your protocol and hands over to a handler as soon as the policyholder asks. It states that it is a machine in its first sentence, on every call — and that is not a setting: since 2 August 2026, article 50(1) of the European AI Regulation requires anyone interacting with an AI system to be informed.
· Post and paper documents: reading what comes in is handled by a dedicated document processing agent, which extracts the data and passes it to me. Your post used to take 2.3 days to become a file; it becomes one on the day it is opened.
What I never guess: a value that cannot be read on a scan. It stays empty and flagged, never filled in from a neighbouring file — a guessed value becomes indistinguishable from a read one three weeks later.
The figure that matters for your department: your 14 handlers were spending the equivalent of a half-time post every week re-keying what arrived by post and by phone. That half-time is back, and it went into working the older files.
What I suggest: opening the same route to the 1,900 commercial multi-risk notifications, where your artisans notify almost everything by phone. You set the date, I am ready in a week.
⛓ Sourced · switchboard logs, scanned incoming post, claims department timesheets
41 % of our notifications arrive with a document missing. We chase, it drags, and the policyholder gets angry.
It is not the missing documents that cost you: it is the NUMBER of requests. I counted, and the figure is clean.
Across your incomplete files last year: 3 chasers on average, and each chaser adds 6 days. Eighteen days lost, not because the policyholder is slow, but because the documents were asked for one after another, as each examination discovered another one.
What I have done: I drew up, by nature of loss and by contract, the list of documents required — 62 standard contracts, 340 items of cover, 19 lists of documents — then matched them against what is already there.
Across the 31 files from last night:
· 18 are complete and go straight to examination.
· 9 are waiting on a single document, and the request is written for all 9, addressed by name, with the upload link.
· 4 are waiting on two or three: a single grouped request, never three messages.
The time this moves: checking the documents in a file used to take 35 % of handling time; it now takes 7 %.
What every request states, and it is what brings the document back: the item of cover that document establishes, spelled out, instead of « please send us your supporting documents ». A policyholder who is told what the document is for replies about twice as fast — and that is checkable in your own returns.
What I suggest: that you review the 13 requests this morning. Sent today, they come back before Friday; sent at first examination, they come back next week. documents-required_by-nature-of-loss.pdf19 lists, every document tied to its cover
⛓ Sourced · incomplete files from the year, standard contracts, chaser log
Across your incomplete files last year: 3 chasers on average, and each chaser adds 6 days. Eighteen days lost, not because the policyholder is slow, but because the documents were asked for one after another, as each examination discovered another one.
What I have done: I drew up, by nature of loss and by contract, the list of documents required — 62 standard contracts, 340 items of cover, 19 lists of documents — then matched them against what is already there.
Across the 31 files from last night:
· 18 are complete and go straight to examination.
· 9 are waiting on a single document, and the request is written for all 9, addressed by name, with the upload link.
· 4 are waiting on two or three: a single grouped request, never three messages.
The time this moves: checking the documents in a file used to take 35 % of handling time; it now takes 7 %.
What every request states, and it is what brings the document back: the item of cover that document establishes, spelled out, instead of « please send us your supporting documents ». A policyholder who is told what the document is for replies about twice as fast — and that is checkable in your own returns.
What I suggest: that you review the 13 requests this morning. Sent today, they come back before Friday; sent at first examination, they come back next week. documents-required_by-nature-of-loss.pdf19 lists, every document tied to its cover
⛓ Sourced · incomplete files from the year, standard contracts, chaser log
Who sends the request to the policyholder? We do not want a letter going out in the mutual's name on its own.
Today, you do. And if you want it to go without you, that is not a technical setting: it is a written mandate you give me, capped, dated, and withdrawn with a single word.
The current arrangement: I build the request, it waits in your handler's queue, he sends it in one click. Measured delay: 4.1 days between the file opening and the first request — the time it takes for the file to come back up the pile.
The mandate I am proposing, written out in full:
· Scope: requests for documents, and nothing else — no decision letter, no refusal, no settlement offer.
· Types of loss covered: escape of water and broken glass only. No bodily injury, no dispute, no suspected fraud: those three are not in the mandate and cannot be added to it.
· Templates: 3 templates your head of department approves once, to the comma. I write not one sentence outside them.
· Cap: 60 sends a week. Beyond that, everything waits for a person — an abnormal volume is first a sign of a problem, not of a need.
· Term: until 31 December, then it must be signed again. A mandate with no end date is a mandate no one reads again.
· Withdrawal: one word, at any time, with no notice and no effect on files under way.
· Log: every send carries the mandate reference, the date and the file. A written mandate IS human approval — given once, for a named scope, and checkable line by line.
What it produced at the two departments that signed it: the first request goes out in 40 minutes instead of 4.1 days. Across 4,200 escape-of-water claims a year, that is close to four days gained on every incomplete file, without a handler touching anything.
The decision stays yours: you sign, or you keep the click. Either way the request is written this morning. mandate_sending-document-requests.pdfScope, cap, term, withdrawn with one word
✎ Framework · draft mandate, departmental letter templates, send log
The current arrangement: I build the request, it waits in your handler's queue, he sends it in one click. Measured delay: 4.1 days between the file opening and the first request — the time it takes for the file to come back up the pile.
The mandate I am proposing, written out in full:
· Scope: requests for documents, and nothing else — no decision letter, no refusal, no settlement offer.
· Types of loss covered: escape of water and broken glass only. No bodily injury, no dispute, no suspected fraud: those three are not in the mandate and cannot be added to it.
· Templates: 3 templates your head of department approves once, to the comma. I write not one sentence outside them.
· Cap: 60 sends a week. Beyond that, everything waits for a person — an abnormal volume is first a sign of a problem, not of a need.
· Term: until 31 December, then it must be signed again. A mandate with no end date is a mandate no one reads again.
· Withdrawal: one word, at any time, with no notice and no effect on files under way.
· Log: every send carries the mandate reference, the date and the file. A written mandate IS human approval — given once, for a named scope, and checkable line by line.
What it produced at the two departments that signed it: the first request goes out in 40 minutes instead of 4.1 days. Across 4,200 escape-of-water claims a year, that is close to four days gained on every incomplete file, without a handler touching anything.
The decision stays yours: you sign, or you keep the click. Either way the request is written this morning. mandate_sending-document-requests.pdfScope, cap, term, withdrawn with one word
✎ Framework · draft mandate, departmental letter templates, send log
And chasing? Some policyholders never reply, and we end up forgetting the file.
No file is forgotten, and above all: a chaser stops on its own the second the document arrives — including at 10 p.m. on a Saturday.
The rhythm I propose, drawn from your own return times: D+4 through the channel the policyholder used to notify · D+10 through another channel · D+15, a call from the voice agent, because after two written attempts with no reply, it is no longer forgetfulness, it is an obstacle — and an obstacle is cleared by talking.
The two automatic actions, and both are reversible in either direction:
· The document arrives, the sequence stops instantly. And the reverse holds too: if the document is unreadable or incomplete, I ask again for that one document, on its own, never for the whole list. Asking again for everything already supplied is the surest way to lose a policyholder acting in good faith.
· The policyholder replies with something other than a document — a question, a challenge, bad news: the sequence stops as well, and the message goes to your handler. A chaser arriving after a reply tells the person that nobody read them.
What the quarter produced: files complete at D+7 rise from 38 % to 79 %, and 2,100 chasers did not need to be sent — either the document had already arrived, or the grouped request had been enough first time.
The figure beyond the delay: your 14 handlers recovered the equivalent of 22 working days over the quarter on this one task. They put it into the 340 files older than two months — the ones that generate complaints.
What I suggest next: allowing documents to be uploaded as a simple photo from the link in the request. Among artisans, 6 documents out of 10 are already photographed on a phone then re-scanned at the branch — we may as well receive them directly. document-tracking_what-stops-a-chaser.pdf38 % → 79 % of files complete at D+7
⛓ Sourced · request and return log, reply times by channel, departmental backlog
The rhythm I propose, drawn from your own return times: D+4 through the channel the policyholder used to notify · D+10 through another channel · D+15, a call from the voice agent, because after two written attempts with no reply, it is no longer forgetfulness, it is an obstacle — and an obstacle is cleared by talking.
The two automatic actions, and both are reversible in either direction:
· The document arrives, the sequence stops instantly. And the reverse holds too: if the document is unreadable or incomplete, I ask again for that one document, on its own, never for the whole list. Asking again for everything already supplied is the surest way to lose a policyholder acting in good faith.
· The policyholder replies with something other than a document — a question, a challenge, bad news: the sequence stops as well, and the message goes to your handler. A chaser arriving after a reply tells the person that nobody read them.
What the quarter produced: files complete at D+7 rise from 38 % to 79 %, and 2,100 chasers did not need to be sent — either the document had already arrived, or the grouped request had been enough first time.
The figure beyond the delay: your 14 handlers recovered the equivalent of 22 working days over the quarter on this one task. They put it into the 340 files older than two months — the ones that generate complaints.
What I suggest next: allowing documents to be uploaded as a simple photo from the link in the request. Among artisans, 6 documents out of 10 are already photographed on a phone then re-scanned at the branch — we may as well receive them directly. document-tracking_what-stops-a-chaser.pdf38 % → 79 % of files complete at D+7
⛓ Sourced · request and return log, reply times by channel, departmental backlog
We have 62 standard contracts and 340 items of cover. Even our old hands do not know them all by heart.
Nobody can know them all, and that is not the point: the point is knowing WHICH one applies to THIS file, in the version of the contract that was in force on the day of the loss.
What I match, on escape-of-water file no. 2611:
· The policyholder's contract in the version applicable on the date of loss, not today's version. That point alone explains 3 of the 18 discrepancies I found in your closed files: the contract had been amended between the loss and its examination.
· The items of cover that may apply — three here —, each with its reference in the contract: the clause, the exact wording, the excess and the limit.
· The document in the file that triggers each one. An item of cover that no document supports is presented as such, together with the document missing to support it.
· The exclusions to check, each set against the element of the file that rules it out or does not.
On an ambiguous clause I do the whole job rather than choose on your behalf: when two readings of the same clause lead to two different items of cover, I give you BOTH, what separates them, what each is worth to the file, and the exact document that would settle it — which I have already requested. Smoothing two readings into one would erase exactly the information the handler decides on.
The time this moves: matching the contract and the cover used to take 40 % of handling time; it now takes 8 %.
What it gives you beyond time: your 3 senior writers had become the bottleneck for every question of cover. Over the quarter they were called on 61 times instead of 380 — and the remaining 61 are the real ones: the ambiguous clauses, where their opinion is worth money.
What I suggest next: that every ruling a senior writer gives on an ambiguous clause becomes a written departmental position, dated and signed. Of the 61 this quarter, 23 concern the same dozen clauses — once written down, I apply them on my own and the writer never sees them again. applicable-cover_file-EW-2611.pdf3 items of cover, their reference, the document that triggers each
⛓ Sourced · standard contracts across their successive versions, the policyholder's endorsements, documents in file 2611
What I match, on escape-of-water file no. 2611:
· The policyholder's contract in the version applicable on the date of loss, not today's version. That point alone explains 3 of the 18 discrepancies I found in your closed files: the contract had been amended between the loss and its examination.
· The items of cover that may apply — three here —, each with its reference in the contract: the clause, the exact wording, the excess and the limit.
· The document in the file that triggers each one. An item of cover that no document supports is presented as such, together with the document missing to support it.
· The exclusions to check, each set against the element of the file that rules it out or does not.
On an ambiguous clause I do the whole job rather than choose on your behalf: when two readings of the same clause lead to two different items of cover, I give you BOTH, what separates them, what each is worth to the file, and the exact document that would settle it — which I have already requested. Smoothing two readings into one would erase exactly the information the handler decides on.
The time this moves: matching the contract and the cover used to take 40 % of handling time; it now takes 8 %.
What it gives you beyond time: your 3 senior writers had become the bottleneck for every question of cover. Over the quarter they were called on 61 times instead of 380 — and the remaining 61 are the real ones: the ambiguous clauses, where their opinion is worth money.
What I suggest next: that every ruling a senior writer gives on an ambiguous clause becomes a written departmental position, dated and signed. Of the 61 this quarter, 23 concern the same dozen clauses — once written down, I apply them on my own and the writer never sees them again. applicable-cover_file-EW-2611.pdf3 items of cover, their reference, the document that triggers each
⛓ Sourced · standard contracts across their successive versions, the policyholder's endorsements, documents in file 2611
So is this file accepted or not?
The file is ready to decide and the step that is missing is already prepared — the acceptance itself is pronounced by your handler: it is money, and it is a policyholder's right.
The file is ready to decide: the three items of cover are matched, two are beyond argument, the third turns on a single fact — whether the leak originates in the private part or the common part of the building. That fact is in none of the documents in the file; it has to be established on site.
What is blocking, named precisely: until that inspection happens, your handler decides on an assumption. In your figures, that inspection takes 5.2 days: the time it takes for someone to think of instructing a loss adjuster.
The route, and it is already prepared: the instruction letter to the adjuster is written, with the precise question to settle — not « inspect the loss », but « establish whether the leak originates in the private or the common part ». An adjuster asked a question comes back with an answer; an adjuster asked for a report comes back with a report.
Instructing an adjuster commits your money, so it is a mandate, and here are its limits: €350 per file maximum · 30 files a month · escape of water and damage to property only · until 31 December · withdrawn with one word · and I notify you on every instruction, with the file and the amount. Beyond any one of those caps, everything waits for your signature.
What it is worth, measured at the two departments that signed it: instruction goes from 5.2 days to 2 hours, and final settlement lands 9 days earlier. Across 1,100 files a year involving an adjuster, that is nine days less waiting for every policyholder, without a euro more spent — the inspections were being ordered anyway, simply later.
What I suggest: the mandate for escape of water first, over three months, then we look at the first 30 files together before extending it. mandate_instructing-a-loss-adjuster.pdf€350 per file, 30 a month, withdrawn with one word
✎ Framework · draft adjuster mandate, instruction letters, instruction delays for the year
The file is ready to decide: the three items of cover are matched, two are beyond argument, the third turns on a single fact — whether the leak originates in the private part or the common part of the building. That fact is in none of the documents in the file; it has to be established on site.
What is blocking, named precisely: until that inspection happens, your handler decides on an assumption. In your figures, that inspection takes 5.2 days: the time it takes for someone to think of instructing a loss adjuster.
The route, and it is already prepared: the instruction letter to the adjuster is written, with the precise question to settle — not « inspect the loss », but « establish whether the leak originates in the private or the common part ». An adjuster asked a question comes back with an answer; an adjuster asked for a report comes back with a report.
Instructing an adjuster commits your money, so it is a mandate, and here are its limits: €350 per file maximum · 30 files a month · escape of water and damage to property only · until 31 December · withdrawn with one word · and I notify you on every instruction, with the file and the amount. Beyond any one of those caps, everything waits for your signature.
What it is worth, measured at the two departments that signed it: instruction goes from 5.2 days to 2 hours, and final settlement lands 9 days earlier. Across 1,100 files a year involving an adjuster, that is nine days less waiting for every policyholder, without a euro more spent — the inspections were being ordered anyway, simply later.
What I suggest: the mandate for escape of water first, over three months, then we look at the first 30 files together before extending it. mandate_instructing-a-loss-adjuster.pdf€350 per file, 30 a month, withdrawn with one word
✎ Framework · draft adjuster mandate, instruction letters, instruction delays for the year
And when it smells of a dispute, or worse, of fraud? That is where we want no automation at all.
Those three families reach the right person on day one instead of day nine — and they enter no automated handling at all: that is by design, not by setting.
What I pass on, and to whom:
· Bodily injury: the whole file to the medical adviser, without my reading or copying a single health document. They stay where they were filed, and I have no access to them.
· Declared or likely dispute: the whole file to the lawyer, with the time-stamped verbatim — that is the piece that counts for them.
· Anything deserving a second look: the whole file to the head of department.
What goes into the policyholder's file, and it is a guarantee you can hold up to them: dated facts, and nothing else. No legal characterisation, no label — no « suspect » marker, no score, nothing that stays. Mechanically: a label written one day follows the policyholder for years and steers the next handler's eye, and that handler will never know what it rested on. I pass on dated facts — three claims on the same cover in fourteen months, a notification made on the day cover started — and it is your teams who characterise them, or do not.
The gain, and it works both ways: over the quarter, 214 files went to the right person on day one, against day nine on average before. For the lawyer that is eight more days to act; for the honest policyholder it is eight fewer days of waiting.
What I suggest: that the head of department review the 12 routing criteria once a quarter — they are written down, short, and they belong to you. None of them changes unless you write the change. routing_bodily-injury-dispute-second-look.pdf12 written criteria, no label in the file
✎ Framework · departmental routing rules, transfer log, the lawyer's caseload
What I pass on, and to whom:
· Bodily injury: the whole file to the medical adviser, without my reading or copying a single health document. They stay where they were filed, and I have no access to them.
· Declared or likely dispute: the whole file to the lawyer, with the time-stamped verbatim — that is the piece that counts for them.
· Anything deserving a second look: the whole file to the head of department.
What goes into the policyholder's file, and it is a guarantee you can hold up to them: dated facts, and nothing else. No legal characterisation, no label — no « suspect » marker, no score, nothing that stays. Mechanically: a label written one day follows the policyholder for years and steers the next handler's eye, and that handler will never know what it rested on. I pass on dated facts — three claims on the same cover in fourteen months, a notification made on the day cover started — and it is your teams who characterise them, or do not.
The gain, and it works both ways: over the quarter, 214 files went to the right person on day one, against day nine on average before. For the lawyer that is eight more days to act; for the honest policyholder it is eight fewer days of waiting.
What I suggest: that the head of department review the 12 routing criteria once a quarter — they are written down, short, and they belong to you. None of them changes unless you write the change. routing_bodily-injury-dispute-second-look.pdf12 written criteria, no label in the file
✎ Framework · departmental routing rules, transfer log, the lawyer's caseload
All this means you read notifications, contracts, sometimes very personal documents. Where exactly does that go?
It does not leave your premises. I run on local inference on a machine at the mutual, and your files are copied nowhere.
Local inference means the model computes on your machine: the text of a notification crosses no outside network to be processed. If you would rather not host a machine, the other route is an isolated resource hosted in France, under French law, dedicated to your mutual — no pooling with another insurer.
What protects the data of your 78,000 policyholders:
· No notification trains a model, neither yours nor anyone else's. What I learn from your files serves your files.
· Encryption in transit and at rest, and role-based access — rights follow the job: a first-level handler opens the files in their own portfolio, not bodily injury files and not files in dispute.
· A full log: which file was built, from which documents, at what time, and who approved what. That log is what answers a complaint, an audit, or a policyholder asking what was done with their data.
· Hosting in France, under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity — including against a US provider hosting in Europe.
· Health documents stay outside the first-level scope, and for bodily injury claims you have the SecNumCloud and reinforced hosting options, to whatever level you require.
The figure that makes this commercial rather than technical: 0 data outside the European Union, and 100 % of hosting in France. Your last claims-delegation tender made it a knock-out criterion; you were able to answer without reservation, evidence attached.
What I suggest: that I keep up to date the record your data protection officer asks for every year — hosting, subcontractors, retention periods, who accesses what. It is asked for once a year and takes three days to assemble. technical-framework_where-notifications-live.pdfLocal inference, role-based access, processing in the EU targeted
✎ Framework · deployment architecture, access log, record of processing activities
Local inference means the model computes on your machine: the text of a notification crosses no outside network to be processed. If you would rather not host a machine, the other route is an isolated resource hosted in France, under French law, dedicated to your mutual — no pooling with another insurer.
What protects the data of your 78,000 policyholders:
· No notification trains a model, neither yours nor anyone else's. What I learn from your files serves your files.
· Encryption in transit and at rest, and role-based access — rights follow the job: a first-level handler opens the files in their own portfolio, not bodily injury files and not files in dispute.
· A full log: which file was built, from which documents, at what time, and who approved what. That log is what answers a complaint, an audit, or a policyholder asking what was done with their data.
· Hosting in France, under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity — including against a US provider hosting in Europe.
· Health documents stay outside the first-level scope, and for bodily injury claims you have the SecNumCloud and reinforced hosting options, to whatever level you require.
The figure that makes this commercial rather than technical: 0 data outside the European Union, and 100 % of hosting in France. Your last claims-delegation tender made it a knock-out criterion; you were able to answer without reservation, evidence attached.
What I suggest: that I keep up to date the record your data protection officer asks for every year — hosting, subcontractors, retention periods, who accesses what. It is asked for once a year and takes three days to assemble. technical-framework_where-notifications-live.pdfLocal inference, role-based access, processing in the EU targeted
✎ Framework · deployment architecture, access log, record of processing activities
And on the policyholder's side? Do they know they are dealing with a machine?
Yes, from the first sentence, every time, and they can ask for a person at any moment. It is not an option you could switch off.
What is said, and when:
· On the phone: the statement opens the call, before any question. « You can ask for an adviser at any time » is in the same sentence, not three screens later.
· On the portal: the statement heads the exchange, in the policyholder's language.
· On the letters and emails I prepare: they go out in the name of a named handler, because he has reviewed them or mandated them. A letter signed by a machine does not exist here.
The basis, and it is dated: since 2 August 2026, article 50(1) of the European AI Regulation requires anyone interacting with an AI system to be informed, unless that is obvious.
What it produces in practice, and the figure is a surprise: 11 % of policyholders ask for an adviser after the statement — they get one in 2 minutes on average. The other 89 % complete their notification, often at 9 p.m. on a Sunday, at the moment they have just found the damage.
And the measure that locks everything down: over the quarter, 0 claims accepted and 0 settlements set without human approval, across 2,840 files built. That is not a cautious setting, it is how I am wired: I have no access to the screen that accepts a claim.
What I suggest: adding to the acknowledgement letter the sentence that tells the policyholder what they can ask for — their notification as recorded, and the list of documents they have already supplied. Two lines, and a third of the anxious calls disappear. informing-the-policyholder_what-is-said-and-when.pdf11 % ask for an adviser, reached in 2 minutes
✎ Framework · call scripts, adviser request log, approval log for the quarter
What is said, and when:
· On the phone: the statement opens the call, before any question. « You can ask for an adviser at any time » is in the same sentence, not three screens later.
· On the portal: the statement heads the exchange, in the policyholder's language.
· On the letters and emails I prepare: they go out in the name of a named handler, because he has reviewed them or mandated them. A letter signed by a machine does not exist here.
The basis, and it is dated: since 2 August 2026, article 50(1) of the European AI Regulation requires anyone interacting with an AI system to be informed, unless that is obvious.
What it produces in practice, and the figure is a surprise: 11 % of policyholders ask for an adviser after the statement — they get one in 2 minutes on average. The other 89 % complete their notification, often at 9 p.m. on a Sunday, at the moment they have just found the damage.
And the measure that locks everything down: over the quarter, 0 claims accepted and 0 settlements set without human approval, across 2,840 files built. That is not a cautious setting, it is how I am wired: I have no access to the screen that accepts a claim.
What I suggest: adding to the acknowledgement letter the sentence that tells the policyholder what they can ask for — their notification as recorded, and the list of documents they have already supplied. Two lines, and a third of the anxious calls disappear. informing-the-policyholder_what-is-said-and-when.pdf11 % ask for an adviser, reached in 2 minutes
✎ Framework · call scripts, adviser request log, approval log for the quarter
Do we have to change our claims handling system?
No. I plug into it read-only, and nothing moves on your side — not your screens, not your habits, not your file references.
What I plug into, and what I do there:
· The claims handling system: I read the contracts and the files, and I attach the built file to them. The accounting entries stay yours.
· The document management system: I index the content of the documents, without renaming or moving a single file.
· The notification portal: I receive, I do not change the form.
· The switchboard: the voice agent answers and hands over.
What that avoids: no migration, no extra subscription for your 14 handlers, no historical data conversion. Your 62 standard contracts stay exactly where they are.
The timescale: a few weeks, after the audit and the design phase where we write down together your notification protocol, your 19 lists of documents and your 12 routing criteria. They are your rules; I apply them, I do not invent them.
What I suggest to start: escape of water alone — 4,200 notifications, the largest type and the best defined. Three months of pilot, then you look at the three figures that matter: time to a complete file, rework rate and chasers avoided. If any one of the three does not move, we stop there — and you keep the 19 document lists and the 12 criteria, which are yours.
✎ Framework · integration plan, pilot scope, the department's notification protocol
What I plug into, and what I do there:
· The claims handling system: I read the contracts and the files, and I attach the built file to them. The accounting entries stay yours.
· The document management system: I index the content of the documents, without renaming or moving a single file.
· The notification portal: I receive, I do not change the form.
· The switchboard: the voice agent answers and hands over.
What that avoids: no migration, no extra subscription for your 14 handlers, no historical data conversion. Your 62 standard contracts stay exactly where they are.
The timescale: a few weeks, after the audit and the design phase where we write down together your notification protocol, your 19 lists of documents and your 12 routing criteria. They are your rules; I apply them, I do not invent them.
What I suggest to start: escape of water alone — 4,200 notifications, the largest type and the best defined. Three months of pilot, then you look at the three figures that matter: time to a complete file, rework rate and chasers avoided. If any one of the three does not move, we stop there — and you keep the 19 document lists and the 12 criteria, which are yours.
✎ Framework · integration plan, pilot scope, the department's notification protocol
After a quarter, what has actually changed for the department?
A file becomes complete in 2.1 days instead of 9.4. Everything else follows from that one figure — and it shows up as far as your lapse rate.
On the three tasks you were measuring:
· Taking and writing up the notification: 55 % → 9 % of first-level handling time.
· Checking the documents in the file: 35 % → 7 %.
· Matching the contract and the cover: 40 % → 8 %.
What those hours became: the backlog of files older than two months fell from 340 to 118. Your handlers did not work less, they worked on what was dragging — and those are the files that generate complaints.
The figure that shows up in your accounts, and it is the only one here that does: your lapse rate in the six months after a claim falls from 9.1 % to 6.4 % across the scope covered. Across 2,840 claimants in the quarter, that is 76 policies kept; scaled to a year at your average premium of €412, roughly 307 policies and €126,000 of premium that stay with you. A policyholder does not leave because they were poorly settled — they leave because they waited without knowing.
And the framework measure your board will ask for: 0 claims accepted without human approval, 0 settlements set automatically, 0 data leaving the European Union, across 2,840 files and 4,100 logged approvals.
What I suggest for the next quarter: extend to the 1,900 commercial multi-risk notifications. That is where your delay is longest — 14.2 days to a complete file — and where your artisans most need a fast answer: a workshop that is shut down does not catch up. quarterly-review_what-the-department-regained.pdf9.4 → 2.1 days, 76 policies kept
⛓ Sourced · claims system timings, departmental backlog, approval log, policyholder portfolio
On the three tasks you were measuring:
· Taking and writing up the notification: 55 % → 9 % of first-level handling time.
· Checking the documents in the file: 35 % → 7 %.
· Matching the contract and the cover: 40 % → 8 %.
What those hours became: the backlog of files older than two months fell from 340 to 118. Your handlers did not work less, they worked on what was dragging — and those are the files that generate complaints.
The figure that shows up in your accounts, and it is the only one here that does: your lapse rate in the six months after a claim falls from 9.1 % to 6.4 % across the scope covered. Across 2,840 claimants in the quarter, that is 76 policies kept; scaled to a year at your average premium of €412, roughly 307 policies and €126,000 of premium that stay with you. A policyholder does not leave because they were poorly settled — they leave because they waited without knowing.
And the framework measure your board will ask for: 0 claims accepted without human approval, 0 settlements set automatically, 0 data leaving the European Union, across 2,840 files and 4,100 logged approvals.
What I suggest for the next quarter: extend to the 1,900 commercial multi-risk notifications. That is where your delay is longest — 14.2 days to a complete file — and where your artisans most need a fast answer: a workshop that is shut down does not catch up. quarterly-review_what-the-department-regained.pdf9.4 → 2.1 days, 76 policies kept
⛓ Sourced · claims system timings, departmental backlog, approval log, policyholder portfolio
Something must have gone wrong. Say so, we would rather know now.
Yes, and the figure is in the review: of 2,840 files built, 176 had to be reworked by a handler because my matching of the cover was incomplete. Six point two per cent.
The cause, and it is precise — not an average: 131 of those 176 reworks fall on 4 standard contracts alone, all predating 2019. Those four exist only as scanned documents, with handwritten endorsements stapled to them: I was reading the standard contract, I was not reading the endorsement that had amended it.
The comparison that explains it: across the 58 standard contracts available in structured form, the rework rate is 1.4 %. It is not the difficulty of the loss that separates the two groups, it is the state of the contract — and that is not an intelligence problem, it is a source problem.
What I have already done about it: the 4 contracts and their 61 endorsements went through the document processing agent, then were read back by your lawyer — four hours of their time, once. On the 2 contracts already handled, the rework rate went from 6.2 % to 1.9 % in five weeks.
What I am asking for: four more hours for the 2 remaining contracts, and the rework rate for the whole department joins 1.4 %. It is the same correction, applied twice more.
Why I publish this figure rather than let it pass: because a rework rate reads both ways. 176 files reworked is 176 times your handler saw what I had not seen — which is exactly what you are buying: a file that arrives done, and a person who decides. reworked-by-the-handler_176-files.pdf131 on 4 old contracts, correction measured
⛓ Sourced · rework log, state of the standard contracts, measurements before and after the document rebuild
The cause, and it is precise — not an average: 131 of those 176 reworks fall on 4 standard contracts alone, all predating 2019. Those four exist only as scanned documents, with handwritten endorsements stapled to them: I was reading the standard contract, I was not reading the endorsement that had amended it.
The comparison that explains it: across the 58 standard contracts available in structured form, the rework rate is 1.4 %. It is not the difficulty of the loss that separates the two groups, it is the state of the contract — and that is not an intelligence problem, it is a source problem.
What I have already done about it: the 4 contracts and their 61 endorsements went through the document processing agent, then were read back by your lawyer — four hours of their time, once. On the 2 contracts already handled, the rework rate went from 6.2 % to 1.9 % in five weeks.
What I am asking for: four more hours for the 2 remaining contracts, and the rework rate for the whole department joins 1.4 %. It is the same correction, applied twice more.
Why I publish this figure rather than let it pass: because a rework rate reads both ways. 176 files reworked is 176 times your handler saw what I had not seen — which is exactly what you are buying: a file that arrives done, and a person who decides. reworked-by-the-handler_176-files.pdf131 on 4 old contracts, correction measured
⛓ Sourced · rework log, state of the standard contracts, measurements before and after the document rebuild
Last question: day to day, what goes out without asking us?
Three actions, not one more, all named and all reversible. Everything else waits for a signature, including where it would save me time.
The three automatic actions:
· The acknowledgement to the policyholder, within 10 minutes, with the file number and what they need to supply. Reversible: if they notify the same loss twice, I do not send two acknowledgements — I link them and say so.
· Stopping a chaser the moment a document arrives, including at night and at weekends. And the reverse: if the document is unreadable, the chaser restarts on that document alone, never on the whole list.
· Filing an incoming document in the file it concerns, with its arrival date. Reversible in one click if the link is wrong, and the history keeps a trace of both.
And the six acts that carry a signature: accepting a claim · setting a settlement · ruling on cover · writing to the policyholder about anything other than a document request · instructing an adjuster beyond the signed mandate · handling a bodily injury file, a file in dispute or a file raised for a second look. Six acts, six people's names — and the file built in advance every time.
What that gives in one line: 2,840 files built, 4,100 approvals logged, 0 decisions taken without a person.
And the next step I propose, with figures: the pilot on commercial multi-risk, three months, with the same capped mandates. If time to a complete file falls there as it did on escape of water — from 14.2 days to under 4 — your 1,900 artisans get back to work a week earlier. You set the start date; everything else is ready. who-decides-what_three-automatic-actions.pdf3 reversible actions, everything else signed
✎ Framework · list of automatic actions, approval log, scope of the next pilot
The three automatic actions:
· The acknowledgement to the policyholder, within 10 minutes, with the file number and what they need to supply. Reversible: if they notify the same loss twice, I do not send two acknowledgements — I link them and say so.
· Stopping a chaser the moment a document arrives, including at night and at weekends. And the reverse: if the document is unreadable, the chaser restarts on that document alone, never on the whole list.
· Filing an incoming document in the file it concerns, with its arrival date. Reversible in one click if the link is wrong, and the history keeps a trace of both.
And the six acts that carry a signature: accepting a claim · setting a settlement · ruling on cover · writing to the policyholder about anything other than a document request · instructing an adjuster beyond the signed mandate · handling a bodily injury file, a file in dispute or a file raised for a second look. Six acts, six people's names — and the file built in advance every time.
What that gives in one line: 2,840 files built, 4,100 approvals logged, 0 decisions taken without a person.
And the next step I propose, with figures: the pilot on commercial multi-risk, three months, with the same capped mandates. If time to a complete file falls there as it did on escape of water — from 14.2 days to under 4 — your 1,900 artisans get back to work a week earlier. You set the start date; everything else is ready. who-decides-what_three-automatic-actions.pdf3 reversible actions, everything else signed
✎ Framework · list of automatic actions, approval log, scope of the next pilot
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, the first level of claims handling. 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 765 € excl. VAT / month
Taking the notification
Gathers date, circumstances and property concerned according to your protocol.
Documents required
Checks the documents in the file and prepares the request for those missing. The request for a document is announced by text message, without ever detailing the claim.
Matching the cover
Sets out the applicable cover with its reference in the contract.
File the claim notification and its documents in the claim file
To extract the data from the documents received, a dedicated document agent takes it on.
Controls and safeguards
These 6 controls are built into the agent: they frame what it does, whatever plan you pick. They are not chosen and are not added to your order.
Human validation, exceptions and escalation
Status, safe closure and audit trail
Sources, access rights and handling of questions with no answer
Implement testing, explainability, human oversight and monitoring
Record models, decisions, incidents and changes
Protect consumers, data and regulated secrets
Other needs our agents cover
Each card says where the matching agent stands: available, on quote, or still being architected.
Banking, finance & insurance
For everything at stake in the sector, see our dedicated page.
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 files can a handler examine on the merits?
By taking on the gathering of the material, the effort shifts towards examining the claim and the relationship with the policyholder. How large the gain is depends on your volume and remains to be confirmed by a pilot.
Taking and writing up the notification
Checking the documents in the file
Matching the contract and the cover
Indicative figures, not contractual, to be confirmed by a pilot on your volume of notifications and the variety of your cover. Accepting a claim and setting a settlement concern money and the policyholder's rights: those decisions belong to the handler.
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 first-level claims handling agent (notification, documents, cover), installed and operated for you.
Agility
Setup + controlled subscription
8,260 € excl. VAT setup
then 765 € 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
1,225 € 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
12,245 € excl. VAT setup
then 986 € 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 policyholders
Your policyholders' data stays with youLocal inference or an isolated resource hosted in France; no notification and no policyholder data entrusted to a third party, no data used to train a model.
Data in France, under French lawPolicyholders' personal data: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The claims handler keeps the decisionThe agent produces claim files that are built, which can be checked and altered; no approval is automated.
Human oversight & traceabilityOn your volume of notifications and the variety of your cover: systematic logging and tracking, in line with the AI Act.
Going further
Related resources
Frequently asked questions
Your questions, our answers
Does the agent decide whether a claim is accepted?
No. It builds the file and sets out the applicable cover with its reference in the contract. Accepting the claim and setting the settlement belong to the handler.
How are the stated circumstances handled?
They are carried over exactly as the policyholder stated them, without rewording. Rewritten circumstances could change how the file reads.
What does it do if a file is doubtful?
It puts the file to the handler with the material gathered and what remains uncertain. Personal injury claims, disputes and suspected fraud are excluded from automated handling from the design stage.
Is our policyholders' data 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. That data is not used to train a third-party model.
Does it connect to our handling system?
Yes, to the common tools on the market, reading the contracts and the files. The integration is defined at the design stage.
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 volume of notifications and the variety of your cover, 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 policyholder side, the agent answers on WhatsApp Business, your website chat and email — people reach you where they already write, with no account to create. A claim is often reported from the scene, phone in hand: gathering the report and the documents where the policyholder actually is saves real time. The circumstances stated are recorded as given, and the decision to cover the claim stays with the handler. 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 policyholders by text message?
Yes, as an option. The agent acknowledges the claim by text message, flags the missing document and announces the loss adjuster's visit. 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 or email. 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 insurance
Let's talk
Let's size up the potential in your claims
15 minutes to frame your protocol and your cover — hosted in France, supervised, with no commitment.