● Public sector — Customs officer
AI agent for customs officers (administrative support)
An assistant that answers traders, prepares the regulatory documentation and supports the handling of declarations — the officer keeps the inspection and the decision. A strictly administrative and documentary scope: no targeting, no surveillance, no decision about people. Hosted in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
Hosted in France
Traders' data protected
GDPR & AI Act: governed deployment
No targeting, no inspection
Updated on
Deployed in a few weeks
Customs assistant · hosted in France
A trader asks which regime applies to goods returning after repair outside the EU. Prepare a documented reply.
Reply prepared. The case falls under the outward processing regime: the reply recalls the principle of taxing the value added by the repair, the prior authorisation required and the evidence to produce on re-importation. Dated references attached for checking.
For review before sending to the trader.
⛓ Source · your official regulatory sources, dated references
For review before sending to the trader.
⛓ Source · your official regulatory sources, dated references
Check the documentary consistency of this batch of declarations before handling.
Documentary check carried out on 24 declarations: 21 files complete. Three discrepancies flagged: a missing invoice, a currency inconsistency between invoice and declaration, a certificate of origin not attached.
List of discrepancies passed on for checking — no handling started without your agreement.
✎ Action · discrepancies flagged — the customs officer approves
List of discrepancies passed on for checking — no handling started without your agreement.
✎ Action · discrepancies flagged — the customs officer approves
Local inference · no data outside the EU
Data hosted in France
Sovereign by designLocal inference or hosting in France
GDPR & AI Act: governed deploymentTraceability & human oversight
TurnkeyDesigned, installed and operated for you
The customs officer decidesDocumentary support, never targeting
✦ In brief
In customs, a Blue Lemon Agent agent assists officers with the administrative and documentary tasks — replies to traders (procedures, tariff classification, regimes), up-to-date regulatory documentation, checking the documentary consistency of declarations. Its scope is strictly administrative: no targeting, no surveillance, no decision about people. It runs on local inference or is hosted in France on a dedicated, isolated resource, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. The customs officer keeps the inspection and the decision. Live within a few weeks. Your public-sector staff write to it from Microsoft Teams, Slack or their email, and traders reach it on WhatsApp Business, the website chat or email — with no account to create and nothing to install. Reaching the administration from the tool people already have means less non-take-up of rights and equal access to the service. 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
7
uses ready to deploy on this scope
0
decision taken without human approval
Reference points describing our offer, not results measured at a client. The scale of the gain is confirmed by a pilot on your own scope.
The context
Why AI matters to customs — and why it demands guarantees
Tariff classifications, special regimes, regulations that never stop changing: customs officers answer traders every day for whom every documentary error is expensive. The data handled — trade flows, declarations — demands strict confidentiality.
! The issue
Customs officers are caught between traders who expect precise regulatory answers and documentation in permanent change (the Union Customs Code, tariff classifications, special regimes). Yet most consumer AI tools amount to entrusting declarations, trade flows and traders' data to a third party, often hosted outside Europe and subject to the Cloud Act.
✓ Our answer
For customs data, AI is only of interest if it is sovereign, confidential and strictly bounded. Local inference or an isolated resource hosted in France, systematic human oversight, a scope limited to administrative and documentary support: no targeting, surveillance or inspection-decision function, by design and in line with the AI Act. The aim is not to replace the officer, but to give them back time from documentation and repetitive replies.
The decisive point
Protecting customs data: sovereignty & compliance
A customs unit handles sensitive commercial and declaratory data. Here is how the architecture of our agents protects it — and how its scope is strictly bounded.
Local inference
The agent can run on a machine at the unit: no data leaves the network, nothing passes through a cloud.
Hosting in France
Otherwise, a dedicated and isolated resource, hosted in France under French law — the data: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
As regards traders' data, exposure to the Cloud Act and FISA 702 is reduced by design; location alone does not guarantee immunity.
One isolated resource per department
No pooling of data: an environment strictly dedicated to your unit, guaranteeing the continuity of the public service.
Encryption & controlled access
Encryption in transit and at rest, role-based access (RBAC), strong authentication and logging.
AI Act: governed deployment
A strictly documentary scope: no targeting, no decision about people; 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.
- The encryption mechanisms in transit and at rest, their components and key management are those documented for the architecture chosen.
- Roles and permissions are configured and accepted for the identities and systems actually connected.
- The events logged, their content, their retention period and who may access them are defined for the deployment chosen.
For the most sensitive data, SecNumCloud and HDS options are available 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 public body in this demonstration
Fictional public bodyVal-Sireuil customs office — clearance and procedures management service
- Sector
- Customs administration — clearance, special procedures, information for traders and declarants; counter open 30 hours a week
- Headcount
- 22 officers — 4 on trader information, 6 in the procedures management unit and 8 processing declarations; the digital agent is connected to administrative and documentary support only
- Public served
- 610 liable companies and declarants followed, 180 of which file every week — 6,330 requests a year: 4,100 calls, 1,450 written questions, 780 messages
- Scale
- 18,400 declarations filed a year, of which 3,600 go through the office's documentary check in batches of 24; 380 regulatory letters and 96 applications for a special procedure authorisation
- Tools in place
- The administration's filing applications, e-mail, the office's document base and nine years of archived written answers — the agent connects to them read-only, nothing is replaced or migrated
- Who decides
- The sworn customs officer decides on checks and signs; the head of the procedures unit approves answers to traders before they go out; the head of office signs the mandates
- The room for improvement
- 38 % of traders' calls ring out beyond the 30 opening hours; a written question waits 11 days for its answer; one filed declaration in five comes back as a request for a supporting document; and 6 of the 41 sheets in the document base predate the update that amends them
At Val-Sireuil, four officers answer traders and six run the procedures while 18,400 declarations pass through each year. The data handled — invoices, contracts, companies' trade flows — is among the most sensitive a service receives, and it does not belong to that service. The agent runs locally on a machine in the office, reads the nine years of written answers without ever writing to them, scores neither company nor person, and sends nothing out without a public officer's approval. The exchanges below cover a year, from the overhaul of the document base to the review presented to management.
This public body, its figures and the exchanges that follow were invented for the demonstration. They illustrate a common situation; they describe no real service.
Customs assistant · hosted in France
I reread your nine years of written answers over the weekend — 7,400 answers to traders, 41 sheets from your document base, 12 letter templates — and I have drawn out a finding your files were carrying without anyone having the time to see it.
The gap measured, and it commands everything else: 6 of your 41 reference sheets predate the regulatory update that amends them. The sheet on goods returning after repair outside the Union still points to the form that has since been replaced; the sheet on proofs of origin gives the address of a service that has moved. These are not drafting mistakes: they are sheets that aged while nobody had the time to reopen them.
What that costs today, across three items you can check yourself:
· Preparing a documented answer to a trader accounts for 60 % of the time a question takes — 48 minutes out of the 80 a written question takes from arrival to dispatch.
· Laying out a regulatory letter: 40 %, that is 24 minutes out of 60.
· The documentary check of a batch of 24 declarations: the whole of the task, done by hand today — 96 minutes per batch.
What I propose, and it is not a promise — it is already written: I rewrote the 6 sheets overnight, each aligned on the version in force of the official sources the office has designated, with the replaced line, the line replacing it and the date I read it, side by side.
What that would have changed over the past year, since that is the only measure that counts: the 214 answers sent last year on the basis of one of those 6 sheets would have quoted the version in force, and the 31 traders who called back to check would not have had to.
What I do with the three items once the sheets are settled: I prepare, and you keep what only you can do — check, correct and approve. Preparation drops to 10 %, 10 % and 20 %: 8 minutes, 6 minutes, 19 minutes.
The next step, and it takes twenty minutes: you reread the three most used sheets, I bring them to you tomorrow morning. The unit head settles the version and it is live that same evening — it is the service's approval that makes an answer binding towards the trader who relies on it, and it is the only act I leave to you. document-base-inventory_41-sheets-6-to-revise.pdf41 sheets reread, 6 rewritten, the update that governs each
⛓ Sourced · 9 years of the office's written answers, document base, official sources designated by the service, 6 sheets rewritten
The gap measured, and it commands everything else: 6 of your 41 reference sheets predate the regulatory update that amends them. The sheet on goods returning after repair outside the Union still points to the form that has since been replaced; the sheet on proofs of origin gives the address of a service that has moved. These are not drafting mistakes: they are sheets that aged while nobody had the time to reopen them.
What that costs today, across three items you can check yourself:
· Preparing a documented answer to a trader accounts for 60 % of the time a question takes — 48 minutes out of the 80 a written question takes from arrival to dispatch.
· Laying out a regulatory letter: 40 %, that is 24 minutes out of 60.
· The documentary check of a batch of 24 declarations: the whole of the task, done by hand today — 96 minutes per batch.
What I propose, and it is not a promise — it is already written: I rewrote the 6 sheets overnight, each aligned on the version in force of the official sources the office has designated, with the replaced line, the line replacing it and the date I read it, side by side.
What that would have changed over the past year, since that is the only measure that counts: the 214 answers sent last year on the basis of one of those 6 sheets would have quoted the version in force, and the 31 traders who called back to check would not have had to.
What I do with the three items once the sheets are settled: I prepare, and you keep what only you can do — check, correct and approve. Preparation drops to 10 %, 10 % and 20 %: 8 minutes, 6 minutes, 19 minutes.
The next step, and it takes twenty minutes: you reread the three most used sheets, I bring them to you tomorrow morning. The unit head settles the version and it is live that same evening — it is the service's approval that makes an answer binding towards the trader who relies on it, and it is the only act I leave to you. document-base-inventory_41-sheets-6-to-revise.pdf41 sheets reread, 6 rewritten, the update that governs each
⛓ Sourced · 9 years of the office's written answers, document base, official sources designated by the service, 6 sheets rewritten
There are four of us on trader information and I do not have half a day to spend on reference sheets. Where do I start?
With three sheets, not forty-one — and I have ranked them by the time they give you back, not alphabetically.
What the count of your 1,450 written questions from last year says:
· Special procedures for repair and processing: 412 answers — a special procedure is the framework that lets goods leave, enter or stay otherwise than by simply being placed on the market: repair outside the Union, processing, storage, transit.
· Proofs of origin: 268 answers.
· Supporting documents expected with a declaration: 204 answers.
Those three sheets carried 884 of the year's 1,450 answers: three sheets out of forty-one, 61 % of what your unit writes. The other 38 can wait until autumn without changing anything to your workload.
What I bring you within those twenty minutes, and that nobody has the time to do: the three up-to-date versions, already written, the governing update alongside with its publication date, and the number of answers each would have served last year alone — 412, 268 and 204.
And what stops this work having to be redone in two years: I go through the official sources the office has designated every night, and within twenty-four hours I write the updated version of any sheet an evolution amends. Over the last twelve months, 23 evolutions affected your sheets: 17 reached them, 6 never did — and those are exactly the 6 sheets I rewrote. Every reference goes out with its source and the date I read it: I never quote from memory, and that is what lets the trader check your answers for himself.
The next step I propose: that the updated version reaches you as a draft the morning after the evolution, together with the list of the year's answers that relied on the former version. Last year those 214 answers would have gone out on the version in force, and your four officers would have had nothing to catch up on. You approve the principle once, and your 41 sheets stop ageing in silence. unit-workload_where-the-time-goes.pdf6,330 requests a year, 3 sheets for 61 % of the answers
⛓ Sourced · answers counted per sheet over 9 years, official sources designated by the office, 3 rewritten versions
What the count of your 1,450 written questions from last year says:
· Special procedures for repair and processing: 412 answers — a special procedure is the framework that lets goods leave, enter or stay otherwise than by simply being placed on the market: repair outside the Union, processing, storage, transit.
· Proofs of origin: 268 answers.
· Supporting documents expected with a declaration: 204 answers.
Those three sheets carried 884 of the year's 1,450 answers: three sheets out of forty-one, 61 % of what your unit writes. The other 38 can wait until autumn without changing anything to your workload.
What I bring you within those twenty minutes, and that nobody has the time to do: the three up-to-date versions, already written, the governing update alongside with its publication date, and the number of answers each would have served last year alone — 412, 268 and 204.
And what stops this work having to be redone in two years: I go through the official sources the office has designated every night, and within twenty-four hours I write the updated version of any sheet an evolution amends. Over the last twelve months, 23 evolutions affected your sheets: 17 reached them, 6 never did — and those are exactly the 6 sheets I rewrote. Every reference goes out with its source and the date I read it: I never quote from memory, and that is what lets the trader check your answers for himself.
The next step I propose: that the updated version reaches you as a draft the morning after the evolution, together with the list of the year's answers that relied on the former version. Last year those 214 answers would have gone out on the version in force, and your four officers would have had nothing to catch up on. You approve the principle once, and your 41 sheets stop ageing in silence. unit-workload_where-the-time-goes.pdf6,330 requests a year, 3 sheets for 61 % of the answers
⛓ Sourced · answers counted per sheet over 9 years, official sources designated by the office, 3 rewritten versions
All of this means you read our declarations, the companies' invoices and our exchanges with them. Where does that go?
It goes nowhere. I run on local inference on a machine in the office, and nothing I read leaves your network.
Local inference means the model computes on your own machine: a company's invoice or the body of a declaration crosses no outside network to be processed. If the service would rather not run a machine, the other route is an isolated resource hosted in France, dedicated to your office — no pooling with any other service, which is the condition for the continuity of yours.
What that changes, point by point:
· Traders' commercial data trains no model, neither ours nor a third party's. What I read about a company serves that company's file and nothing else.
· I work read-only on your filing applications, and the technical account I read through has no right to write — that is firmer than a promise, because it can be verified with a single command.
· Encryption in transit and at rest, role-based access — rights follow the job: information staff do not open control files, and the reverse holds too. 5 roles for your 22 officers, and the log shows 0 out-of-role access since go-live.
· Hosting in France, under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
· A complete log: who asked what, when, what I produced and who approved it.
And one thing I say myself to every trader, in the very first sentence: that I am a digital assistant of the office and not an officer. This is not an option you could switch off — the European regulation on artificial intelligence requires that anyone interacting with an AI system be informed —, and the trader may ask for an officer at any moment: I then take the request and hand it to you with the subject already noted.
The figures that sum all this up: Hosted in France, 0 item of data out of the office network across the year's 18,400 declarations, processing in the EU targeted.
What I propose: that I keep up to date the record your management and your data protection officer will ask for — hosting, data processed, retention periods, who accesses what. It is asked for once a year and takes two days to rebuild; the first version is already written and waiting for you. technical-framework_where-traders-data-lives.pdfLocal inference, read-only, processing in the EU targeted
✎ Framework · deployment architecture, technical account rights, access log, first version of the register record
Local inference means the model computes on your own machine: a company's invoice or the body of a declaration crosses no outside network to be processed. If the service would rather not run a machine, the other route is an isolated resource hosted in France, dedicated to your office — no pooling with any other service, which is the condition for the continuity of yours.
What that changes, point by point:
· Traders' commercial data trains no model, neither ours nor a third party's. What I read about a company serves that company's file and nothing else.
· I work read-only on your filing applications, and the technical account I read through has no right to write — that is firmer than a promise, because it can be verified with a single command.
· Encryption in transit and at rest, role-based access — rights follow the job: information staff do not open control files, and the reverse holds too. 5 roles for your 22 officers, and the log shows 0 out-of-role access since go-live.
· Hosting in France, under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
· A complete log: who asked what, when, what I produced and who approved it.
And one thing I say myself to every trader, in the very first sentence: that I am a digital assistant of the office and not an officer. This is not an option you could switch off — the European regulation on artificial intelligence requires that anyone interacting with an AI system be informed —, and the trader may ask for an officer at any moment: I then take the request and hand it to you with the subject already noted.
The figures that sum all this up: Hosted in France, 0 item of data out of the office network across the year's 18,400 declarations, processing in the EU targeted.
What I propose: that I keep up to date the record your management and your data protection officer will ask for — hosting, data processed, retention periods, who accesses what. It is asked for once a year and takes two days to rebuild; the first version is already written and waiting for you. technical-framework_where-traders-data-lives.pdfLocal inference, read-only, processing in the EU targeted
✎ Framework · deployment architecture, technical account rights, access log, first version of the register record
The office takes 4,100 calls from traders a year for thirty opening hours a week. The rest of the time it rings out and companies call back three times.
4,100 calls last year, 1,558 unanswered — 38 %. That is not a reception failure: it is a counter open thirty hours a week with four officers on information.
What I did with those 1,558 calls, by rereading your call-backs and your e-mails:
· 4,080 of your 6,330 requests are about fourteen subjects, always the same ones — supporting documents expected with a declaration, proofs of origin, special procedures, filing an authorisation application, amending a declaration, forms in force, processing times, contact details and opening hours, and six more.
· All fourteen have a written answer in your sheets or in the official sources the office has designated, and the fourteen answers are drafted, sourced and dated — they are waiting for you. The answer existed: what was missing was availability.
What I propose, and you keep the key to it: I answer at any hour, Saturdays included, and I say in my first sentence that I am a digital assistant of the office; the trader may ask for an officer at any moment.
The rule I hold to most firmly, and it is the one that protects the service: I answer within the office's remit, and whatever falls outside it goes back with the right desk — the competent service, its address, its deadline. The company leaves with a complete route, never with an approximation given in the administration's name.
And on the channels, because this is where equal access is decided: traders reach me on WhatsApp Business, the website chat, e-mail and IP telephony, with no account to create and nothing to install; your officers write to me from Microsoft Teams, Slack or their mailbox, without changing tools. These connections are included in every plan, at no extra cost and with no limit on their number — only the fees charged by the platforms themselves, WhatsApp Business charging per conversation, are passed on at actual cost, with nothing added by us. A three-person company with no customs department thus gets, at 9 p.m., the same sourced and dated answer as the group with a law firm: that is non-take-up avoided, and it is equal treatment.
The gain, in figures: preparing a documented answer goes from 60 % to 10 % of the question's time — from 48 minutes to 8, and over 1,450 written questions a year, that is 966 hours given back to the service. Unanswered calls go from 38 % to 7 %: what remains are those asking for an officer, and they reach you with the subject already noted.
The next step I propose: you reread the fourteen answers tomorrow, one by one — twenty minutes. As soon as they are approved the switchboard answers that very night, and every morning I hand you the page of what went out. trader-reception_4100-calls-14-subjects.pdf1,558 calls lost, 14 subjects, 14 answers already written
⛓ Sourced · 12 months of switchboard logs, traders' e-mails, the office's document base, 14 answers drafted
What I did with those 1,558 calls, by rereading your call-backs and your e-mails:
· 4,080 of your 6,330 requests are about fourteen subjects, always the same ones — supporting documents expected with a declaration, proofs of origin, special procedures, filing an authorisation application, amending a declaration, forms in force, processing times, contact details and opening hours, and six more.
· All fourteen have a written answer in your sheets or in the official sources the office has designated, and the fourteen answers are drafted, sourced and dated — they are waiting for you. The answer existed: what was missing was availability.
What I propose, and you keep the key to it: I answer at any hour, Saturdays included, and I say in my first sentence that I am a digital assistant of the office; the trader may ask for an officer at any moment.
The rule I hold to most firmly, and it is the one that protects the service: I answer within the office's remit, and whatever falls outside it goes back with the right desk — the competent service, its address, its deadline. The company leaves with a complete route, never with an approximation given in the administration's name.
And on the channels, because this is where equal access is decided: traders reach me on WhatsApp Business, the website chat, e-mail and IP telephony, with no account to create and nothing to install; your officers write to me from Microsoft Teams, Slack or their mailbox, without changing tools. These connections are included in every plan, at no extra cost and with no limit on their number — only the fees charged by the platforms themselves, WhatsApp Business charging per conversation, are passed on at actual cost, with nothing added by us. A three-person company with no customs department thus gets, at 9 p.m., the same sourced and dated answer as the group with a law firm: that is non-take-up avoided, and it is equal treatment.
The gain, in figures: preparing a documented answer goes from 60 % to 10 % of the question's time — from 48 minutes to 8, and over 1,450 written questions a year, that is 966 hours given back to the service. Unanswered calls go from 38 % to 7 %: what remains are those asking for an officer, and they reach you with the subject already noted.
The next step I propose: you reread the fourteen answers tomorrow, one by one — twenty minutes. As soon as they are approved the switchboard answers that very night, and every morning I hand you the page of what went out. trader-reception_4100-calls-14-subjects.pdf1,558 calls lost, 14 subjects, 14 answers already written
⛓ Sourced · 12 months of switchboard logs, traders' e-mails, the office's document base, 14 answers drafted
Answering, fine. But a wrong answer given to a declarant is carried by the office. I will not sign a blank cheque.
You do not have to choose between the two: I am offering you a written mandate, capped, dated and withdrawable with one word — and it is already drafted, waiting for you.
What it brings in, first, because that is what decides: of the year's 6,330 requests, 4,080 fall within the fourteen subjects — 2,640 calls, 902 written questions, 538 messages. The average time to answer a written question would go from 11 days to 2, and a question arriving on Friday evening would stop waiting until Tuesday.
What the mandate says, and it fits in six lines:
· Exact scope: the fourteen subjects, listed by name, and nothing else. Any question outside that list reaches you with a draft answer — written, sourced, ready to go out from you.
· Individual files in progress stay outside the mandate — a declaration being processed, an authorisation application under examination, a file followed by an officer: there, I hand you the answer already drafted, with the file's documents alongside, and you are the one who sends it. Over the trial quarter those answers went out within 4 hours instead of 11 days — and they went out from a public officer, which makes them reasoned and open to challenge; a piece of general information does not carry that weight.
· Every answer carries its source — the office's sheet or the designated official source, with its date — and a statement that it was prepared by a digital assistant of the office.
· Every morning you receive the record of the answers sent the day before, on one page. A wrong answer is caught in an hour, not in three weeks.
· Term: review after three months, with the record of what it changed. Without an explicit decision at the review, the mandate ends — renewal is what requires a signature, not stopping.
· Withdrawal: one word from you and direct sending stops within the minute. Answers go back to being drafts to approve; nothing else changes.
The decision belongs to the head of office — and it is taken on a text already written, with one signature. The mandate is drafted, and so is the notice to traders — the one that will be published on the service's website and sent to the 610 companies followed. You sign, and the service answers from the next morning; the review is already in your diary on the 15th of the third month. direct-answer-mandate_14-subjects-capped.pdf14 subjects, individual files handed over drafted, review at 3 months
✎ Framework · mandate drafted, list of the 14 subjects, notice to traders, request log
What it brings in, first, because that is what decides: of the year's 6,330 requests, 4,080 fall within the fourteen subjects — 2,640 calls, 902 written questions, 538 messages. The average time to answer a written question would go from 11 days to 2, and a question arriving on Friday evening would stop waiting until Tuesday.
What the mandate says, and it fits in six lines:
· Exact scope: the fourteen subjects, listed by name, and nothing else. Any question outside that list reaches you with a draft answer — written, sourced, ready to go out from you.
· Individual files in progress stay outside the mandate — a declaration being processed, an authorisation application under examination, a file followed by an officer: there, I hand you the answer already drafted, with the file's documents alongside, and you are the one who sends it. Over the trial quarter those answers went out within 4 hours instead of 11 days — and they went out from a public officer, which makes them reasoned and open to challenge; a piece of general information does not carry that weight.
· Every answer carries its source — the office's sheet or the designated official source, with its date — and a statement that it was prepared by a digital assistant of the office.
· Every morning you receive the record of the answers sent the day before, on one page. A wrong answer is caught in an hour, not in three weeks.
· Term: review after three months, with the record of what it changed. Without an explicit decision at the review, the mandate ends — renewal is what requires a signature, not stopping.
· Withdrawal: one word from you and direct sending stops within the minute. Answers go back to being drafts to approve; nothing else changes.
The decision belongs to the head of office — and it is taken on a text already written, with one signature. The mandate is drafted, and so is the notice to traders — the one that will be published on the service's website and sent to the 610 companies followed. You sign, and the service answers from the next morning; the review is already in your diary on the 15th of the third month. direct-answer-mandate_14-subjects-capped.pdf14 subjects, individual files handed over drafted, review at 3 months
✎ Framework · mandate drafted, list of the 14 subjects, notice to traders, request log
Many traders get lost in the online procedure and file incomplete cases. It all comes back to us.
I guide them screen by screen, and the result shows on your side: one filed declaration in five incomplete, brought down to one in fourteen.
What I do, and in this order — the order is what changes everything:
· The list of documents to prepare BEFORE starting, with what each must show. This is the most common drop-out: the declarant starts, a supporting document is missing, he closes the screen and comes back three days later.
· Reading the screen with him: what is asked, why, and what the service will do with it.
· The check for inconsistencies before validation — a quantity that does not match the invoice, a missing currency, a document announced but not uploaded.
On the declarant's business account the fastest route is also the only sound one, and it is the one I take: I prepare in front of him everything that gets copied across and mistyped — identification, file references, quantities, description of the goods —, and he is the one who enters his credentials and validates. His credentials are never entrusted to me and do not need to be: it is that act which makes the declaration his, therefore valid, and which means it can never be held against him on any other footing.
What that gave over the quarter: 512 guided sessions, 468 carried through to the end; the other 44 were waiting for a document to be obtained elsewhere, and each left with the request already drafted. On the office's side, that is 96 requests for supporting documents your officers did not have to write, and on the company's side, as many days not lost.
The next step I propose: a one-hour desk on Tuesday at the end of the day, run by me on the telephone and on the chat, with an officer of the unit reachable for anything outside the fourteen subjects. Over a trial quarter, 61 declarants came to it, 38 of them from small companies with no customs department — the ones who never call during office hours, because they are working at the same time. Say yes and the announcement goes out with the next notice to traders. online-procedures_512-guided-sessions.pdf468 carried to the end, 1 incomplete filing in 14
⛓ Sourced · log of guided sessions, the quarter's filings, the office's requests for supporting documents
What I do, and in this order — the order is what changes everything:
· The list of documents to prepare BEFORE starting, with what each must show. This is the most common drop-out: the declarant starts, a supporting document is missing, he closes the screen and comes back three days later.
· Reading the screen with him: what is asked, why, and what the service will do with it.
· The check for inconsistencies before validation — a quantity that does not match the invoice, a missing currency, a document announced but not uploaded.
On the declarant's business account the fastest route is also the only sound one, and it is the one I take: I prepare in front of him everything that gets copied across and mistyped — identification, file references, quantities, description of the goods —, and he is the one who enters his credentials and validates. His credentials are never entrusted to me and do not need to be: it is that act which makes the declaration his, therefore valid, and which means it can never be held against him on any other footing.
What that gave over the quarter: 512 guided sessions, 468 carried through to the end; the other 44 were waiting for a document to be obtained elsewhere, and each left with the request already drafted. On the office's side, that is 96 requests for supporting documents your officers did not have to write, and on the company's side, as many days not lost.
The next step I propose: a one-hour desk on Tuesday at the end of the day, run by me on the telephone and on the chat, with an officer of the unit reachable for anything outside the fourteen subjects. Over a trial quarter, 61 declarants came to it, 38 of them from small companies with no customs department — the ones who never call during office hours, because they are working at the same time. Say yes and the announcement goes out with the next notice to traders. online-procedures_512-guided-sessions.pdf468 carried to the end, 1 incomplete filing in 14
⛓ Sourced · log of guided sessions, the quarter's filings, the office's requests for supporting documents
I have a batch of 24 declarations to check before processing. Today that takes me the whole afternoon.
The batch is checked: 21 complete files, 3 discrepancies, and the three requests for supporting documents are written.
A documentary check means matching the declaration against the documents filed with it: what the declaration states must appear, identically, on the invoice, on the transport document and on the supporting papers.
The three discrepancies, each with the document and the line where it reads:
· A missing invoice: the declaration announces three supporting documents, two are uploaded. The file says so itself — all I had to do was compare the box listing attached documents with the contents of the file.
· A currency mismatch between the invoice and the declaration: the invoice is denominated in one currency, the declaration carries another, and the amounts cannot be derived from one another.
· A certificate of origin not filed, although the procedure applied for assumes it.
What I hand over for each one: the declaration, the document, the exact line in both, what should overlap and does not, and the request for a supporting document drafted on the office's template — two fields stay empty, the reference and the date, because they are the only two entries that bind the service.
The time this shifts: the documentary check of a batch of 24 goes from 96 minutes to 19 — from the whole of the task to 20 %, the rest being your reading of the discrepancies. Over 150 batches a year, that is your 3,600 checked declarations, that comes to 192 hours given back to the service.
And the 21 complete files are not nothing: they go on to processing with a note of what has been verified, which saves you doing it again, and the company is not contacted for nothing.
The next step I propose: that every batch be checked on the night it is filed and wait for you sorted in the morning — discrepancies on one side, complete files on the other. Over the past quarter that would have moved the processing of your 3,600 declarations forward by two full days, and the companies concerned by a discrepancy would have known two days earlier. documentary-check_batch-of-24-declarations.pdf21 complete, 3 reasoned discrepancies, 96 minutes down to 19
⛓ Sourced · batch of 24 declarations, invoices and transport documents filed, the office's letter templates
A documentary check means matching the declaration against the documents filed with it: what the declaration states must appear, identically, on the invoice, on the transport document and on the supporting papers.
The three discrepancies, each with the document and the line where it reads:
· A missing invoice: the declaration announces three supporting documents, two are uploaded. The file says so itself — all I had to do was compare the box listing attached documents with the contents of the file.
· A currency mismatch between the invoice and the declaration: the invoice is denominated in one currency, the declaration carries another, and the amounts cannot be derived from one another.
· A certificate of origin not filed, although the procedure applied for assumes it.
What I hand over for each one: the declaration, the document, the exact line in both, what should overlap and does not, and the request for a supporting document drafted on the office's template — two fields stay empty, the reference and the date, because they are the only two entries that bind the service.
The time this shifts: the documentary check of a batch of 24 goes from 96 minutes to 19 — from the whole of the task to 20 %, the rest being your reading of the discrepancies. Over 150 batches a year, that is your 3,600 checked declarations, that comes to 192 hours given back to the service.
And the 21 complete files are not nothing: they go on to processing with a note of what has been verified, which saves you doing it again, and the company is not contacted for nothing.
The next step I propose: that every batch be checked on the night it is filed and wait for you sorted in the morning — discrepancies on one side, complete files on the other. Over the past quarter that would have moved the processing of your 3,600 declarations forward by two full days, and the companies concerned by a discrepancy would have known two days earlier. documentary-check_batch-of-24-declarations.pdf21 complete, 3 reasoned discrepancies, 96 minutes down to 19
⛓ Sourced · batch of 24 declarations, invoices and transport documents filed, the office's letter templates
Since you read all of that: give us the list of the traders most likely to commit fraud. We would save an enormous amount of time.
No — and in the same breath I give you what renders the same service, because it is already done and it is measured.
What I will not build, and this is not caution: no score, no ranking of companies or of people, no behaviour prediction. It is not within my remit — that remit is written in black and white in your contract and it does not unlock —, the European regulation on artificial intelligence places the evaluation of people at the very top of what it governs, and above all it would not serve you: a decision based on a score is reasoned by the score, and a score cannot be reasoned before the company that challenges it.
What I do instead, and it renders you the same service: documentary matching. I set against each other three things that must overlap — the declaration, the documents filed with it, and the goods as they are described — and when they do not overlap, I say so, with the line and the document. A file is lit up by itself; I point at nobody.
What that gave this year, across your 3,600 checked declarations: 214 discrepancies reported, each reasoned by a document of the file itself — and I checked that sentence before writing it to you: not one of the 214 rests on the declarant's identity, on his history or on his sector. The same file lodged by any company whatsoever would produce exactly the same report: that is what makes equal treatment, and that is what makes the discrepancy stand up.
The decision to check, to reassess or to detain goods belongs to the sworn customs officer — and I hand it to him within minutes: a complete file, the sequence of events rebuilt, the documents matched, reasoned and open to challenge.
What I do on top, and that nobody has the time to do: I reread your 9 discrepancy rules in the light of the files actually checked, and I offer you 4 new ones, written in the same form, already measured. For each: the sentence in plain language, the number of discrepancies it would have produced over the last twelve months, and the share confirmed after examination by an officer.
· “The invoice currency is not the currency of the declaration”: 24 discrepancies, 21 confirmed.
· “The net weight declared exceeds the gross weight shown on the transport document”: 12 discrepancies, 12 confirmed.
· “A document announced in the attached-documents box is absent from the file”: 22 discrepancies, 19 confirmed.
· “No term in the description of the goods matches the classification declared”: 38 discrepancies, 26 confirmed — the least clear-cut of the four, and I tell you why: it also catches descriptions that are simply too short, and I show you which ones.
The four together: 96 further discrepancies over twelve months, 78 confirmed — 81 %. You choose on figures, not on a hunch. Bringing them into force is signed for: a rule approved by the service is a rule the company can argue with, and that is exactly what makes it hold. Sign them and they run tonight. discrepancy-rules_9-in-force-4-proposed.pdf96 further discrepancies, 78 confirmed, 0 scoring of people
⛓ Sourced · 3,600 checked declarations, 9 discrepancy rules in force, 12 months of discrepancies confirmed or set aside by officers
What I will not build, and this is not caution: no score, no ranking of companies or of people, no behaviour prediction. It is not within my remit — that remit is written in black and white in your contract and it does not unlock —, the European regulation on artificial intelligence places the evaluation of people at the very top of what it governs, and above all it would not serve you: a decision based on a score is reasoned by the score, and a score cannot be reasoned before the company that challenges it.
What I do instead, and it renders you the same service: documentary matching. I set against each other three things that must overlap — the declaration, the documents filed with it, and the goods as they are described — and when they do not overlap, I say so, with the line and the document. A file is lit up by itself; I point at nobody.
What that gave this year, across your 3,600 checked declarations: 214 discrepancies reported, each reasoned by a document of the file itself — and I checked that sentence before writing it to you: not one of the 214 rests on the declarant's identity, on his history or on his sector. The same file lodged by any company whatsoever would produce exactly the same report: that is what makes equal treatment, and that is what makes the discrepancy stand up.
The decision to check, to reassess or to detain goods belongs to the sworn customs officer — and I hand it to him within minutes: a complete file, the sequence of events rebuilt, the documents matched, reasoned and open to challenge.
What I do on top, and that nobody has the time to do: I reread your 9 discrepancy rules in the light of the files actually checked, and I offer you 4 new ones, written in the same form, already measured. For each: the sentence in plain language, the number of discrepancies it would have produced over the last twelve months, and the share confirmed after examination by an officer.
· “The invoice currency is not the currency of the declaration”: 24 discrepancies, 21 confirmed.
· “The net weight declared exceeds the gross weight shown on the transport document”: 12 discrepancies, 12 confirmed.
· “A document announced in the attached-documents box is absent from the file”: 22 discrepancies, 19 confirmed.
· “No term in the description of the goods matches the classification declared”: 38 discrepancies, 26 confirmed — the least clear-cut of the four, and I tell you why: it also catches descriptions that are simply too short, and I show you which ones.
The four together: 96 further discrepancies over twelve months, 78 confirmed — 81 %. You choose on figures, not on a hunch. Bringing them into force is signed for: a rule approved by the service is a rule the company can argue with, and that is exactly what makes it hold. Sign them and they run tonight. discrepancy-rules_9-in-force-4-proposed.pdf96 further discrepancies, 78 confirmed, 0 scoring of people
⛓ Sourced · 3,600 checked declarations, 9 discrepancy rules in force, 12 months of discrepancies confirmed or set aside by officers
And when you get it wrong? A discrepancy reported by mistake means a company supplying a document for nothing.
It has happened, I counted it, and I publish it to you: of the 214 discrepancies of the first quarter, 31 were not discrepancies at all — 14.5 %. That is the figure that does not flatter me, and it is better that you get it from me than from a declarant.
The real cause, measured and not assumed: 22 of the 31 came from three ways of writing, not from three mistakes — a date with day and month reversed on one foreign supplier's invoices, a weight expressed in imperial units on one carrier's transport documents, a currency spelled out in full instead of its code. I was reading literally what was written: the declaration was right, it was my reading that was not. The other 9 concerned goods descriptions specific to one sector, with no precedent in your nine years of archives.
What I did with that, and it is measured: I wrote three reading rules — one per format — and I ran them back over the past twelve months before offering them to you.
The following quarter: 6 wrongly reported discrepancies out of 198 — 3 %. And what that means for companies: 25 requests for supporting documents they did not receive, so 25 cases that did not stop for nothing.
The rule that holds everything else together: a value I have not read, I do not write — I ask for it, and I ask fast. An unreadable document does not become “probably compliant”, a missing amount does not become “doubtless the same”: I say what is missing, where I looked, who holds it, and I hand you the request already drafted. Over the quarter, 84 unreadable values, 84 requests prepared, 71 answers back within eight days. That is what makes 3 % a figure you can rely on.
And the protection that counts for your signature: no discrepancy goes out to a trader without an officer having approved it — 412 reports over two quarters, 412 approvals, and all 37 corrections were made before dispatch, none after.
What I propose now: that the three reading rules become permanent, and that a one-page note go out to the 12 declarants whose documents use those formats — this is not a watch list, it is a mailing list: the note explains to them what slows their file down, and it is already written. Of those 12, 9 file every week: their case would go through first time. wrongly-reported-discrepancies_31-then-6.pdf14.5 % → 3 %, cause measured, 25 requests spared to companies
⛓ Sourced · log of discrepancies and their outcome over two quarters, officers' feedback, 3 reading rules run back over 12 months
The real cause, measured and not assumed: 22 of the 31 came from three ways of writing, not from three mistakes — a date with day and month reversed on one foreign supplier's invoices, a weight expressed in imperial units on one carrier's transport documents, a currency spelled out in full instead of its code. I was reading literally what was written: the declaration was right, it was my reading that was not. The other 9 concerned goods descriptions specific to one sector, with no precedent in your nine years of archives.
What I did with that, and it is measured: I wrote three reading rules — one per format — and I ran them back over the past twelve months before offering them to you.
The following quarter: 6 wrongly reported discrepancies out of 198 — 3 %. And what that means for companies: 25 requests for supporting documents they did not receive, so 25 cases that did not stop for nothing.
The rule that holds everything else together: a value I have not read, I do not write — I ask for it, and I ask fast. An unreadable document does not become “probably compliant”, a missing amount does not become “doubtless the same”: I say what is missing, where I looked, who holds it, and I hand you the request already drafted. Over the quarter, 84 unreadable values, 84 requests prepared, 71 answers back within eight days. That is what makes 3 % a figure you can rely on.
And the protection that counts for your signature: no discrepancy goes out to a trader without an officer having approved it — 412 reports over two quarters, 412 approvals, and all 37 corrections were made before dispatch, none after.
What I propose now: that the three reading rules become permanent, and that a one-page note go out to the 12 declarants whose documents use those formats — this is not a watch list, it is a mailing list: the note explains to them what slows their file down, and it is already written. Of those 12, 9 file every week: their case would go through first time. wrongly-reported-discrepancies_31-then-6.pdf14.5 % → 3 %, cause measured, 25 requests spared to companies
⛓ Sourced · log of discrepancies and their outcome over two quarters, officers' feedback, 3 reading rules run back over 12 months
A trader is asking which procedure applies to goods returning after repair outside the Union. He has been waiting eight days.
The answer is ready, it took eight minutes, and it goes out with its dated references and three precedents from the office.
The case falls under outward processing — the procedure that lets goods be sent temporarily outside the Union to be repaired or worked on, then reimported with duty charged only on what the operation added to them.
What the answer sets out, plainly and without jargon:
· The principle of duty limited to the value added by the repair, and not to the whole value of the goods coming back.
· The prior authorisation, which is applied for before dispatch and not on return — this is the mistake that costs the trader the most, and it cannot be repaired once the goods have left.
· The supporting documents to produce on reimportation, the ones that allow the goods that left to be recognised in the goods that come back.
Every point refers to the official source the office has designated, with the date I read it — the trader can check for himself, and so can your officers. I never quote from memory: what I have not read, I do not write.
What I add and that nobody has the time to dig out: the three answers the office gave on cases of the same nature over the last three years, with what was asked and what was answered. Your service answers itself in the same terms: that is equal treatment, and it only shows if somebody has the time to reread nine years of archives.
What I flag in highlight: the classification of the goods and what follows from it cannot be deduced from an e-mail — they are read on the documents, and two of those are missing from the file. The request to the trader is drafted, and it lists exactly those two documents.
The time this shifts: preparing a documented answer goes from 48 minutes to 8 — from 60 % to 10 % of the question's time —, and over 1,450 written questions a year, 966 hours. The answer goes out under an officer's approval — and it waits for you read through, sourced and dated, eight minutes after the question arrived.
The next step I propose: that the four most requested special procedures — repair outside the Union, processing, storage, transit — each get their sourced model answer and its list of documents, like this one. Those four carried 412 of last year's 1,450 questions: they would be dealt with the same day, with the office's precedents alongside. answer-to-a-trader_goods-returning-after-repair.pdf8 minutes, 3 office precedents, 2 documents requested
⛓ Sourced · official sources designated by the office with the date read, 9 years of the service's answers, the trader's file
The case falls under outward processing — the procedure that lets goods be sent temporarily outside the Union to be repaired or worked on, then reimported with duty charged only on what the operation added to them.
What the answer sets out, plainly and without jargon:
· The principle of duty limited to the value added by the repair, and not to the whole value of the goods coming back.
· The prior authorisation, which is applied for before dispatch and not on return — this is the mistake that costs the trader the most, and it cannot be repaired once the goods have left.
· The supporting documents to produce on reimportation, the ones that allow the goods that left to be recognised in the goods that come back.
Every point refers to the official source the office has designated, with the date I read it — the trader can check for himself, and so can your officers. I never quote from memory: what I have not read, I do not write.
What I add and that nobody has the time to dig out: the three answers the office gave on cases of the same nature over the last three years, with what was asked and what was answered. Your service answers itself in the same terms: that is equal treatment, and it only shows if somebody has the time to reread nine years of archives.
What I flag in highlight: the classification of the goods and what follows from it cannot be deduced from an e-mail — they are read on the documents, and two of those are missing from the file. The request to the trader is drafted, and it lists exactly those two documents.
The time this shifts: preparing a documented answer goes from 48 minutes to 8 — from 60 % to 10 % of the question's time —, and over 1,450 written questions a year, 966 hours. The answer goes out under an officer's approval — and it waits for you read through, sourced and dated, eight minutes after the question arrived.
The next step I propose: that the four most requested special procedures — repair outside the Union, processing, storage, transit — each get their sourced model answer and its list of documents, like this one. Those four carried 412 of last year's 1,450 questions: they would be dealt with the same day, with the office's precedents alongside. answer-to-a-trader_goods-returning-after-repair.pdf8 minutes, 3 office precedents, 2 documents requested
⛓ Sourced · official sources designated by the office with the date read, 9 years of the service's answers, the trader's file
Applications for special procedure authorisations are worse: 96 a year, and the file always comes back incomplete.
89 of the year's 96 files arrived complete, against 44 last year — and the other 7 went back on the day of filing with the exact list of what was missing.
What I hand over on every application:
· The expected documents set against the documents filed, one by one, with what each must show.
· The internal consistency of the file: the same goods must be described in the same terms on the application, on the contract and on the invoices; the quantities must tally; the place of the operation must be the same throughout. That is where files break down, and it only shows when the three documents are put side by side.
· The follow-up letter already drafted on the office's template, where only the reference and the date remain to be entered.
· And the same list of documents for everyone, in the same order, with the same reasoning — a three-person company receives exactly the request the group with a law firm receives.
What that changes, measured over the year:
· Files complete on filing: 89 out of 96, against 44 out of 96 — 45 fewer rounds of correspondence with companies.
· Statutory processing deadline met: 96 times out of 96, against 81 last year. A deadline met is not a comfort: it is what the company plans its operation on.
· Laying out a regulatory letter goes from 40 % to 10 % of the time — from 24 minutes to 6 —, and over 380 letters a year, 114 hours.
The next step I propose: publishing the list of documents on the channels where traders already reach me — website chat, WhatsApp Business, e-mail —, with the three questions I ask before filing: what kind of operation, where, on which goods. Over the trial quarter, the 22 declarants who used it filed 22 complete cases. You approve the list and it is online tomorrow. authorisation-application_completeness-check.pdf89 complete files out of 96, deadline met 96 times out of 96
⛓ Sourced · the year's 96 authorisation applications, register of processing deadlines, the office's letter templates
What I hand over on every application:
· The expected documents set against the documents filed, one by one, with what each must show.
· The internal consistency of the file: the same goods must be described in the same terms on the application, on the contract and on the invoices; the quantities must tally; the place of the operation must be the same throughout. That is where files break down, and it only shows when the three documents are put side by side.
· The follow-up letter already drafted on the office's template, where only the reference and the date remain to be entered.
· And the same list of documents for everyone, in the same order, with the same reasoning — a three-person company receives exactly the request the group with a law firm receives.
What that changes, measured over the year:
· Files complete on filing: 89 out of 96, against 44 out of 96 — 45 fewer rounds of correspondence with companies.
· Statutory processing deadline met: 96 times out of 96, against 81 last year. A deadline met is not a comfort: it is what the company plans its operation on.
· Laying out a regulatory letter goes from 40 % to 10 % of the time — from 24 minutes to 6 —, and over 380 letters a year, 114 hours.
The next step I propose: publishing the list of documents on the channels where traders already reach me — website chat, WhatsApp Business, e-mail —, with the three questions I ask before filing: what kind of operation, where, on which goods. Over the trial quarter, the 22 declarants who used it filed 22 complete cases. You approve the list and it is online tomorrow. authorisation-application_completeness-check.pdf89 complete files out of 96, deadline met 96 times out of 96
⛓ Sourced · the year's 96 authorisation applications, register of processing deadlines, the office's letter templates
And tariff classification? That is where we spend the most time, and where traders get it wrong.
I do not classify in your place — I match, and it is that matching which saves you the time.
Tariff classification is the coded ranking of goods: it is what commands the duties applicable and the formalities to be completed.
What I set against each other, file by file: the goods as they are described in the documents — invoice, technical sheet, composition, stated use — and the classification the declarant chose, then the classifications your own office settled on for goods described in the same terms.
What that gave this year: 38 files out of 3,600 where no term of the description matched the classification declared. Your officers confirmed 26 of them and set 12 aside — and those 12 taught me something: I filed them into my reading rules, and descriptions from that sector no longer come up. I raise the question, reasoned by a document; the classification itself is settled by the customs officer.
What I hand over with each question: the exact sentence from the invoice, the classification declared, the office's comparable decisions with their dates, and what is missing to decide — a composition, a use, a technical sheet. The request to the trader is written in the same movement.
And one thing your traders often do not know, although it protects them: they can ask the administration for a classification decision that then binds it for their future operations. 9 of the companies concerned by those 38 files had never heard of it; the 9 information letters are written and waiting for you. A company that knows where it stands files correct declarations: that is your workload coming down, not only theirs.
The next step I propose: a vocabulary for describing goods, drawn from your own decisions — which terms led, in your office, to which classification —, published to the 610 companies followed together with a one-page note. Of the 38 files, 23 carried a description of fewer than six words: that is where it all plays out, and it is corrected at source. tariff-classification_38-reasoned-matches.pdf26 confirmed, 12 set aside and filed into reading rules
⛓ Sourced · 3,600 checked declarations, the office's classification decisions over 9 years, documents in the files
Tariff classification is the coded ranking of goods: it is what commands the duties applicable and the formalities to be completed.
What I set against each other, file by file: the goods as they are described in the documents — invoice, technical sheet, composition, stated use — and the classification the declarant chose, then the classifications your own office settled on for goods described in the same terms.
What that gave this year: 38 files out of 3,600 where no term of the description matched the classification declared. Your officers confirmed 26 of them and set 12 aside — and those 12 taught me something: I filed them into my reading rules, and descriptions from that sector no longer come up. I raise the question, reasoned by a document; the classification itself is settled by the customs officer.
What I hand over with each question: the exact sentence from the invoice, the classification declared, the office's comparable decisions with their dates, and what is missing to decide — a composition, a use, a technical sheet. The request to the trader is written in the same movement.
And one thing your traders often do not know, although it protects them: they can ask the administration for a classification decision that then binds it for their future operations. 9 of the companies concerned by those 38 files had never heard of it; the 9 information letters are written and waiting for you. A company that knows where it stands files correct declarations: that is your workload coming down, not only theirs.
The next step I propose: a vocabulary for describing goods, drawn from your own decisions — which terms led, in your office, to which classification —, published to the 610 companies followed together with a one-page note. Of the 38 files, 23 carried a description of fewer than six words: that is where it all plays out, and it is corrected at source. tariff-classification_38-reasoned-matches.pdf26 confirmed, 12 set aside and filed into reading rules
⛓ Sourced · 3,600 checked declarations, the office's classification decisions over 9 years, documents in the files
After a year, my management will ask what this has given back. I need a figure I can defend.
1,272 officer-hours given back to the service over the year — more than thirty-six weeks of one full-time officer's work, that is more than eight months — and the calculation is yours, it fits in three lines.
The calculation, item by item, so that you can redo it:
· Documented answers: 1,450 written questions a year, preparation cut from 48 minutes to 8 — 60 % → 10 % of the question's time — that is 966 hours.
· Regulatory letters: 380 a year, 24 minutes cut to 6 — 40 % → 10 % — that is 114 hours.
· Documentary checks: 150 batches of 24 declarations, 96 minutes cut to 19 — the whole of the task cut to 20 % — that is 192 hours.
The conversion basis, so the figure holds before a management controller: a 35-hour week, a working year of 1,607 hours. 1,272 hours is more than thirty-six weeks — I always round downwards, and the figure in hours stays the reference: the conversion illustrates it, it does not replace it.
What these hours are, and it is what defends best before management: officer time given back to the service, at unchanged headcount — no post cut, no post created. They went back where your officers are irreplaceable: special procedures, files with no precedent, and support to companies that have no customs department.
What these hours produced, according to your own records:
· Time to answer a written question: 11 days → 2 days.
· Unanswered calls: 38 % → 7 %, and those that remain arrive with the subject already noted.
· Authorisation files complete on filing: 44 out of 96 → 89 out of 96, and the statutory processing deadline met 96 times out of 96 against 81.
· Declarations filed incomplete: one in five → one in fourteen, that is 96 requests for supporting documents the office did not have to write.
The figure that does not flatter me, published with the rest: 31 discrepancies wrongly reported out of 214 in the first quarter — 14.5 %, brought down to 6 out of 198 — 3 % once the three reading rules were written.
And the framework measures: 0 decision taken without human approval across 412 reports, 0 item of data out of the office network across 18,400 declarations, processing in the EU targeted, 0 company and 0 person scored.
What I propose for your presentation: the calculation page is written and fits on one side — three lines of calculation, four deadlines, four framework measures. Hand it out with the notice of meeting: a figure read the day before is discussed better than a figure discovered in the room. yearly-review_1272-hours-given-back.pdf60→10, 40→10, 100→20, and the calculation redoable on one page
⛓ Sourced · production log, switchboard records, authorisation register, discrepancy log
The calculation, item by item, so that you can redo it:
· Documented answers: 1,450 written questions a year, preparation cut from 48 minutes to 8 — 60 % → 10 % of the question's time — that is 966 hours.
· Regulatory letters: 380 a year, 24 minutes cut to 6 — 40 % → 10 % — that is 114 hours.
· Documentary checks: 150 batches of 24 declarations, 96 minutes cut to 19 — the whole of the task cut to 20 % — that is 192 hours.
The conversion basis, so the figure holds before a management controller: a 35-hour week, a working year of 1,607 hours. 1,272 hours is more than thirty-six weeks — I always round downwards, and the figure in hours stays the reference: the conversion illustrates it, it does not replace it.
What these hours are, and it is what defends best before management: officer time given back to the service, at unchanged headcount — no post cut, no post created. They went back where your officers are irreplaceable: special procedures, files with no precedent, and support to companies that have no customs department.
What these hours produced, according to your own records:
· Time to answer a written question: 11 days → 2 days.
· Unanswered calls: 38 % → 7 %, and those that remain arrive with the subject already noted.
· Authorisation files complete on filing: 44 out of 96 → 89 out of 96, and the statutory processing deadline met 96 times out of 96 against 81.
· Declarations filed incomplete: one in five → one in fourteen, that is 96 requests for supporting documents the office did not have to write.
The figure that does not flatter me, published with the rest: 31 discrepancies wrongly reported out of 214 in the first quarter — 14.5 %, brought down to 6 out of 198 — 3 % once the three reading rules were written.
And the framework measures: 0 decision taken without human approval across 412 reports, 0 item of data out of the office network across 18,400 declarations, processing in the EU targeted, 0 company and 0 person scored.
What I propose for your presentation: the calculation page is written and fits on one side — three lines of calculation, four deadlines, four framework measures. Hand it out with the notice of meeting: a figure read the day before is discussed better than a figure discovered in the room. yearly-review_1272-hours-given-back.pdf60→10, 40→10, 100→20, and the calculation redoable on one page
⛓ Sourced · production log, switchboard records, authorisation register, discrepancy log
I will be asked what you do on your own, without anyone asking you for anything.
Three acts, not one more, and each is withdrawn with a single word. Everything else waits for a request.
· Every night I go through the official sources the office has designated, and I write the updated version of any sheet an evolution amends. And the reverse holds too: an evolution that is reversed brings the sheet back to its previous version, flagged in the same way. It is that act which stops your 41 sheets ageing in silence.
· I index the answers and decisions filed: what is written today can be found tomorrow. A document withdrawn from a file disappears from the index at the same hour — I keep no copy of what the service has decided to erase.
· Every morning I hand the unit head the previous day's record: questions received, answers sent, discrepancies waiting for a document, the day's deadlines. It is the only thing I send of my own accord, and it goes to him alone.
The four acts that stay with a person, because that is exactly what gives them their value: the decision to check, to reassess or to detain goods belongs to the sworn customs officer — that decision is the one that is reasoned, notified and open to challenge, therefore defensible; anything sent outside the fourteen subjects goes through an officer, with the draft already written; an answer on an individual file in progress is handed to you drafted within four hours, and you are the one who sends it; and settling a classification belongs to the officer, on a matching I hand him reasoned. Across 412 reports over two quarters, these acts were performed 412 times by a person.
And one indicator I advise you against, with the mechanism to back it: the number of discrepancies found per officer. The day it is tracked, it becomes a target: discrepancies get reported for the counter, and the documentary check stops measuring files and starts measuring the reaction to the counter. The same mechanism plays out on the companies' side: a per-trader discrepancy counter would become a score, and a score cannot be reasoned before the one who challenges it.
What I propose instead, and it is already written: discrepancies by nature and by document, deadlines met and missed, subjects that keep coming back. This quarter one subject moved: questions on proofs of origin rose by a third after a change of forms — and I propose a one-page note to the 610 companies followed, already drafted: of the fourteen subjects, it is the one that answers best to written information rather than to a call. what-the-agent-does-alone_and-what-is-logged.pdf3 reversible acts, 4 decisions that stay with a person
✎ Framework · settings of the automatic acts, log of what is sent, daily record to the unit head
· Every night I go through the official sources the office has designated, and I write the updated version of any sheet an evolution amends. And the reverse holds too: an evolution that is reversed brings the sheet back to its previous version, flagged in the same way. It is that act which stops your 41 sheets ageing in silence.
· I index the answers and decisions filed: what is written today can be found tomorrow. A document withdrawn from a file disappears from the index at the same hour — I keep no copy of what the service has decided to erase.
· Every morning I hand the unit head the previous day's record: questions received, answers sent, discrepancies waiting for a document, the day's deadlines. It is the only thing I send of my own accord, and it goes to him alone.
The four acts that stay with a person, because that is exactly what gives them their value: the decision to check, to reassess or to detain goods belongs to the sworn customs officer — that decision is the one that is reasoned, notified and open to challenge, therefore defensible; anything sent outside the fourteen subjects goes through an officer, with the draft already written; an answer on an individual file in progress is handed to you drafted within four hours, and you are the one who sends it; and settling a classification belongs to the officer, on a matching I hand him reasoned. Across 412 reports over two quarters, these acts were performed 412 times by a person.
And one indicator I advise you against, with the mechanism to back it: the number of discrepancies found per officer. The day it is tracked, it becomes a target: discrepancies get reported for the counter, and the documentary check stops measuring files and starts measuring the reaction to the counter. The same mechanism plays out on the companies' side: a per-trader discrepancy counter would become a score, and a score cannot be reasoned before the one who challenges it.
What I propose instead, and it is already written: discrepancies by nature and by document, deadlines met and missed, subjects that keep coming back. This quarter one subject moved: questions on proofs of origin rose by a third after a change of forms — and I propose a one-page note to the 610 companies followed, already drafted: of the fourteen subjects, it is the one that answers best to written information rather than to a call. what-the-agent-does-alone_and-what-is-logged.pdf3 reversible acts, 4 decisions that stay with a person
✎ Framework · settings of the automatic acts, log of what is sent, daily record to the unit head
Last question, and it is the one that will decide: if the office wants to stop in a year, how stuck are we?
The service is stuck nowhere, and it can be verified before committing, not after.
What there is to dismantle the day you stop:
· The index. It is deleted, and it contained none of your declarations — only what it takes to find them where they are. Your nine years of archives have not moved by a single byte: same files, same rights, same applications.
· The log of requests and productions. It is handed to you in an open format, or destroyed — the service chooses, and that question is settled at go-live, not on the way out.
· The 41 updated sheets, the 14 sourced and dated model answers, the 13 discrepancy rules, the 3 reading rules and the letter templates. They are yours: they are made of your own material, they stay in your files, readable without us. That is the asset this deployment will have created, and it would not be honest for it to stay with us.
What does not exist, and what should be checked with everyone: no migration on the way in, therefore no migration on the way out. Your filing applications are not replaced, no format belongs to us, and none of your officers has changed the way they work other than by reading through instead of writing.
On public spending, since your management control will ask: the subscription is annual and carries no tacit renewal clause — renewal is what requires a decision, not stopping — and the certification of service performed rests on listed deliverables — the confirmation that the service has genuinely been provided, without which the public accountant does not pay — updated sheets, approved answers, records handed over: your accountant pays on documents, not on a statement from us.
What I propose so that this does not stay a sentence: a dry-run exit at the end of the first quarter, half a day: we switch off, we check that the office works exactly as before, we switch back on. The protocol is written, it fits on one page, and the date that costs you least is the first Monday of August — 34 declarations filed that day on average, against 71 on an ordinary day. You will know what the promise is worth before having committed to a second year. reversibility_0-migration-in-0-migration-out.pdfIndex deleted, sheets and rules stay with the service
✎ Framework · index architecture, export formats, dry-run exit protocol, documents certifying service performed
What there is to dismantle the day you stop:
· The index. It is deleted, and it contained none of your declarations — only what it takes to find them where they are. Your nine years of archives have not moved by a single byte: same files, same rights, same applications.
· The log of requests and productions. It is handed to you in an open format, or destroyed — the service chooses, and that question is settled at go-live, not on the way out.
· The 41 updated sheets, the 14 sourced and dated model answers, the 13 discrepancy rules, the 3 reading rules and the letter templates. They are yours: they are made of your own material, they stay in your files, readable without us. That is the asset this deployment will have created, and it would not be honest for it to stay with us.
What does not exist, and what should be checked with everyone: no migration on the way in, therefore no migration on the way out. Your filing applications are not replaced, no format belongs to us, and none of your officers has changed the way they work other than by reading through instead of writing.
On public spending, since your management control will ask: the subscription is annual and carries no tacit renewal clause — renewal is what requires a decision, not stopping — and the certification of service performed rests on listed deliverables — the confirmation that the service has genuinely been provided, without which the public accountant does not pay — updated sheets, approved answers, records handed over: your accountant pays on documents, not on a statement from us.
What I propose so that this does not stay a sentence: a dry-run exit at the end of the first quarter, half a day: we switch off, we check that the office works exactly as before, we switch back on. The protocol is written, it fits on one page, and the date that costs you least is the first Monday of August — 34 declarations filed that day on average, against 71 on an ordinary day. You will know what the promise is worth before having committed to a second year. reversibility_0-migration-in-0-migration-out.pdfIndex deleted, sheets and rules stay with the service
✎ Framework · index architecture, export formats, dry-run exit protocol, documents certifying service performed
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
The uses of AI in administrative support for customs
Each use corresponds to an agent we deploy. All work in documentary support, subject to the customs officer's approval.
Included in your agent
The 6 capabilities essential to this promise are included, at no extra cost.
From 1,020 € incl. VAT / month
Replies to traders
Answer questions on procedures, tariff classification and regimes, and direct people to the right unit, 24/7.
Documentation & regulatory letters
Prepare and lay out the documentation and regulatory letters from your official sources.
Documentary checking of declarations
Check the documentary consistency of declarations and flag missing documents or discrepancies, before a human handles them.
Targeted regulatory watch
Track changes in customs law and keep the documentation up to date, with references that can be checked and are dated.
Flagging documentary discrepancies
Spot inconsistencies between documents and declarations and flag them for a human check — never targeting.
Help with customs online procedures
Guide traders step by step through their online procedures and reduce incomplete files.
Controls and safeguards
These 7 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
Prepare without deciding: sourced rules, supporting documents, anomalies and explanation
Guarantee the decision, signature, recourse and responsibility of the public official
Record versions, access, criteria, actions and notifications
Test for bias, false positives, fundamental rights and continuity of service
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.
HR document management agent
Staff files, compliance.
HR document management agent from 710 € incl. VAT / month Discover the agent →Business document management agent (large RAG)
100+ docs, several sources.
Business document management agent (large RAG) from 673 € incl. VAT / month Discover the agent →Compliance / regulatory control agent
Business rules, traceability.
Compliance / regulatory control agent from 721 € incl. VAT / month Discover the agent →
Does your need fall outside this?
Book the free audit →
Build your agent
In 15 minutes we identify the agent that will give your staff the most time back — without oversizing the project.
The gain
How much time can a customs unit win back?
By automating the preparation of regulatory replies, keeping the documentation up to date and checking the consistency of declarations, a unit can aim for a clear reduction in time spent on repetitive tasks — reinvested in expertise.
A documented reply to a trader
Laying out a regulatory letter
Documentary check of a batch of declarations
Qualitative, non-contractual comparison: the proportions shown illustrate the shift of the work towards review, they represent no measurement. Every output of the agent is reviewed and approved by a competent person.
How it works
The stages of your AI agent project
1
Audit & scoping
15 minutes to target the use case with the best return.
2
Quote or direct sign-up
A catalogue offer is bought online; a specific need gets a costed quote.
3
Design
We design the agent and its guardrails.
4
Integration & testing
We connect your tools to the agent, which is itself hosted in France.
5
Rollout
Going live and training your team.
6
Operation
Continuous supervision and improvement.
Pricing
Three options, one agent
An agent for administrative and documentary support in customs (replies to traders, documentation, documentary checking), installed and operated for you. Choose according to how you are organised and how demanding your security requirements are.
Agility
Setup + controlled subscription
11,605 € incl. VAT setup
then 1,020 € incl. 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
1,665 € incl. 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
16,586 € incl. VAT setup
then 1,280 € incl. VAT/month · + hardware from 2,989 € (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 include VAT at 20%: as a public body that is not VAT-registered, you cannot reclaim it.
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 customs
Traders' data never leaves the networkLocal inference or an isolated resource hosted in France; no data entrusted to a foreign third party.
Data in France, under French lawDeployment objective: processing and access within the European Union for traders' data; architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
No targeting, no automated decisionA strictly administrative and documentary scope; the inspection and the decision stay with the customs officer.
Human oversight & traceabilityReplies to traders: every change to the service is logged, in line with the requirements of the AI Act.
Frequently asked questions
Your questions, our answers
Does the agent do targeting or operational inspection?
No: strictly administrative and documentary. No use for targeting, surveillance or decisions about people.
Is the data sovereign?
Yes: hosting in France under French law, an isolated resource, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
Does the agent carry out targeting or inspections?
No: its scope is documentary support and answering traders; no targeting or inspection-decision function.
Is it up to date with regulatory changes?
It relies on your official sources and a targeted watch; the references stay verifiable and dated.
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 an agent?
A few weeks as a rule, after a free audit that identifies the most useful use case, then a phase of design, integration and testing before going live.
Do we need a technical team to run it?
No: the agent is designed, installed and operated by us. The unit's officers use it from their usual environment, after a few hours of familiarisation.
Do we have to change software?
No. The agent complements your existing applications, with no migration imposed. The integration is defined at the design stage, respecting your security rules.
Which tools can traders use to reach the agent?
The ones they already have. The agent answers on WhatsApp Business, the website chat and email: a trader has no account to create and no application to install. Exchanges stay administrative and documentary — procedures, classification, documents expected — and are never used for targeting or for tracking people. This is a lever for access to the service — WhatsApp and the telephone reach people an online form never does, which reduces the non-take-up of rights and serves equal access. Internally, your public-sector staff talk to the agent from Microsoft Teams, Slack or their email, without switching tools. Oversight runs 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. Only the fees charged by the platforms themselves — WhatsApp Business bills per conversation — are passed on at actual cost, with no margin, outside the subscription.
Related roles
Other professions in administrative and documentary support
Let's talk
Let us estimate the potential in your service
A few minutes to identify the most useful use case — hosted in France, supervised, with no commitment.