We build Agents. You build the Future.
● B2B offer — Insurance & settlementClaims handling: the file worked, the decision prepared
Working a claim follows a protocol: notification taken, documents checked, a loss adjustment where needed, cover matched, costing prepared. Your agent runs that protocol from end to end, stops at every approval point you have placed and keeps the record of each step. Data hosted in France; the AI model is called at its publisher, except with the Sovereign model option. With the 100% Sovereign package, your policyholders' data stays with you. The settlement decision belongs to the department.
Updated on
Deployed in a few weeks
Claims handling · hosted in France
Fictional simulation — not a client result
Where does file 7314 stand?
Steps completed: notification taken, documents checked, loss adjustment received, cover matched.
The costing prepared takes the items from the adjuster's report and the limits in the contract, with their reference.
The next step is an approval point: the settlement decision.
🔗 Sourced · the handling protocol and the documents in the file
The costing prepared takes the items from the adjuster's report and the limits in the contract, with their reference.
The next step is an approval point: the settlement decision.
🔗 Sourced · the handling protocol and the documents in the file
Why not go all the way to payment?
The costing is ready and documented item by item, with the applicable limits.
Setting a settlement concerns money and the policyholder's rights: it is an approval point you placed there, and it stays one.
✎ Support · costing prepared, the department's decision
Setting a settlement concerns money and the policyholder's rights: it is an approval point you placed there, and it stays one.
✎ Support · costing prepared, the department's decision
Data hosted in France · sovereign model as an option
Files hosted in France
Hosted in FranceAI model called at its publisher, except with the Sovereign model option
Policyholder data protected
GDPR & AI Act: governed deploymentTraceability & human oversight
TurnkeyDesigned, installed and operated for you
The settlement department decidesThe agent carries out what you have authorised
For an organisation established outside the European Union, the law of its own country applies; at its request, we put equivalent safeguards in place (human review before any action, a log of the criteria, information for the people concerned).
✦ In brief
A Blue Lemon Agent claims handling agent runs your handling protocol from notification to costing: documents checked, loss adjustment built in, cover matched and items costed with their reference in the contract. It stops at every approval point and keeps the record of each step. Data hosted in France; the AI model is called at its publisher, except with the Sovereign model option. With the 100% Sovereign package, your policyholders' data stays 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 with the Sovereign model option (target architecture)
4
settlement uses ready to deploy
0
decision with a legal or significant effect taken by an agent
These figures describe our offer, not results measured at a client. How large the gain is on your volume of files and the complexity of your protocols is confirmed by a pilot.
Indicative estimate, to be confirmed by a pilot on your real files (general terms of sale, article 4).
Pricing
One package, one agent
A claims handling agent (protocol, loss adjustment, costing), installed and operated for you.
Agility
Setup + controlled subscription
4,310 € excl. VAT setup
then 396 € 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
- Hosting in France, a dedicated and isolated resource
The simplest
Serenity
All inclusive, no setup fee
636 € 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
- Hosting in France, managed end to end
100% Sovereign
On site, you own it
6,505 € excl. VAT setup
then 507 € 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.
- Hardware installed on your premises (you own it)
- French / European AI models run locally
- Secure remote maintenance (Pro support included)
Included in the package: hosting and AI consumption up to the volume of your capacity tier, in documents, exchanges, tokens and storage; beyond that, re-invoiced at real cost with no margin, tracked in your client area as usage statements are issued. 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 for the package chosen, in France by default, 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.
Use cases
What does the agent actually do?
One agent, the whole handling chain. The agent carries out by itself what you authorise it to do; you decide what it does on its own and what it submits to you. The compensation decision belongs to the department.
Included in your agent
The 4 capabilities essential to this promise are included, at no extra cost.
From 396 € excl. VAT / month
Running the protocol
Chains together the steps of your handling protocol, file by file.
Building in the loss adjustment
Takes the items from the adjuster's report into the costing prepared.
Approval points
Stops at every step you have defined as committing.
Build the claim file from the documents received
To extract the data from the documents received, a dedicated document agent completes the picture.
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.
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 how the matching agent can be obtained today: available, on quote, or not yet in the catalogue.
Banking, finance & insurance
For everything at stake in the sector, see our dedicated page.
Available View the agent page →Need to go further?
These agents handle a different business process, with their own owner and their own price. They are added to this one.
Does your need fall outside this?
Build your agent
In 30 minutes we identify the most relevant agent — without oversizing the project.
Demonstration
See the agent at work
5 everyday 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 companySévrelle Assurances — regional property and casualty insurer
- Sector
- Property and casualty insurance — home, motor and commercial multi-risk; claims handled in house
- Headcount
- 218 staff, including 16 in the claims department: 10 handlers, 4 costing writers and 2 technical referents; 26 appointed loss adjusters outside
- Policyholders served
- 128,000 policies, private and business, across four départements; average annual premium of €420
- Order of magnitude
- 7,200 claims notified a year, 600 a month, of which 2,400 involve a loss adjuster; 34 days of average settlement time; commitment to a first reply to the policyholder within 5 working days
- Tools in place
- Claims management system, document upload platform, policy and cover repository, adjuster network and six years of closed files — the agent plugs into them read-only, nothing is replaced or migrated
- Who decides
- The handler validates each point of the file; the claims manager sets the settlement amount; the technical referent rules on disputed cover; bodily injury claims, disputes and suspected fraud are handled entirely by a person
- Room for improvement
- 65 % of the time spent on a file goes into chaining the steps; 27 of the 34 days of delay are waiting between two steps; a file waits 11 days between the adjuster's report and its costing; 38 % of files go back out for a missing document
Sévrelle Assurances is not looking to settle differently, but to settle faster and to spend its handlers' time on what only they do: assessing and talking to the policyholder. The agent runs on local inference on a machine of the company and plugs read-only into the claims system, the document platform and the cover repository: it runs the protocol, stops at every validation point placed by the department and keeps the trace of every step. The exchanges below cover one quarter, from the reading of six years of closed files to the review presented to management.
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
Fictional simulation — not a client result
I read six years of your closed files — 41,200 claims — and timed each of the eleven steps of your handling protocol. The result fits in one sentence: of your 34 days of average settlement time, 27 are waiting between two steps.
The handling protocol is the sequence of steps your department applies to a claim: notification taken, documents checked, adjuster if needed, cover matched, costing prepared, decision. Your handlers are not slow: they wait, and they chase.
The five waits that make 80 % of the delay, measured on your own files:
· From notification to a complete file: 8.4 days, because 38 % of files go back out for a missing document.
· From a complete file to appointing the adjuster: 3.2 days.
· From the adjuster's report to its costing: 11 days — the longest of all, and the one the policyholder feels most, since they know the adjuster has been round.
· From costing to decision: 2.6 days.
· From decision to payment: 2.1 days.
What I take on, and what it shifts: chaining the steps accounts for 65 % of the handling time of a file — 55 minutes out of the 84 it takes. It now accounts for 9 %: 7 minutes 30 seconds. The rest is assessment and the call to the policyholder: the work only your handlers can do.
What it does to the delay: 34 days of average settlement brought down to 12, and the commitment to a first reply within 5 working days met 98 times out of 100 instead of 71.
What I propose: that the claims manager reads the map of the eleven steps and places the stopping points himself — I run the protocol between two points and stop at every one. The map is written, it can be corrected with a single word, and I apply it from the next notification. protocol-map_where-the-34-days-go.pdf11 steps timed, 27 waiting days out of 34
⛓ Sourced · 6 years of closed files (41,200), timestamps of the 11 steps, log of document requests
The handling protocol is the sequence of steps your department applies to a claim: notification taken, documents checked, adjuster if needed, cover matched, costing prepared, decision. Your handlers are not slow: they wait, and they chase.
The five waits that make 80 % of the delay, measured on your own files:
· From notification to a complete file: 8.4 days, because 38 % of files go back out for a missing document.
· From a complete file to appointing the adjuster: 3.2 days.
· From the adjuster's report to its costing: 11 days — the longest of all, and the one the policyholder feels most, since they know the adjuster has been round.
· From costing to decision: 2.6 days.
· From decision to payment: 2.1 days.
What I take on, and what it shifts: chaining the steps accounts for 65 % of the handling time of a file — 55 minutes out of the 84 it takes. It now accounts for 9 %: 7 minutes 30 seconds. The rest is assessment and the call to the policyholder: the work only your handlers can do.
What it does to the delay: 34 days of average settlement brought down to 12, and the commitment to a first reply within 5 working days met 98 times out of 100 instead of 71.
What I propose: that the claims manager reads the map of the eleven steps and places the stopping points himself — I run the protocol between two points and stop at every one. The map is written, it can be corrected with a single word, and I apply it from the next notification. protocol-map_where-the-34-days-go.pdf11 steps timed, 27 waiting days out of 34
⛓ Sourced · 6 years of closed files (41,200), timestamps of the 11 steps, log of document requests
Where has file 7314 got to? It is a water damage claim notified twelve days ago, and the policyholder has called twice.
File 7314 is at the validation point that precedes the decision, and it arrived there yesterday. Here is the exact state, step by step.
· Notification taken on the 3rd at 9:12 am, through the online form. Water damage, kitchen and living room, source identified: washing machine connector.
· Documents checked on the 3rd at 9:26 am — fourteen minutes after notification. The furniture invoice was missing: the request went out the same day, the document came back on the 5th.
· Adjuster appointed on the 5th, report received on the 9th.
· Cover matched on the 9th: water damage covered under the policy, €150 excess, €12,000 furniture limit, depreciation deducted per the policy scale and recoverable on proof of replacement. Every line refers to its clause.
· Costing prepared on the 9th: 4 items taken from the adjuster's report, €6,320 before excess, €6,170 after.
· Next step: the validation point you placed there — the settlement decision.
What I did without being asked, and it explains the twelve days: I chased the adjuster on the morning of the 8th, his report was due on the 7th. And I flagged a discrepancy: the adjuster costs a « ceiling paint » item that the leak-detection report does not mention. It is set apart in the costing, with both documents side by side — €340 that deserve your eye, not mine.
What I propose now: the settlement proposal is drafted and so is the letter to the policyholder. You rule on the paint item, you approve, and both go out within the minute — the policyholder who has called twice will have her answer today, twelve days after notification instead of thirty-four. file-7314_status-and-prepared-costing.pdf6 steps completed, 1 discrepancy set apart, decision with the department
⛓ Sourced · file 7314, adjuster's report of the 9th, policy and endorsement, chasing log
· Notification taken on the 3rd at 9:12 am, through the online form. Water damage, kitchen and living room, source identified: washing machine connector.
· Documents checked on the 3rd at 9:26 am — fourteen minutes after notification. The furniture invoice was missing: the request went out the same day, the document came back on the 5th.
· Adjuster appointed on the 5th, report received on the 9th.
· Cover matched on the 9th: water damage covered under the policy, €150 excess, €12,000 furniture limit, depreciation deducted per the policy scale and recoverable on proof of replacement. Every line refers to its clause.
· Costing prepared on the 9th: 4 items taken from the adjuster's report, €6,320 before excess, €6,170 after.
· Next step: the validation point you placed there — the settlement decision.
What I did without being asked, and it explains the twelve days: I chased the adjuster on the morning of the 8th, his report was due on the 7th. And I flagged a discrepancy: the adjuster costs a « ceiling paint » item that the leak-detection report does not mention. It is set apart in the costing, with both documents side by side — €340 that deserve your eye, not mine.
What I propose now: the settlement proposal is drafted and so is the letter to the policyholder. You rule on the paint item, you approve, and both go out within the minute — the policyholder who has called twice will have her answer today, twelve days after notification instead of thirty-four. file-7314_status-and-prepared-costing.pdf6 steps completed, 1 discrepancy set apart, decision with the department
⛓ Sourced · file 7314, adjuster's report of the 9th, policy and endorsement, chasing log
All of this assumes you read notifications, documents, sometimes medical certificates. Where do they go?
With the 100% Sovereign package, they do not leave your walls. I run on local inference on a machine belonging to Sévrelle, and no policyholder file is copied anywhere else.
Local inference means the model computes on your machine: the text of a notification or of a document crosses no external network to be processed. If you would rather not host a machine, the other route is an isolated resource hosted in France, dedicated to Sévrelle — no pooling with another insurer.
What protects your policyholders' data, point by point:
· No file trains any model. What I learn from your 41,200 files serves your policyholders, and nothing you entrust to me surfaces anywhere else.
· I read, I do not write in your files: I post drafts, and your handler is the one who records them — the technical account through which I read has no write permission, and that can be checked with a single command.
· Reinforced ring-fencing of health data, with role-based access — rights follow the job: a costing writer opens material damage items, not the medical documents of a bodily injury file.
· Encryption in transit and at rest, and a complete log: which document was read, by which process, at what time. That log is what lets you reconstruct a file for an audit.
· Hosting in France, under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity, including against an American provider hosting in Europe.
And a design decision that protects you, taken before go-live: bodily injury claims, disputed files and suspected fraud leave the assisted workflow — they go to entirely human handling. That is not a caveat, it is time gained: the sorting happens at notification, in 4 minutes. Last year your 800 bodily injury files were spotted on the ninth day on average, after following the ordinary circuit; they now go to the referent the same day, file assembled.
What I propose: that I keep up to date the processing sheet your data protection officer and your internal audit will ask for — data processed, retention periods, who has access to what. The first version is written and you have it to hand. technical-framework_where-your-policyholder-files-live.pdf100% Sovereign package: local inference, health ring-fencing, processing in the EU targeted
✎ Framework · deployment architecture, technical account permissions, health data ring-fencing, access log
Local inference means the model computes on your machine: the text of a notification or of a document crosses no external network to be processed. If you would rather not host a machine, the other route is an isolated resource hosted in France, dedicated to Sévrelle — no pooling with another insurer.
What protects your policyholders' data, point by point:
· No file trains any model. What I learn from your 41,200 files serves your policyholders, and nothing you entrust to me surfaces anywhere else.
· I read, I do not write in your files: I post drafts, and your handler is the one who records them — the technical account through which I read has no write permission, and that can be checked with a single command.
· Reinforced ring-fencing of health data, with role-based access — rights follow the job: a costing writer opens material damage items, not the medical documents of a bodily injury file.
· Encryption in transit and at rest, and a complete log: which document was read, by which process, at what time. That log is what lets you reconstruct a file for an audit.
· Hosting in France, under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity, including against an American provider hosting in Europe.
And a design decision that protects you, taken before go-live: bodily injury claims, disputed files and suspected fraud leave the assisted workflow — they go to entirely human handling. That is not a caveat, it is time gained: the sorting happens at notification, in 4 minutes. Last year your 800 bodily injury files were spotted on the ninth day on average, after following the ordinary circuit; they now go to the referent the same day, file assembled.
What I propose: that I keep up to date the processing sheet your data protection officer and your internal audit will ask for — data processed, retention periods, who has access to what. The first version is written and you have it to hand. technical-framework_where-your-policyholder-files-live.pdf100% Sovereign package: local inference, health ring-fencing, processing in the EU targeted
✎ Framework · deployment architecture, technical account permissions, health data ring-fencing, access log
Fictional simulation — not a client result
The completeness check now happens at notification, in fourteen minutes, and it has brought document requests down from 38 % to 9 % of files.
Completeness is the check that every document required to handle the claim is present; an incomplete file does not move forward, and the check at first notification is what makes that known straight away, rather than on the day someone opens it.
What I check on arrival, claim by claim: the documents required for that type of claim and for that policy, the consistency of dates between notification and evidence, the presence of the policy and its endorsements in the version in force on the day of the loss — the point that gains most from being settled at first notification.
What the policyholder receives when a document is missing: a single letter, listing ALL the missing documents, with where to find each one and what it is for. Your files went back out 1.7 times on average for a missing document: they now go back 1.05 times — because the request was partial, not because the policyholder was careless.
What that gives, quantified: the wait between notification and a complete file drops from 8.4 days to 2.3. Across 5,100 standard files a year, that is six days given back to every policyholder, and as many chasing calls that no longer reach your switchboard.
And cover matching follows immediately: for every complete file, I set out the applicable cover, the policy clause, the excess, the limit and the depreciation scale — every line refers to its clause and to the version of the policy in force on the day of the loss.
The next step I propose: that the six document request letters be reviewed once and for all — I have rewritten them, each with the complete list and a plain-language explanation. You approve them in twenty minutes and the policyholder stops receiving three letters for one file. completeness-check_38-percent-down-to-9.pdfDocuments required by claim type, 6 letters rewritten
⛓ Sourced · 41,200 closed files, log of document requests, policy and endorsement repository
Completeness is the check that every document required to handle the claim is present; an incomplete file does not move forward, and the check at first notification is what makes that known straight away, rather than on the day someone opens it.
What I check on arrival, claim by claim: the documents required for that type of claim and for that policy, the consistency of dates between notification and evidence, the presence of the policy and its endorsements in the version in force on the day of the loss — the point that gains most from being settled at first notification.
What the policyholder receives when a document is missing: a single letter, listing ALL the missing documents, with where to find each one and what it is for. Your files went back out 1.7 times on average for a missing document: they now go back 1.05 times — because the request was partial, not because the policyholder was careless.
What that gives, quantified: the wait between notification and a complete file drops from 8.4 days to 2.3. Across 5,100 standard files a year, that is six days given back to every policyholder, and as many chasing calls that no longer reach your switchboard.
And cover matching follows immediately: for every complete file, I set out the applicable cover, the policy clause, the excess, the limit and the depreciation scale — every line refers to its clause and to the version of the policy in force on the day of the loss.
The next step I propose: that the six document request letters be reviewed once and for all — I have rewritten them, each with the complete list and a plain-language explanation. You approve them in twenty minutes and the policyholder stops receiving three letters for one file. completeness-check_38-percent-down-to-9.pdfDocuments required by claim type, 6 letters rewritten
⛓ Sourced · 41,200 closed files, log of document requests, policy and endorsement repository
And when the cover is arguable? Our two technical referents spend their days on that.
Of the 5,100 standard files, 4,480 leave no room for argument — the cover is named, the clause is cited, your handler checks in two minutes. The other 620 reach your referent ALREADY argued from both sides.
What I hand over for each of those 620:
· The two readings, written out: the one that triggers the cover and the one that excludes it, each with the exact clause it rests on.
· The internal precedents: the twelve comparable files of the last six years and what the referent decided in each, with the date. That is your own doctrine, as your referents applied it: it lived in their decisions, and it has been written down since this morning.
· The amount at stake, so that the ruling is made in full knowledge of what it involves.
What that changes for your referents: a ruling used to take 40 minutes, it takes 6 — 620 rulings, 351 hours a year given back to the two people who hold the technical doctrine of the house.
And the figure that opens the real saving: 9 clauses concentrate 71 % of those 620 arguments. The same questions come back: the line between seepage and water damage, the definition of business furniture, depreciation on goods less than two years old…
What I do on top, and it is the work that never gets done between two files: I have drafted the 9 corresponding doctrine notes, each grounded in the decisions your referents actually made — for each one, the number of files it would have settled without a ruling over the last six years, and the files it would have left to a ruling. Your referents choose on figures, not on a hunch.
The next step: your two referents read the 9 notes — half a day. Signed, they bring rulings down from 620 to 180 a year, and they remain your doctrine, enforceable and editable with a single word. disputed-cover_9-doctrine-notes.pdf620 rulings, 40 minutes down to 6, 9 clauses making 71 %
⛓ Sourced · 620 files with disputed cover, referents' decisions over 6 years, clauses of the cover repository
What I hand over for each of those 620:
· The two readings, written out: the one that triggers the cover and the one that excludes it, each with the exact clause it rests on.
· The internal precedents: the twelve comparable files of the last six years and what the referent decided in each, with the date. That is your own doctrine, as your referents applied it: it lived in their decisions, and it has been written down since this morning.
· The amount at stake, so that the ruling is made in full knowledge of what it involves.
What that changes for your referents: a ruling used to take 40 minutes, it takes 6 — 620 rulings, 351 hours a year given back to the two people who hold the technical doctrine of the house.
And the figure that opens the real saving: 9 clauses concentrate 71 % of those 620 arguments. The same questions come back: the line between seepage and water damage, the definition of business furniture, depreciation on goods less than two years old…
What I do on top, and it is the work that never gets done between two files: I have drafted the 9 corresponding doctrine notes, each grounded in the decisions your referents actually made — for each one, the number of files it would have settled without a ruling over the last six years, and the files it would have left to a ruling. Your referents choose on figures, not on a hunch.
The next step: your two referents read the 9 notes — half a day. Signed, they bring rulings down from 620 to 180 a year, and they remain your doctrine, enforceable and editable with a single word. disputed-cover_9-doctrine-notes.pdf620 rulings, 40 minutes down to 6, 9 clauses making 71 %
⛓ Sourced · 620 files with disputed cover, referents' decisions over 6 years, clauses of the cover repository
Our internal audit will ask us to reconstruct a file. Today that takes us two hours per file.
Three minutes, without opening a folder: the trail is built during handling, not afterwards.
What the trail carries, for every file:
· The steps completed, timestamped, each with its author — the agent when it is the agent, the person when it is a person.
· The documents received, their arrival date and the request that called for them.
· The costed items, each with its cover clause, the version of the policy in force on the day of the loss, the excess and the limit applied.
· The validation points, with who approved, when, and what changed between the draft and the decision.
What it used to cost and what it costs: building the trail accounted for 30 % of the handling time of a file — 25 minutes out of 84. It accounts for 4 %: 3 minutes 20 seconds. Across 5,100 files, that is 1,842 hours a year no longer spent retyping what has just been done.
What it is worth on audit day: across the 2,550 files handled this quarter, 100 % of costed items refer to a cover clause — that is the rule I hold most firmly: an amount without its reference to the contract enters no proposal.
And an effect that only shows on the day you need it: when a policyholder challenges a decision, the reconstruction goes out with the answer. Of the 34 challenges this quarter, 29 stopped at the reading of the item detail — a policyholder who sees where every euro comes from rarely challenges twice.
The next step I propose: that the trail go out automatically with every settlement proposal. It is already written in a form a policyholder can read — and that is the version I suggest trialling for one month, on files above €3,000. handling-trail_reconstruction-in-3-minutes.pdf100 % of items referred to a clause, 34 challenges, 29 closed
⛓ Sourced · handling log of the 2,550 files of the quarter, challenges received, cover repository
What the trail carries, for every file:
· The steps completed, timestamped, each with its author — the agent when it is the agent, the person when it is a person.
· The documents received, their arrival date and the request that called for them.
· The costed items, each with its cover clause, the version of the policy in force on the day of the loss, the excess and the limit applied.
· The validation points, with who approved, when, and what changed between the draft and the decision.
What it used to cost and what it costs: building the trail accounted for 30 % of the handling time of a file — 25 minutes out of 84. It accounts for 4 %: 3 minutes 20 seconds. Across 5,100 files, that is 1,842 hours a year no longer spent retyping what has just been done.
What it is worth on audit day: across the 2,550 files handled this quarter, 100 % of costed items refer to a cover clause — that is the rule I hold most firmly: an amount without its reference to the contract enters no proposal.
And an effect that only shows on the day you need it: when a policyholder challenges a decision, the reconstruction goes out with the answer. Of the 34 challenges this quarter, 29 stopped at the reading of the item detail — a policyholder who sees where every euro comes from rarely challenges twice.
The next step I propose: that the trail go out automatically with every settlement proposal. It is already written in a form a policyholder can read — and that is the version I suggest trialling for one month, on files above €3,000. handling-trail_reconstruction-in-3-minutes.pdf100 % of items referred to a clause, 34 challenges, 29 closed
⛓ Sourced · handling log of the 2,550 files of the quarter, challenges received, cover repository
Fictional simulation — not a client result
The adjuster's report on file 7314 arrived yesterday at 4:40 pm; the costing was ready at 5:05 pm. Twenty-five minutes, where that report used to wait eleven days.
The adjuster's report is the document in which the appointed loss adjuster describes the damage and costs it item by item; an item is one line of damage — furniture, paintwork, leak detection.
What I do with the report, line by line:
· I take every item costed by the adjuster and apply the cover, the excess, the limit and the depreciation scale of the policy — with the clause next to every line.
· I separate deducted depreciation from recoverable depreciation — the first is taken off the immediate payment, the second comes back to the policyholder on proof of replacement — and the policyholder reads both amounts, which spares you tomorrow's phone call.
· I flag discrepancies between the report and the rest of the file: an item the adjuster costs that the notification does not mention, an item notified that the report does not include. Across the 2,400 files with an adjuster last year there would have been 214 — now flagged as the report is read, rather than at payment time, when the policyholder calls.
The time this shifts: preparing the costing accounted for 45 % of the time of a file with an adjuster — 54 minutes. It accounts for 8 %: 9 minutes 40 seconds. Across 2,400 files, that is 1,773 hours a year.
And the delay, which is what the policyholder feels: the wait between the report received and its costing drops from 11 days to 1. It is the largest of the five waits in your protocol, and the one most visible from outside: the policyholder knows the adjuster has been round.
The next step I propose: that the costing go to the handler as soon as the report arrives, without waiting its turn in the queue. Across the 2,400 files with an adjuster, that brings average settlement time from 34 days to 12 — and you still approve at the same place, simply ten days earlier. prepared-costing_item-by-item-with-its-clause.pdf4 items, excess, limit, deducted and recoverable depreciation
⛓ Sourced · adjuster's report on file 7314, 2,400 files with an adjuster this year, policy depreciation scales
The adjuster's report is the document in which the appointed loss adjuster describes the damage and costs it item by item; an item is one line of damage — furniture, paintwork, leak detection.
What I do with the report, line by line:
· I take every item costed by the adjuster and apply the cover, the excess, the limit and the depreciation scale of the policy — with the clause next to every line.
· I separate deducted depreciation from recoverable depreciation — the first is taken off the immediate payment, the second comes back to the policyholder on proof of replacement — and the policyholder reads both amounts, which spares you tomorrow's phone call.
· I flag discrepancies between the report and the rest of the file: an item the adjuster costs that the notification does not mention, an item notified that the report does not include. Across the 2,400 files with an adjuster last year there would have been 214 — now flagged as the report is read, rather than at payment time, when the policyholder calls.
The time this shifts: preparing the costing accounted for 45 % of the time of a file with an adjuster — 54 minutes. It accounts for 8 %: 9 minutes 40 seconds. Across 2,400 files, that is 1,773 hours a year.
And the delay, which is what the policyholder feels: the wait between the report received and its costing drops from 11 days to 1. It is the largest of the five waits in your protocol, and the one most visible from outside: the policyholder knows the adjuster has been round.
The next step I propose: that the costing go to the handler as soon as the report arrives, without waiting its turn in the queue. Across the 2,400 files with an adjuster, that brings average settlement time from 34 days to 12 — and you still approve at the same place, simply ten days earlier. prepared-costing_item-by-item-with-its-clause.pdf4 items, excess, limit, deducted and recoverable depreciation
⛓ Sourced · adjuster's report on file 7314, 2,400 files with an adjuster this year, policy depreciation scales
The real gap is before that: we take three days to appoint an adjuster, and nobody knows why.
Because the mandate to do it alone has not been delegated to anyone yet. Give it to me: it is written, capped, dated, withdrawable with a single word, and you have it in front of you.
What the mandate covers, and nothing else:
· Appointing an adjuster according to your scale, the one your handlers already apply: claim type, estimated amount above the threshold you set, geographic rota among your 26 adjusters.
· Sending the six standard letters you will have approved — acknowledgement of notification, document request, notice of the adjuster's visit, delay information, request for proof of replacement, closure at the policyholder's request.
What caps it, line by line:
· Fees capped at the scale of your adjustment agreement; beyond it, the file returns to the handler with the proposed adjuster already chosen.
· Never on a bodily injury claim, a disputed file or a suspected fraud — those are not in the assisted workflow, and the mandate does not pull them back in.
· One appointment per file only. A second opinion is decided, it does not follow on automatically.
· Review after three months, figures in hand. Without an explicit decision at the review, the mandate stops: renewal is what takes a signature, not termination.
· Withdrawal: one word, effective within the minute; appointment reverts to a proposal awaiting approval, and nothing else changes.
What it is worth, on your own figures: appointment drops from 3.2 days to one hour, and the adjuster visits the policyholder three days earlier on average. Across the 2,400 files with an adjuster, that is 7,200 days of policyholder waiting removed from the clock.
And the settlement decision does not move a millimetre: it stays at the validation point where you placed it — 0 settlement set without a human decision across the 2,550 files of the quarter. Sign the mandate and the first adjuster goes out this afternoon. appointment-mandate_capped-and-dated.pdf3.2 days down to 1 hour, 6 standard letters, review at 3 months
✎ Framework · drafted mandate, the company's appointment scale, adjustment agreement, rota of the 26 adjusters
What the mandate covers, and nothing else:
· Appointing an adjuster according to your scale, the one your handlers already apply: claim type, estimated amount above the threshold you set, geographic rota among your 26 adjusters.
· Sending the six standard letters you will have approved — acknowledgement of notification, document request, notice of the adjuster's visit, delay information, request for proof of replacement, closure at the policyholder's request.
What caps it, line by line:
· Fees capped at the scale of your adjustment agreement; beyond it, the file returns to the handler with the proposed adjuster already chosen.
· Never on a bodily injury claim, a disputed file or a suspected fraud — those are not in the assisted workflow, and the mandate does not pull them back in.
· One appointment per file only. A second opinion is decided, it does not follow on automatically.
· Review after three months, figures in hand. Without an explicit decision at the review, the mandate stops: renewal is what takes a signature, not termination.
· Withdrawal: one word, effective within the minute; appointment reverts to a proposal awaiting approval, and nothing else changes.
What it is worth, on your own figures: appointment drops from 3.2 days to one hour, and the adjuster visits the policyholder three days earlier on average. Across the 2,400 files with an adjuster, that is 7,200 days of policyholder waiting removed from the clock.
And the settlement decision does not move a millimetre: it stays at the validation point where you placed it — 0 settlement set without a human decision across the 2,550 files of the quarter. Sign the mandate and the first adjuster goes out this afternoon. appointment-mandate_capped-and-dated.pdf3.2 days down to 1 hour, 6 standard letters, review at 3 months
✎ Framework · drafted mandate, the company's appointment scale, adjustment agreement, rota of the 26 adjusters
And where do we put the stopping points? I do not want a file that walks itself all the way to the cheque.
You place them, and I hold them without exception: the map of the eleven steps carries your stopping points in red, and I stop at every one even when everything is ready.
The five points your department has placed, and what I hand over at each:
· Classifying the claim — I hand over the proposed classification and the two indications that ground it; that is where bodily injury, disputes and suspicions leave the assisted workflow.
· A disputed cover match — the two readings and the twelve precedents.
· Committing an expense — adjuster, leak detection, protective measure: covered by the mandate below your threshold, handed to the handler above it.
· The settlement decision — the complete costing, item by item, with clause, excess and limit. This point is not delegated and appears in no mandate.
· Closing the file — the summary of what was paid and what remains open, notably recoverable depreciation.
What those points cost you in delay, and that is the measure that matters: 2.6 days on average between the costing being ready and the decision, out of the original 34 days. They are at 0.4 day — not because the decision is rushed, but because the file reaches the queue complete, costed and reasoned. A handler approves in 4 minutes what used to take 26.
What I suggest adding, and only you can decide it: a sixth stopping point above €8,000. Across your six years of files, amounts above €8,000 are 4 % of files and 31 % of sums paid — that is the threshold where a second look changes something. You place it with a click, I apply it to the next file. validation-points_the-five-stops-of-the-protocol.pdf5 points placed by the department, 26 minutes down to 4
✎ Framework · map of the 11 steps and their stopping points, approval log of the quarter, distribution of amounts over 6 years
The five points your department has placed, and what I hand over at each:
· Classifying the claim — I hand over the proposed classification and the two indications that ground it; that is where bodily injury, disputes and suspicions leave the assisted workflow.
· A disputed cover match — the two readings and the twelve precedents.
· Committing an expense — adjuster, leak detection, protective measure: covered by the mandate below your threshold, handed to the handler above it.
· The settlement decision — the complete costing, item by item, with clause, excess and limit. This point is not delegated and appears in no mandate.
· Closing the file — the summary of what was paid and what remains open, notably recoverable depreciation.
What those points cost you in delay, and that is the measure that matters: 2.6 days on average between the costing being ready and the decision, out of the original 34 days. They are at 0.4 day — not because the decision is rushed, but because the file reaches the queue complete, costed and reasoned. A handler approves in 4 minutes what used to take 26.
What I suggest adding, and only you can decide it: a sixth stopping point above €8,000. Across your six years of files, amounts above €8,000 are 4 % of files and 31 % of sums paid — that is the threshold where a second look changes something. You place it with a click, I apply it to the next file. validation-points_the-five-stops-of-the-protocol.pdf5 points placed by the department, 26 minutes down to 4
✎ Framework · map of the 11 steps and their stopping points, approval log of the quarter, distribution of amounts over 6 years
Fictional simulation — not a client result
The storm of the 14th produced 1,400 notifications in three days — four months of ordinary volume. Your department absorbed them with the same headcount, and the five-day commitment held 98 times out of 100.
What I did over those three days, while your handlers were calling policyholders:
· 1,400 completeness checks, and 212 document requests sent the same day, each one complete.
· The sorting by nature: 1,090 standard material damage files, 218 files needing an adjuster, 92 out of the assisted workflow — including 11 bodily injury files sent to the referent within the hour.
· Grouping by town and by street: the 340 files in one neighbourhood went to the same adjuster — 4 trips instead of 340 appointments booked one by one.
What it spared you, in euros: last year a comparable peak took 640 hours of temporary staff and €46,000. This year: no temporary hours at all, and first-response time stayed at 3.1 days.
And the measure that speaks to your sales management: the lapse rate in the twelve months following a claim went from 9.4 % to 6.1 %. Across 5,100 standard files, that is 168 policies kept — €70,560 of annual premium that stays with you. A policyholder settled in twelve days does not shop around; one who has chased three times does.
The next step I propose: a flood and storm protocol, written from those three days — trigger thresholds, geographic grouping, bulk letters already drafted, order of priority for files. It runs to two pages and it is ready: approve it before autumn, and the next peak will be an ordinary Sunday. claim-peak_1400-notifications-in-3-days.pdf0 temporary hours, €46,000 avoided, 3.1 days of first response
⛓ Sourced · 1,400 notifications of the 14th, 15th and 16th, temporary staff records of the previous year, post-claim lapses over 24 months
What I did over those three days, while your handlers were calling policyholders:
· 1,400 completeness checks, and 212 document requests sent the same day, each one complete.
· The sorting by nature: 1,090 standard material damage files, 218 files needing an adjuster, 92 out of the assisted workflow — including 11 bodily injury files sent to the referent within the hour.
· Grouping by town and by street: the 340 files in one neighbourhood went to the same adjuster — 4 trips instead of 340 appointments booked one by one.
What it spared you, in euros: last year a comparable peak took 640 hours of temporary staff and €46,000. This year: no temporary hours at all, and first-response time stayed at 3.1 days.
And the measure that speaks to your sales management: the lapse rate in the twelve months following a claim went from 9.4 % to 6.1 %. Across 5,100 standard files, that is 168 policies kept — €70,560 of annual premium that stays with you. A policyholder settled in twelve days does not shop around; one who has chased three times does.
The next step I propose: a flood and storm protocol, written from those three days — trigger thresholds, geographic grouping, bulk letters already drafted, order of priority for files. It runs to two pages and it is ready: approve it before autumn, and the next peak will be an ordinary Sunday. claim-peak_1400-notifications-in-3-days.pdf0 temporary hours, €46,000 avoided, 3.1 days of first response
⛓ Sourced · 1,400 notifications of the 14th, 15th and 16th, temporary staff records of the previous year, post-claim lapses over 24 months
And recovery against liable third parties? We know we forget some, and it goes straight to loss.
410 recovery actions were left to be brought over six years, and I have put a figure on them: €214,000 you were entitled to recover, part of which is still claimable today.
Recovery is the process by which an insurer who has settled with its policyholder turns to the third party responsible for the damage to get back what it paid.
What reading your six years shows: 1,180 files involved an identified third party — a neighbour, a contractor who worked on site, a managing agent, a supplier. 770 led to a recovery. 410 did not, for an average amount of €522. These are not careless omissions: they are files where the third party appeared in a document nobody reopened at closing time.
What I do now, from notification onwards: I spot the third party in the file documents — accident report, adjuster's report, contractor invoice — and I build the recovery file alongside the handling, with the documents, the amount paid and the basis. It waits for your signature at closing, instead of never existing.
What that gave over the quarter: 96 recovery files built, 71 successful, €39,400 collected — and the remaining 25 are under way, none dropped for want of being seen.
What I suggest on top, and it is money lying idle: of the 410 recoveries never pursued, 138 are still within your internal claim deadlines. Those 138 files are built, documents included — €72,000 claimable. You sign as many as you want, and you can start with the 40 largest, which alone make €41,000.
The next step: that spotting the third party become a step of the protocol, at notification. It is the only place where it is free: later, a closed file has to be reopened. third-party-recovery_410-never-pursued.pdf€214,000 not recovered, 138 files still claimable
⛓ Sourced · 41,200 closed files, documents naming a third party, log of recoveries pursued, amounts paid
Recovery is the process by which an insurer who has settled with its policyholder turns to the third party responsible for the damage to get back what it paid.
What reading your six years shows: 1,180 files involved an identified third party — a neighbour, a contractor who worked on site, a managing agent, a supplier. 770 led to a recovery. 410 did not, for an average amount of €522. These are not careless omissions: they are files where the third party appeared in a document nobody reopened at closing time.
What I do now, from notification onwards: I spot the third party in the file documents — accident report, adjuster's report, contractor invoice — and I build the recovery file alongside the handling, with the documents, the amount paid and the basis. It waits for your signature at closing, instead of never existing.
What that gave over the quarter: 96 recovery files built, 71 successful, €39,400 collected — and the remaining 25 are under way, none dropped for want of being seen.
What I suggest on top, and it is money lying idle: of the 410 recoveries never pursued, 138 are still within your internal claim deadlines. Those 138 files are built, documents included — €72,000 claimable. You sign as many as you want, and you can start with the 40 largest, which alone make €41,000.
The next step: that spotting the third party become a step of the protocol, at notification. It is the only place where it is free: later, a closed file has to be reopened. third-party-recovery_410-never-pursued.pdf€214,000 not recovered, 138 files still claimable
⛓ Sourced · 41,200 closed files, documents naming a third party, log of recoveries pursued, amounts paid
And doubtful files? We let some through, and we know it.
Your four flagging criteria produced 210 flags last year, of which 38 were confirmed after review. I have read your six years and I propose six more, written in the same form and already quantified.
For each of the six you have three numbers before deciding: the number of flags it would have produced over six years, the share confirmed after review by your referents, and the files it would have let through. You choose on figures, not on a hunch.
· Criterion « notification within 30 days of taking out the policy or of an extension endorsement »: 186 flags over six years, 61 confirmed — 33 %. It is the best of the six, and it was not on your list.
· Criterion « same invoice beneficiary across three unrelated claims »: 74 flags, 29 confirmed — 39 %, but it only fires at the third file: so I raise it from the second one, as a watch item, without it counting as a flag.
· The other four, between 12 % and 21 % confirmation, each with its volume and its review cost.
And what I propose on one of yours, figures in hand: your criterion « amount above twice the average for that claim type » produced 96 flags for 4 confirmations — 4 %. I have written the tightened version and run it over your six years: 22 flags instead of 96, the 4 confirmations kept, 74 honest policyholders no longer looked at closely. Sign it and it is in force tonight.
What does not change, and it is what protects the house: a flagged file leaves the assisted workflow and goes to entirely human handling, with the file assembled and the indication that raised it. A flag is never a conclusion, and it never has any effect on the policyholder until a person has reviewed it.
The next step: your referents read the six criteria and the tightened version — one hour. Over six years, the set would have produced 108 more confirmations for 74 fewer pointless reviews. flagging-criteria_6-proposed-1-tightened.pdf96 flags down to 22, the 4 confirmations kept
⛓ Sourced · the 4 criteria in force and their flags, 6 years of closed files and referent reviews
For each of the six you have three numbers before deciding: the number of flags it would have produced over six years, the share confirmed after review by your referents, and the files it would have let through. You choose on figures, not on a hunch.
· Criterion « notification within 30 days of taking out the policy or of an extension endorsement »: 186 flags over six years, 61 confirmed — 33 %. It is the best of the six, and it was not on your list.
· Criterion « same invoice beneficiary across three unrelated claims »: 74 flags, 29 confirmed — 39 %, but it only fires at the third file: so I raise it from the second one, as a watch item, without it counting as a flag.
· The other four, between 12 % and 21 % confirmation, each with its volume and its review cost.
And what I propose on one of yours, figures in hand: your criterion « amount above twice the average for that claim type » produced 96 flags for 4 confirmations — 4 %. I have written the tightened version and run it over your six years: 22 flags instead of 96, the 4 confirmations kept, 74 honest policyholders no longer looked at closely. Sign it and it is in force tonight.
What does not change, and it is what protects the house: a flagged file leaves the assisted workflow and goes to entirely human handling, with the file assembled and the indication that raised it. A flag is never a conclusion, and it never has any effect on the policyholder until a person has reviewed it.
The next step: your referents read the six criteria and the tightened version — one hour. Over six years, the set would have produced 108 more confirmations for 74 fewer pointless reviews. flagging-criteria_6-proposed-1-tightened.pdf96 flags down to 22, the 4 confirmations kept
⛓ Sourced · the 4 criteria in force and their flags, 6 years of closed files and referent reviews
Fictional simulation — not a client result
One quarter: 7,650 handling hours returned over the year, and the calculation is yours — it fits in three lines you can redo.
The three items you were measuring:
· Chaining the handling steps: 65 % → 9 % of the handling time of a file, 55 minutes brought down to 7 minutes 30 seconds, across 5,100 standard files — 4,037 hours.
· Preparing the costing: 45 % → 8 %, 54 minutes brought down to 9 minutes 40 seconds, across 2,400 files with an adjuster — 1,773 hours.
· Building the audit trail: 30 % → 4 %, 25 minutes brought down to 3 minutes 20 seconds, across 5,100 files — 1,842 hours.
What that is worth at your fully loaded rate of €42: €321,000 of handling capacity returned a year, for €4,752 of annual subscription. These hours are not a payroll saving: they went into assessment, the telephone and recoveries — your department absorbed the storm without a single temporary hour, where the previous peak had cost €46,000.
The four measures your policyholders feel:
· Average settlement time: 34 days → 12.
· First reply within 5 working days: met 71 % → 98 %.
· Files going back out for a missing document: 38 % → 9 %.
· Lapse within twelve months of a claim: 9.4 % → 6.1 %, that is 168 policies kept and €70,560 of annual premium.
And two figures for your finance department: €39,400 of recoveries collected this quarter, and €72,000 still claimable on the 138 files reopened.
The next step I propose: the same report by product — home, motor, commercial multi-risk. It is the view your underwriting committee lacks, and it is already written for the three products. quarterly-review_7650-hours-returned.pdf65→9, 45→8, 30→4, and the calculation redoable on one page
⛓ Sourced · handling log of the quarter, delay records, post-claim lapses, recovery log
The three items you were measuring:
· Chaining the handling steps: 65 % → 9 % of the handling time of a file, 55 minutes brought down to 7 minutes 30 seconds, across 5,100 standard files — 4,037 hours.
· Preparing the costing: 45 % → 8 %, 54 minutes brought down to 9 minutes 40 seconds, across 2,400 files with an adjuster — 1,773 hours.
· Building the audit trail: 30 % → 4 %, 25 minutes brought down to 3 minutes 20 seconds, across 5,100 files — 1,842 hours.
What that is worth at your fully loaded rate of €42: €321,000 of handling capacity returned a year, for €4,752 of annual subscription. These hours are not a payroll saving: they went into assessment, the telephone and recoveries — your department absorbed the storm without a single temporary hour, where the previous peak had cost €46,000.
The four measures your policyholders feel:
· Average settlement time: 34 days → 12.
· First reply within 5 working days: met 71 % → 98 %.
· Files going back out for a missing document: 38 % → 9 %.
· Lapse within twelve months of a claim: 9.4 % → 6.1 %, that is 168 policies kept and €70,560 of annual premium.
And two figures for your finance department: €39,400 of recoveries collected this quarter, and €72,000 still claimable on the 138 files reopened.
The next step I propose: the same report by product — home, motor, commercial multi-risk. It is the view your underwriting committee lacks, and it is already written for the three products. quarterly-review_7650-hours-returned.pdf65→9, 45→8, 30→4, and the calculation redoable on one page
⛓ Sourced · handling log of the quarter, delay records, post-claim lapses, recovery log
And your mistakes? On a costing, a mistake costs money or costs a policyholder.
Across the 1,240 costings prepared in the first quarter, 112 had to be corrected on substance — 9 %. That is the figure I publish about myself, and it is better that you get it from me than from a policyholder.
The real cause, measured rather than assumed: 84 of the 112 concerned policies whose recent endorsement was not carried into the cover repository. Your repository updates once a month; an endorsement signed on the 3rd only appeared there on the 1st of the following month. I was faithfully reading a source that was a month out of date. The other 28 concerned depreciation scales specific to three old commercial policies.
What I did about it, and it is measured: I now read the endorsement attached to the file, not the monthly repository alone; the 340 policies carrying an endorsement less than a year old have been read one by one; and the three old scales have been written in the same form as the others.
The following quarter: 19 costings corrected out of 1,310 — 1.5 %.
And the point that matters most for your liability: none of those 112 produced a wrong payment. A costing stops at the validation point, and every item carries its clause: your handler rules out a wrong line in two minutes — that is exactly what the reference to the contract is for, and it is why it is not negotiable.
The framework measures, over the quarter: 2,550 files handled, 2,550 decisions taken by a person, 0 settlement set without human approval, 0 payment triggered automatically, and 0 policyholder data leaving the network.
The next step: that the cover repository update when the endorsement is signed, not once a month. It is not me that this fixes: it is a source your handlers consult too — across the six years read, 61 files had been costed by hand on an endorsement that was not carried over. costings-corrected_112-then-19.pdf9 % → 1.5 %, measured cause, 0 wrong payment
⛓ Sourced · log of costings and their corrections over two quarters, cover repository, 340 policies with endorsements
The real cause, measured rather than assumed: 84 of the 112 concerned policies whose recent endorsement was not carried into the cover repository. Your repository updates once a month; an endorsement signed on the 3rd only appeared there on the 1st of the following month. I was faithfully reading a source that was a month out of date. The other 28 concerned depreciation scales specific to three old commercial policies.
What I did about it, and it is measured: I now read the endorsement attached to the file, not the monthly repository alone; the 340 policies carrying an endorsement less than a year old have been read one by one; and the three old scales have been written in the same form as the others.
The following quarter: 19 costings corrected out of 1,310 — 1.5 %.
And the point that matters most for your liability: none of those 112 produced a wrong payment. A costing stops at the validation point, and every item carries its clause: your handler rules out a wrong line in two minutes — that is exactly what the reference to the contract is for, and it is why it is not negotiable.
The framework measures, over the quarter: 2,550 files handled, 2,550 decisions taken by a person, 0 settlement set without human approval, 0 payment triggered automatically, and 0 policyholder data leaving the network.
The next step: that the cover repository update when the endorsement is signed, not once a month. It is not me that this fixes: it is a source your handlers consult too — across the six years read, 61 files had been costed by hand on an endorsement that was not carried over. costings-corrected_112-then-19.pdf9 % → 1.5 %, measured cause, 0 wrong payment
⛓ Sourced · log of costings and their corrections over two quarters, cover repository, 340 policies with endorsements
Last question, the one that will decide management: what do you do without asking us, and how do we leave if we want to leave?
Three gestures without asking, two acts under a signed mandate, and an exit that can be checked before committing rather than after.
The three gestures, each withdrawable with a single word:
· I read the files filed every night and I build the status of each. And the reverse is true too: a file you delete leaves the index at the same hour — I keep no copy of what you decided to erase.
· I chase any adjuster whose report is past its announced date, once, politely, with the file number. That gesture is what took three days out of the adjustment wait.
· I hand you every morning the list of files that have reached a validation point, ordered by age. It is the only thing I send of my own accord, and it goes only to your department.
What stays with a person, and it is what gives your decisions their value: setting the settlement amount · triggering a payment · ruling on disputed cover · reviewing a flagged file · handling a bodily injury claim, a dispute or a suspected fraud, from end to end. Over the quarter: 2,550 files, 2,550 human decisions.
The exit, since that is what decides a management board: no migration on the way in, therefore none on the way out. Your claims management system is not replaced, nor is your document platform, the index is deleted and contained no file — only what is needed to find them where they are, and the map of the eleven steps, the 9 doctrine notes, the 6 rewritten letters, the flagging criteria and the storm protocol stay yours, readable without us. That is the asset this quarter created, and it would not be honest for it to remain with us.
What I propose so that this is not just a sentence: a dry-run exit at the end of the first quarter, half a day — we switch off, we check that the department handles exactly as before, we switch back on. The protocol is written and the date that costs you least is the first Tuesday of August: it has been your notification trough for six years, 9 files on average. Management will know what the promise is worth before committing to a second year. technical-framework_where-your-policyholder-files-live.pdfReversibility: 0 migration in, 0 migration out
✎ Framework · configuration of the standing gestures, decision log, dry-run exit protocol, export formats
The three gestures, each withdrawable with a single word:
· I read the files filed every night and I build the status of each. And the reverse is true too: a file you delete leaves the index at the same hour — I keep no copy of what you decided to erase.
· I chase any adjuster whose report is past its announced date, once, politely, with the file number. That gesture is what took three days out of the adjustment wait.
· I hand you every morning the list of files that have reached a validation point, ordered by age. It is the only thing I send of my own accord, and it goes only to your department.
What stays with a person, and it is what gives your decisions their value: setting the settlement amount · triggering a payment · ruling on disputed cover · reviewing a flagged file · handling a bodily injury claim, a dispute or a suspected fraud, from end to end. Over the quarter: 2,550 files, 2,550 human decisions.
The exit, since that is what decides a management board: no migration on the way in, therefore none on the way out. Your claims management system is not replaced, nor is your document platform, the index is deleted and contained no file — only what is needed to find them where they are, and the map of the eleven steps, the 9 doctrine notes, the 6 rewritten letters, the flagging criteria and the storm protocol stay yours, readable without us. That is the asset this quarter created, and it would not be honest for it to remain with us.
What I propose so that this is not just a sentence: a dry-run exit at the end of the first quarter, half a day — we switch off, we check that the department handles exactly as before, we switch back on. The protocol is written and the date that costs you least is the first Tuesday of August: it has been your notification trough for six years, 9 files on average. Management will know what the promise is worth before committing to a second year. technical-framework_where-your-policyholder-files-live.pdfReversibility: 0 migration in, 0 migration out
✎ Framework · configuration of the standing gestures, decision log, dry-run exit protocol, export formats
Data hosted in France · sovereign model as an option
Your case is not here? That is exactly what a 30-minute conversation is for. Discuss your project →
How it works
The stages of your AI agent project
1
Audit & scoping
30 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.
The context
What does an AI agent bring to working your claims?
A file worked without dead time, with a record of each step, shortens the time the policyholder waits for an answer.
! The issue
Working a claim chains together several steps and several sources — notification, documents, loss adjustment, contract. The waiting between those steps makes up most of the delay the policyholder feels. The agent runs the chain, keeps the file's state and prepares the costing item by item with the applicable limits.
✓ Our answer
The settlement department receives files that are worked and a documented costing, and spends its time on the decision. Setting a settlement concerns money and the policyholder's rights: that approval point stays human, as do the files excluded from automated handling. Data hosted in France on an isolated resource; the AI model is called at its publisher, except with the Sovereign model option. With that option, your policyholders' personal and health data is entrusted to no third party.
The decisive point
Policyholders' personal and health data: sovereignty & compliance
A claim file may contain health data and engages contractual rights. Here is how it is protected.
Local inference
With the 100% Sovereign package, the agent runs on a machine belonging to your organisation: no policyholder data and no document from a file leaves the network.
Hosting in France
Otherwise, with the Agility and Serenity packages, a dedicated, isolated resource hosted in France, under French law: data hosted in France; the AI model is called at its publisher, except with the Sovereign model option.
Reduced extraterritorial exposure
For policyholders' personal and health data, the architecture aims to reduce exposure to the Cloud Act and FISA 702; being located in France or in the European Union does not, on its own, guarantee immunity.
Isolated resource
No pooling: an environment strictly dedicated to your organisation and its handling protocols.
Every step recorded and referenced
Steps completed, documents received and items costed are kept with their reference in the contract; encryption, role-based access and logging that can be used in an inspection.
AI Act: governed deployment
The agent carries out what you have authorised; no settlement is set and no payment is triggered 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.
The gain
How many files can a department carry to a decision?
By taking on the chaining of the steps, the effort shifts towards judgement and the relationship with the policyholder. How large the gain is depends on your volume and remains to be confirmed by a pilot.
Chaining the handling steps
Preparing the costing
Building the file's record
Indicative figures, not contractual, to be confirmed by a pilot on your volume of files and the complexity of your protocols. Setting a settlement concerns money and the policyholder's rights: that decision belongs to the settlement department.
Indicative estimate, to be confirmed by a pilot on your real files (general terms of sale, article 4).
Our commitment
Four guarantees that matter to your policyholders
With 100% Sovereign, your policyholders' data stays with youData hosted in France on an isolated resource; the AI model is called at its publisher, except with the Sovereign model option. With that option, no policyholder data entrusted to a third party, no data used to train a model.
Data in France, under French lawPolicyholders' personal and health data: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The settlement department keeps the decisionThe agent produces files worked through to the decision, which can be checked and altered; the decisions remain yours.
Human oversight & traceabilityOn your volume of files and the complexity of your protocols: systematic logging and tracking, in line with the AI Act.
Going further
Related resources
Frequently asked questions
Your questions, our answers
Does the agent set the settlement?
No. It prepares the costing item by item with the applicable limits and their reference in the contract. Setting the settlement is an approval point that stays human.
How does this differ from level 1 claims?
Level 1 builds the file; this offer runs the full handling, loss adjustment and costing included, up to the point of decision.
Which files are excluded from automated handling?
Personal injury claims, disputed files and suspected fraud, excluded from the design stage and routed to entirely human handling.
Can the record be used in an inspection?
Yes: steps completed, documents received and items costed are kept with their reference, which makes it possible to reconstruct the handling afterwards.
Is health data protected?
Yes. The agent is hosted in France, on local inference or an isolated resource, with reinforced partitioning for these categories, with the deployment objective of processing and access operated within the European Union and an architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
How long does it take to deploy this agent?
A few months as a rule, depending on the volume of files and the complexity of your protocols, after a meeting with our sales team then phases of design, integration and testing.
Related use cases
Other agents for insurance
Let's talk
Let's size up the potential in your claims
30 minutes to frame your protocols and your volumes — hosted in France, supervised, with no commitment.