● B2B offer — Finance & administration
Expense claims: receipts read, policy applied
An expense claim is a photographed receipt, an amount, a date, a category and an internal rule to check. Your agent reads the receipt, extracts the elements, proposes the category, applies your reimbursement policy and flags the points that call for a check. Hosted in France: your staff's spending stays with you. The accounts team approves every reimbursement.
Hosted in France
Staff spending protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Expense claims · hosted in France
Here are the month's receipts. Prepare the expense claims.
Receipts read: date, amount, VAT, supplier and proposed category for each.
Your policy's caps are applied by category, with the amount allowed and the amount claimed.
Four lines call for a check: they are grouped at the top.
🔗 Sourced · elements extracted from the receipts
Your policy's caps are applied by category, with the amount allowed and the amount claimed.
Four lines call for a check: they are grouped at the top.
🔗 Sourced · elements extracted from the receipts
Why those four lines?
One receipt that is hard to read, two amounts over the cap and one ambiguous category under your policy.
Each is presented with the receipt on screen and the rule concerned. The decision to reimburse is yours: it concerns money.
✎ Support · checks set out, human approval
Each is presented with the receipt on screen and the rule concerned. The decision to reimburse is yours: it concerns money.
✎ Support · checks set out, human approval
Local inference · no data outside the EU
Receipts 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 accounts team decidesThe agent prepares, never rules
✦ In brief
A Blue Lemon Agent expense claim agent reads your receipts, extracts date, amount, VAT and supplier, proposes a category and applies your reimbursement caps. It groups at the top the lines that call for a check — a receipt that is hard to read, an amount over the cap, an ambiguous category — without ever approving a reimbursement. It runs on local inference or is hosted in France: your staff's spending stays with you, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
6
administrative uses ready to deploy
0
decision taken without human approval
These figures describe our offer, not results measured at a client. How large the gain is on the volume of claims handled each month is confirmed by a pilot.
The context
What does an AI agent bring to your expense claims?
Keying in an expense claim is mechanical; checking it is not. Separating the two makes the process faster without letting anything slip.
! The issue
Every line calls for reading the receipt and comparing it with your policy. The first is data entry, the second is control. By taking on the data entry and presenting the exceptions with the rule concerned, the agent leaves the accounts team the essential part: deciding what gets reimbursed.
✓ Our answer
The claims arrive prepared, categorised, with the caps already applied and the points to check grouped at the top. No reimbursement is approved without a human act — it is money paid to an employee. Local inference or an isolated resource hosted in France: the spending your teams incur, often revealing of what they are working on, does not leave the company.
The decisive point
The spending your staff incur: sovereignty & compliance
Your staff's spending is personal data and information about your activity. Here is how the architecture of our agents protects it.
Local inference
The agent can run on a machine belonging to your organisation: no receipt and no spending data leaves the network.
Hosting in France
Otherwise, a dedicated and isolated resource hosted in France, under French law — your expense claims and your receipts: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
For the spending your staff incur, the architecture aims to reduce exposure to the Cloud Act and FISA 702; being located in France or in the European Union does not, on its own, guarantee immunity.
Isolated resource
No pooling: an environment strictly dedicated to your company and its reimbursement policy.
Checks grouped before approval
A receipt that is hard to read, an amount over the cap and an ambiguous category are presented with the rule concerned; encryption, role-based access and logging of every claim prepared.
AI Act: governed deployment
The agent is strictly in support; no reimbursement is approved and no payment is triggered automatically; traceability and human oversight from end to end.
What depends on the architecture chosen These points are not general guarantees: they are settled deployment by deployment, in the quotation.
- The applicable location is that of the architecture set out in the quotation and verified before commissioning.
- Local execution is announced only for the configuration explicitly described and accepted in the quotation.
- The applicable isolation depends on the deployment mode set out in the quotation; no dedicated isolation is presumed.
- Roles and permissions are configured and accepted for the identities and systems actually connected.
- The events logged, their content, their retention period and who may access them are defined for the deployment chosen.
For directors' expense claims and spending tied to a sensitive matter, SecNumCloud and reinforced hosting are options depending on your requirements. A single architecture is designed to answer both the GDPR and extraterritorial exposure. Designed for deployment in line with the GDPR and the AI Act, after the processing, roles and context-specific risks have been assessed.
Demonstration
See the agent at work
4 real situations, taken from those that come up most often. Pick one: the exchange unfolds as it would in your organisation.
A scripted demonstration. These exchanges show how the agent behaves — its sources, its refusals, what it leaves to your teams. Nothing is sent from this page, no model is queried here, and the matters named are fictional. That is precisely what we promise your data.
The behaviours shown here — monitoring, automation rules, routing and reminders — are configured with you during deployment, from your tools, your rules and your thresholds.
The architecture points named in these exchanges — location, local execution, isolation, encryption, role-based access, logging — are not a guarantee attached to the demonstration: they are those of the architecture set out in your quotation, and verified before commissioning.
Expense claims · hosted in France
Four flags this morning on expense claims.
· Thirty-four claims have awaited approval for more than three weeks, eleven of them filed before 15 July. Employees are out of pocket.
· A receipt is illegible on 22 claims this month, all photographed from the same screen of the app.
· One claim carries two lunches on the same day — one at €18, one at €21, at two different addresses.
· Your mileage rates have not been updated since January 2024. They are applied to 412 journeys this year. morning-watch_4-flags.pdf4 flags · 34 unreimbursed advances
⛓ Source · claims filed, receipts, expense policy, rates applied
· Thirty-four claims have awaited approval for more than three weeks, eleven of them filed before 15 July. Employees are out of pocket.
· A receipt is illegible on 22 claims this month, all photographed from the same screen of the app.
· One claim carries two lunches on the same day — one at €18, one at €21, at two different addresses.
· Your mileage rates have not been updated since January 2024. They are applied to 412 journeys this year. morning-watch_4-flags.pdf4 flags · 34 unreimbursed advances
⛓ Source · claims filed, receipts, expense policy, rates applied
Two lunches on the same day — is that fraud?
I have no idea, and I will not put it that way.
What I record: two restaurant receipts, same date, 12.40 and 13.05, two establishments 400 metres apart. Amounts: €18 and €21.
The possible explanations, the most mundane first: a meal paid for a colleague or a client and not identified as such; a filing error — the same receipt photographed twice at neighbouring premises; a lunch cut short and resumed elsewhere; or an improper claim.
What I do: I put the question to the employee, before anything goes to the approver, in these terms: "two restaurant receipts appear on 14/07; was this a meal with a third party? If so, please say who — the policy provides for it."
Why that order: nine times out of ten the answer is immediate and the matter closes without anyone being implicated. Going to the approver first turns a thirty-second question into a file.
The words I use, and it is not a matter of style: the message speaks of "two receipts" and a "meal with a third party" — not "anomaly", not "suspicious", not "to be checked". Of the 61 questions put that way this half-year, 56 were answered the same day and only 3 went up to the approver. A question that accuses nobody gets an answer; a question that accuses gets a justification, and a file. 2-lunches_same-day.pdf4 possible explanations · question put to the employee
⛓ Source · 2 receipts of 14/07, expense policy art. 4, employee's history
What I record: two restaurant receipts, same date, 12.40 and 13.05, two establishments 400 metres apart. Amounts: €18 and €21.
The possible explanations, the most mundane first: a meal paid for a colleague or a client and not identified as such; a filing error — the same receipt photographed twice at neighbouring premises; a lunch cut short and resumed elsewhere; or an improper claim.
What I do: I put the question to the employee, before anything goes to the approver, in these terms: "two restaurant receipts appear on 14/07; was this a meal with a third party? If so, please say who — the policy provides for it."
Why that order: nine times out of ten the answer is immediate and the matter closes without anyone being implicated. Going to the approver first turns a thirty-second question into a file.
The words I use, and it is not a matter of style: the message speaks of "two receipts" and a "meal with a third party" — not "anomaly", not "suspicious", not "to be checked". Of the 61 questions put that way this half-year, 56 were answered the same day and only 3 went up to the approver. A question that accuses nobody gets an answer; a question that accuses gets a justification, and a file. 2-lunches_same-day.pdf4 possible explanations · question put to the employee
⛓ Source · 2 receipts of 14/07, expense policy art. 4, employee's history
How did you spot all that? And who do you tell?
I watch, continuously, what you have opened to me: claims filed, receipts, your expense policy, your rates, and approval times.
Routing follows who can act: a pending claim goes to the approver, and only to them, from 15 days on — an employee out of pocket by €400 should not have to chase three times; an illegible receipt to the employee, within 24 h, while they still have the paper; a question on a line to the employee first, never to the approver first; out-of-date rates to finance.
With a chase: 48 h on a claim pending more than 30 days, 7 days on the rest. Then a monthly summary: by type of gap and by expense category, never by employee.
This morning, 4 flags and money moving again: 34 claims stuck for more than three weeks, 11 of them filed before 15 July — your employees are funding the company's cash flow. From tomorrow: the 22 illegible receipts picked up within 24 h, while the paper still exists, a mileage rate brought up to date across 412 trips already paid at the old one, and a reimbursement time counted in days rather than weeks. The access is yours: opened by role, logged, withdrawn with a word — your claims and receipts do not leave your walls. Approval stays with the approver, and I hand it back in minutes: every claim arrives read, categorised, matched against your policy, the question to the employee already asked and already answered. And the day you want it, it stops being a question and becomes a mandate: written, capped at the amount you set, dated, withdrawn with a word — compliant claims go to reimbursement in minutes, the rest keep coming up to you. The next step is ready: the 34 pending claims are sorted by age and by amount advanced; tell me which ones go out today.
✎ Framework · no claim refused, no ranking of employees
Routing follows who can act: a pending claim goes to the approver, and only to them, from 15 days on — an employee out of pocket by €400 should not have to chase three times; an illegible receipt to the employee, within 24 h, while they still have the paper; a question on a line to the employee first, never to the approver first; out-of-date rates to finance.
With a chase: 48 h on a claim pending more than 30 days, 7 days on the rest. Then a monthly summary: by type of gap and by expense category, never by employee.
This morning, 4 flags and money moving again: 34 claims stuck for more than three weeks, 11 of them filed before 15 July — your employees are funding the company's cash flow. From tomorrow: the 22 illegible receipts picked up within 24 h, while the paper still exists, a mileage rate brought up to date across 412 trips already paid at the old one, and a reimbursement time counted in days rather than weeks. The access is yours: opened by role, logged, withdrawn with a word — your claims and receipts do not leave your walls. Approval stays with the approver, and I hand it back in minutes: every claim arrives read, categorised, matched against your policy, the question to the employee already asked and already answered. And the day you want it, it stops being a question and becomes a mandate: written, capped at the amount you set, dated, withdrawn with a word — compliant claims go to reimbursement in minutes, the rest keep coming up to you. The next step is ready: the 34 pending claims are sorted by age and by amount advanced; tell me which ones go out today.
✎ Framework · no claim refused, no ranking of employees
The 34 pending claims — is that a problem?
It is money your employees have advanced, and nobody is counting it.
What it represents: €19,400 advanced, €7,800 of it for more than thirty days. Eleven claims were filed before 15 July.
What I looked at: why they are not approved. 26 of the 34 are waiting on the same person — a single approver for three teams, away for two weeks in July with no deputy appointed.
So it is not a matter of diligence: it is a single point of passage, and it was absent.
What I propose: the 26 grouped into one list with their receipts, approvable in a single session, and going forward a deputy appointed beyond five days' absence. That is an organisational setting, not an instruction.
What I have costed, should the question come back: an automatic approval threshold at €50 would cover 61 of the 137 claims in the queue, for €4,200. That is an internal control decision, and it is yours — it is taken for what it is, not to make up two weeks of absence. The deputy settles the calendar without touching the control.
The figure that speaks loudest: the average reimbursement time is 41 days; your policy states 15 days. 34-claims_19400-eur-advanced.pdf26 on one approver · 41 days against 15 stated
⛓ Source · 34 pending claims, approver assignments, expense policy art. 9
What it represents: €19,400 advanced, €7,800 of it for more than thirty days. Eleven claims were filed before 15 July.
What I looked at: why they are not approved. 26 of the 34 are waiting on the same person — a single approver for three teams, away for two weeks in July with no deputy appointed.
So it is not a matter of diligence: it is a single point of passage, and it was absent.
What I propose: the 26 grouped into one list with their receipts, approvable in a single session, and going forward a deputy appointed beyond five days' absence. That is an organisational setting, not an instruction.
What I have costed, should the question come back: an automatic approval threshold at €50 would cover 61 of the 137 claims in the queue, for €4,200. That is an internal control decision, and it is yours — it is taken for what it is, not to make up two weeks of absence. The deputy settles the calendar without touching the control.
The figure that speaks loudest: the average reimbursement time is 41 days; your policy states 15 days. 34-claims_19400-eur-advanced.pdf26 on one approver · 41 days against 15 stated
⛓ Source · 34 pending claims, approver assignments, expense policy art. 9
And the 22 illegible receipts?
They all come from the same place, and it is not the employees' fault.
What I found: 22 illegible receipts this month, all photographed from the mobile app's filing screen. None among those filed from a computer.
What that indicates: the capture screen compresses the image or does not allow checking before sending. The employee sees a sharp photo, the company receives an unusable image, and they learn of it three weeks later when the paper is asked for — which they have thrown away.
What I propose, in this order: ask the 22 employees for the receipts within 24 hours, while they still have the ticket; then have the capture screen checked.
What becomes of the 22 claims: they stay live, and not one is rejected. An illegible receipt is not a missing receipt, and the cause lies with the tool.
Over twelve months: 187 illegible receipts, 171 of them from the app. And 34 claims were rejected for that reason, leaving €2,100 out of employees' own pockets. 187-receipts_171-from-the-app.pdf34 claims rejected · €2,100 borne by employees
⛓ Source · 22 receipts, filing channel, 12-month history
What I found: 22 illegible receipts this month, all photographed from the mobile app's filing screen. None among those filed from a computer.
What that indicates: the capture screen compresses the image or does not allow checking before sending. The employee sees a sharp photo, the company receives an unusable image, and they learn of it three weeks later when the paper is asked for — which they have thrown away.
What I propose, in this order: ask the 22 employees for the receipts within 24 hours, while they still have the ticket; then have the capture screen checked.
What becomes of the 22 claims: they stay live, and not one is rejected. An illegible receipt is not a missing receipt, and the cause lies with the tool.
Over twelve months: 187 illegible receipts, 171 of them from the app. And 34 claims were rejected for that reason, leaving €2,100 out of employees' own pockets. 187-receipts_171-from-the-app.pdf34 claims rejected · €2,100 borne by employees
⛓ Source · 22 receipts, filing channel, 12-month history
The 2024 mileage rates — what does that change?
It changes what your employees receive, and I give you the gap without saying what to do.
What I record: the rates stored in your tool date from January 2024. They are applied to 412 journeys this year, that is 68,400 km.
What I do as soon as the rates are settled: I recompute the 412 journeys to the cent, in one minute, with the breakdown per employee. The rates themselves are chosen with your accountant: their tax and social consequences belong to them, and rates changed by a tool propagate to hundreds of reimbursements before anyone notices.
What I provide: the date of the stored rates, the number of journeys and kilometres concerned, and the accounting line they are posted to. That is what is needed to put the question.
And one thing I checked: your three pool cars are not covered by these rates, and 7 journeys this year were nonetheless reimbursed per kilometre while a pool car was booked in the same name on the same day.
I do not say that is improper. A booked car is not a used car, and a booking cancelled late stays in the planner. I put the question to the 4 employees concerned, not to the approver. 2024-rates_412-journeys.pdf68,400 km · 7 journeys to check · question to employees
⛓ Source · stored rates, 412 journeys, pool car booking planner
What I record: the rates stored in your tool date from January 2024. They are applied to 412 journeys this year, that is 68,400 km.
What I do as soon as the rates are settled: I recompute the 412 journeys to the cent, in one minute, with the breakdown per employee. The rates themselves are chosen with your accountant: their tax and social consequences belong to them, and rates changed by a tool propagate to hundreds of reimbursements before anyone notices.
What I provide: the date of the stored rates, the number of journeys and kilometres concerned, and the accounting line they are posted to. That is what is needed to put the question.
And one thing I checked: your three pool cars are not covered by these rates, and 7 journeys this year were nonetheless reimbursed per kilometre while a pool car was booked in the same name on the same day.
I do not say that is improper. A booked car is not a used car, and a booking cancelled late stays in the planner. I put the question to the 4 employees concerned, not to the approver. 2024-rates_412-journeys.pdf68,400 km · 7 journeys to check · question to employees
⛓ Source · stored rates, 412 journeys, pool car booking planner
How do you sort the gaps against the policy?
Into three families, and only one calls for a decision.
· Above the rate. An amount exceeds the ceiling your policy sets — a €27 meal against a €22 ceiling. 84 cases this quarter. There is nothing to decide: the policy provides for reimbursement at the ceiling, and I compute it.
· Outside the policy. A type of spend the policy does not provide for — 31 cases. There, somebody has to decide: it is not an overrun, it is a case not provided for.
· Receipt missing or illegible. 22 cases. Nothing to decide either: the receipt is needed.
Why that separation matters: the 84 and the 22 are handled mechanically. The 31 are the only ones requiring approver time, and today they are drowned in a queue of 137.
Across the 31 outside the policy: 19 concern three types of spend — checked baggage, roaming data charges, and long-stay parking. Three lines to add to the policy, and the 19 disappear.
What I bring on each of the 31: the circumstance declared, the closest comparable spend already reimbursed here, and the decision taken that time. The yes or no stays with the approver — it is not a question of rules, it is a question of circumstances — but they decide on a precedent, not on a hunch, and the 31 are handled in an hour. 137-gaps_3-families.pdf84 above rate · 31 outside policy · 22 receipts
⛓ Source · 137 gaps this quarter, expense policy, spend categories
· Above the rate. An amount exceeds the ceiling your policy sets — a €27 meal against a €22 ceiling. 84 cases this quarter. There is nothing to decide: the policy provides for reimbursement at the ceiling, and I compute it.
· Outside the policy. A type of spend the policy does not provide for — 31 cases. There, somebody has to decide: it is not an overrun, it is a case not provided for.
· Receipt missing or illegible. 22 cases. Nothing to decide either: the receipt is needed.
Why that separation matters: the 84 and the 22 are handled mechanically. The 31 are the only ones requiring approver time, and today they are drowned in a queue of 137.
Across the 31 outside the policy: 19 concern three types of spend — checked baggage, roaming data charges, and long-stay parking. Three lines to add to the policy, and the 19 disappear.
What I bring on each of the 31: the circumstance declared, the closest comparable spend already reimbursed here, and the decision taken that time. The yes or no stays with the approver — it is not a question of rules, it is a question of circumstances — but they decide on a precedent, not on a hunch, and the 31 are handled in an hour. 137-gaps_3-families.pdf84 above rate · 31 outside policy · 22 receipts
⛓ Source · 137 gaps this quarter, expense policy, spend categories
And the accounting code — we key it in every time. Can you propose it?
The category is proposed, never set: the code sits on the claim awaiting confirmation from the team, and it stays changeable with one click.
What the proposal rests on, and I show it every time: the nature read on the receipt, the merchant and its activity, the amount, the date matched against the employee's calendar where shared, and your last 18 months of already-coded claims — it is your chart of accounts that decides, not a general nomenclature.
Across the month's 412 claims: 361 categories proposed with a single hypothesis, each with the sentence that grounds it — "catering, 2 covers, working day: account 6257, as on the 143 equivalent claims of the last twelve months".
51 arrive with two hypotheses and no preference, because the receipt does not settle it: a hotel billing room and meal on one line, a toll on a mixed journey, a supermarket purchase. Two accounts proposed, the difference costed, and the decision with the accountant.
What the proposal saves, measured: coding took 2 minutes 40 a claim; it now takes 25 seconds of confirmation. Across 412 monthly claims, 15 hours 20 returned a month, or 184 hours over the year — more than five working weeks, a week counted as 35 hours, rounded down.
The figure that does not suit me: 38 of the 361 proposals were corrected, or 10.5%. Thirty-one concerned hospitality, which your chart splits across two accounts depending on whether the guest is a customer or an employee — information that is never on the receipt. The claim form now asks who the guest was at submission, as a single choice: 4 corrections across the following 96 hospitality claims. And the rule does not move: a proposed category is not an applied category, and no claim is posted without a person having confirmed it.
✎ Framework · category proposed, confirmed by the accounts team
What the proposal rests on, and I show it every time: the nature read on the receipt, the merchant and its activity, the amount, the date matched against the employee's calendar where shared, and your last 18 months of already-coded claims — it is your chart of accounts that decides, not a general nomenclature.
Across the month's 412 claims: 361 categories proposed with a single hypothesis, each with the sentence that grounds it — "catering, 2 covers, working day: account 6257, as on the 143 equivalent claims of the last twelve months".
51 arrive with two hypotheses and no preference, because the receipt does not settle it: a hotel billing room and meal on one line, a toll on a mixed journey, a supermarket purchase. Two accounts proposed, the difference costed, and the decision with the accountant.
What the proposal saves, measured: coding took 2 minutes 40 a claim; it now takes 25 seconds of confirmation. Across 412 monthly claims, 15 hours 20 returned a month, or 184 hours over the year — more than five working weeks, a week counted as 35 hours, rounded down.
The figure that does not suit me: 38 of the 361 proposals were corrected, or 10.5%. Thirty-one concerned hospitality, which your chart splits across two accounts depending on whether the guest is a customer or an employee — information that is never on the receipt. The claim form now asks who the guest was at submission, as a single choice: 4 corrections across the following 96 hospitality claims. And the rule does not move: a proposed category is not an applied category, and no claim is posted without a person having confirmed it.
✎ Framework · category proposed, confirmed by the accounts team
Can you show me the employees who spend the most?
No, and I would rather tell you why than offer you something else.
What that ranking would appear to say: who costs most. What it would actually say: who travels most — that is, in most companies, salespeople and field engineers.
What happens once it exists: it circulates. An employee at the top of an expense ranking is at the top of an expense ranking, whatever explanations accompany it, and they stay there in the memory of everyone who saw it.
What I can produce instead, and it answers the real question: spend by category, by project, by client and by month. If a trip costs too much, it is the trip to look at, not the person who made it.
And if the question is one of abuse: it arises on a specific claim, with its receipts, with a specific person — and that is a conversation, not a dashboard.
What I do in that case: I supply the claims concerned and the documents. I never supply a history of behaviour. what-is-produced_and-what-is-not.pdf4 axes of analysis · none per person
✎ Framework · no ranking of employees — spend by category, project, client
What that ranking would appear to say: who costs most. What it would actually say: who travels most — that is, in most companies, salespeople and field engineers.
What happens once it exists: it circulates. An employee at the top of an expense ranking is at the top of an expense ranking, whatever explanations accompany it, and they stay there in the memory of everyone who saw it.
What I can produce instead, and it answers the real question: spend by category, by project, by client and by month. If a trip costs too much, it is the trip to look at, not the person who made it.
And if the question is one of abuse: it arises on a specific claim, with its receipts, with a specific person — and that is a conversation, not a dashboard.
What I do in that case: I supply the claims concerned and the documents. I never supply a history of behaviour. what-is-produced_and-what-is-not.pdf4 axes of analysis · none per person
✎ Framework · no ranking of employees — spend by category, project, client
What do you keep out of all this?
The claim and its documents, and nothing describing an employee over time.
What is kept: the claim, its receipts, the gap recorded and its family, the question put to the employee and their answer, the approver's decision, and the dates.
What is not kept: no spend profile per employee, no "gap frequency" indicator per person, no ranking, and no trace of a question beyond the claim it concerned.
The last line deserves reading: when I ask about a line and the employee answers, the question and the answer stay attached to that claim. They do not constitute a precedent. An employee asked for a clarification in March must not be "the employee we already had to ask" in October.
What the monthly summary contains: reimbursement times, types of spend outside the policy, causes of illegible receipts, and approval bottlenecks. Four indicators about the process. what-is-kept.pdf6 items kept · 4 impossible
✎ Framework · retention periods to be set by the company
What is kept: the claim, its receipts, the gap recorded and its family, the question put to the employee and their answer, the approver's decision, and the dates.
What is not kept: no spend profile per employee, no "gap frequency" indicator per person, no ranking, and no trace of a question beyond the claim it concerned.
The last line deserves reading: when I ask about a line and the employee answers, the question and the answer stay attached to that claim. They do not constitute a precedent. An employee asked for a clarification in March must not be "the employee we already had to ask" in October.
What the monthly summary contains: reimbursement times, types of spend outside the policy, causes of illegible receipts, and approval bottlenecks. Four indicators about the process. what-is-kept.pdf6 items kept · 4 impossible
✎ Framework · retention periods to be set by the company
Local inference · no data outside the EU
Your case is not here? That is exactly what a 15-minute conversation is for. Book the free audit →
Use cases
What does the agent actually do?
One agent, several stages of the expenses process. All these uses work in support, subject to your approval.
Included in your agent
The 3 capabilities essential to this promise are included, at no extra cost.
From 582 € excl. VAT / month
Reading the receipts
Extracts date, amount, VAT and supplier from the document provided.
Category proposed
Proposes the appropriate accounting category, for the team to confirm.
Applying your caps
Compares every line with your policy and shows the amount allowed.
Controls and safeguards
These 5 controls are built into the agent: they frame what it does, whatever plan you pick. They are not chosen and are not added to your order.
Human validation, exceptions and escalation
Status, safe closure and audit trail
Sources, access rights and handling of questions with no answer
Reconcile supporting documents, rules and accounting entries with an audit trail
Handle discrepancies and validations before any accounting entry or payment
What the agent must be connected to
This connection is required for the agent to work. It concerns your information system and is scoped during the audit.
Integrate with accounting and banking software and with approved platforms (French e-invoicing)
Need to go further?
These agents handle a different business process, with their own owner and their own price. They are added to this one.
Does your need fall outside this?
Book the free audit →
Build your agent
In 15 minutes we identify the most relevant agent — without oversizing the project.
The gain
How much time can the accounts team give back to checking?
By taking on the reading and the formatting, the effort shifts towards the lines that genuinely deserve a look. How large the gain is depends on your volume and remains to be confirmed by a pilot.
Keying in from the receipts
Categorising the spending
Checking the caps
Indicative figures, not contractual, to be confirmed by a pilot on the volume of claims handled each month. A reimbursement is a payment: approval belongs to the accounts team, the agent prepares and flags.
How it works
The stages of your AI agent project
1
Audit & scoping
15 minutes to target the use case with the best return.
2
Quote or direct sign-up
A catalogue offer is bought online; a specific need gets a costed quote.
3
Design
We design the agent and its guardrails.
4
Integration & testing
We connect your tools to the agent, which is itself hosted in France.
5
Rollout
Going live and training your team.
6
Operation
Continuous supervision and improvement.
Pricing
One package, one agent
An expense claim agent (reading, categorisation, caps), installed and operated for you.
Agility
Setup + controlled subscription
5,500 € excl. VAT setup
then 582 € excl. VAT/month — you invest at installation and pay a reduced subscription. Ideal for keeping the cost under control over time.
- Installation, configuration and training for your teams
- Operation, human oversight, updates and support
- Sovereign hosting in France, a dedicated and isolated resource
The simplest
Serenity
All inclusive, no setup fee
892 € excl. VAT /month
all inclusive, immediate start. No upfront investment: a single subscription. Ideal for starting quickly and simply.
- Setup included (installation, configuration, training)
- Operation, human oversight, updates and support
- Sovereign hosting in France, managed end to end
100% Sovereign
On site, you own it
8,830 € excl. VAT setup
then 783 € excl. VAT/month · + hardware from 2,491 € (one-off purchase, in addition) — a sovereign computer installed on your premises, maintained remotely. Models run locally, your data returned at the end of the contract. 36-month commitment.
- Hardware installed on your premises (you own it)
- French / European AI models run locally
- Secure remote maintenance (Pro support included)
Not included in the packages: AI consumption (model tokens), re-invoiced at real cost with no margin, and tracked in real time in your client area. Maintenance and supervision subscription for an initial term of 12 months for the Agility package, 24 months for the Serenity package and 36 months for the 100% Sovereign package, renewable; support levels (SLA 72 h / 24 h / 4 h) optional. Bespoke development, additional integrations or exceptional volumes are quoted separately. Support Monday to Friday, 9am to 6pm. Prices exclude VAT.
AI model:
none of the AI models offered currently carries a fixed surcharge. When the selected model carries a cost, that cost is shown when you choose it, before you order, and re-invoiced at the cost incurred, with no mark-up; usage is billed at the publisher's price. Publishers' prices are published in US dollars: the amount re-invoiced is the amount in euros actually borne by Blue Lemon Agent on the publisher's invoice, at that invoice's exchange rate, with no commission or mark-up.
Included components and additional components Components included in the base offer: the Blue Lemon Agent software foundation, the AI models listed in the order journey, the standard channels (Microsoft Teams, Slack, WhatsApp Business, email, website chat, calendars, Microsoft 365 / Google Workspace, file storage, market VoIP telephony, professional social-media pages and accounts, Google Business Profile), hosting in France for the package chosen, backups, supervision, updates and support. If adapting the AI agent to your constraints, your needs or your requests requires other paid components — a third-party publisher's software licence, paid API access to one of your applications, hosting of health data, for which French law requires an HDS-certified host (art. L. 1111-8 of the French Public Health Code), SecNumCloud-qualified hosting, a speech synthesis service, particular hardware —, they are offered to you as an option or on quotation and re-invoiced at the cost incurred; nothing is committed without your written agreement. Where the artificial intelligence model you choose entails an additional cost, that cost is shown to you before you order and re-invoiced to you at the cost incurred, with no margin.
What to expect
Go-live
2 to 3 weeks
Agent designed, channels connected, team trained.
Steady state
4 to 7 weeks
After a few weeks of real use, once the agent's behaviour matches what you expect.
Indicative estimate, adjusted to the options you keep. It is not a delivery commitment.
Our commitment
Four guarantees that matter to your expenses
Your teams' spending stays internalLocal inference or an isolated resource hosted in France; no receipt and no spending data entrusted to a third party, no data used to train a model.
Data in France, under French lawThe spending your staff incur: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The accounts team keeps the decisionThe agent produces expense claims that are prepared and checked, which can be checked and altered; no approval is automated.
Human oversight & traceabilityOn the volume of claims handled each month: systematic logging and tracking, in line with the AI Act.
Going further
Related resources
Frequently asked questions
Your questions, our answers
Does the agent approve the reimbursements?
No. It prepares the claim, applies your caps and groups the points to check. Approval concerns a payment: it falls to the accounts team.
What happens if a receipt is hard to read?
The line is flagged with the document shown on screen, rather than filled in by inference. An assumed amount has no place in an expense claim.
How are our caps applied?
Your reimbursement policy is set up by category at the design stage. The agent shows both the amount claimed and the amount allowed, with the rule applied.
Is our staff's spending protected?
Yes. The agent is hosted in France, on local inference or an isolated resource, with the deployment objective of processing and access operated within the European Union and an architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. That data is not used to train a third-party model.
Does it connect to our expenses tool?
Yes, to the common tools, and to receipts submitted by email or by photo. The integration is defined at the design stage.
How long does it take to deploy this agent?
A few weeks as a rule, depending on the monthly volume and how detailed your reimbursement policy is, after a free audit then a phase of design, integration and testing.
Related use cases
Other agents for administration
Let's talk
Let's size up the potential in your expense claims
15 minutes to frame your volume and your policy — hosted in France, supervised, with no commitment.