● B2B offer — After-sales & customer relations
After-sales: every return qualified, every warranty checked
A return request is handled quickly once you know which product it is, when it was bought and what the warranty covers. Your agent finds those elements, applies your warranty terms, proposes the reason for return and prepares the file. Hosted in France: your customers' contact details and their purchase history stay with you. The after-sales team decides the outcome.
Hosted in France
Customer history protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
After-sales & warranties · hosted in France
A customer is asking to return a product bought eight months ago.
Product and order identified: item, date of purchase, sales channel.
Your warranty terms are applied: the item is covered over this period for this type of fault.
Return file prepared, with the reason proposed and the documents expected from the customer.
🔗 Sourced · the order and the warranty terms
Your warranty terms are applied: the item is covered over this period for this type of fault.
Return file prepared, with the reason proposed and the documents expected from the customer.
🔗 Sourced · the order and the warranty terms
And if the fault is not covered by the warranty?
The file is presented with the applicable term cited, and the material supplied by the customer.
A goodwill gesture remains possible and is decided case by case: that judgement belongs to the after-sales team.
✎ Support · file presented, gesture decided by you
A goodwill gesture remains possible and is decided case by case: that judgement belongs to the after-sales team.
✎ Support · file presented, gesture decided by you
Local inference · no data outside the EU
After-sales files hosted in France
Sovereign by designLocal inference or hosting in France
GDPR & AI Act: governed deploymentTraceability & human oversight
TurnkeyDesigned, installed and operated for you
The after-sales team decidesThe agent prepares, never rules
✦ In brief
A Blue Lemon Agent after-sales agent identifies the product and order concerned, applies your warranty terms, proposes the reason for return and prepares the file with the documents expected. The goodwill gesture and the outcome of the file stay with the after-sales team. It runs on local inference or is hosted in France: customer contact details and purchase history stay with you, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
6
after-sales uses ready to deploy
0
decision taken without human approval
These figures describe our offer, not results measured at a client. How large the gain is on your volume of returns and the size of your catalogue is confirmed by a pilot.
The context
What does an AI agent bring to your after-sales service?
A return handled quickly and well leaves an excellent impression. What takes time is finding the order and the applicable term.
! The issue
Handling a return means bringing together the product, the date of purchase and the applicable warranty term. That search takes up most of the handling, while the decision itself takes moments. The agent gathers the material, cites the term used and prepares the file ready to be settled.
✓ Our answer
The after-sales team receives complete files and decides the outcome: acceptance, repair, exchange or a goodwill gesture. Those choices commit the company and the customer relationship, and stay human. Local inference or an isolated resource hosted in France: your customers' contact details and their purchase history are entrusted to no third party.
The decisive point
Your customers' contact details and their purchase history: sovereignty & compliance
Your customers' contact details and their purchase history are personal data to be protected. Here is how the architecture of our agents goes about it.
Local inference
The agent can run on a machine belonging to your organisation: no customer data and no after-sales file leaves the network.
Hosting in France
Otherwise, a dedicated and isolated resource hosted in France, under French law — your product returns and your warranties: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
For your customers' contact details and their purchase history, 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 warranty terms.
The warranty term always cited
Every file states the term used and the original order; encryption, role-based access and logging of every file prepared.
AI Act: governed deployment
The agent is strictly in support; no claim is accepted and no refund is decided 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 disputed files and complaints bearing on a product's safety, SecNumCloud and reinforced hosting are options depending on your requirements. A single architecture is designed to answer both the GDPR and extraterritorial exposure. Designed for deployment in line with the GDPR and the AI Act, after the processing, roles and context-specific risks have been assessed.
Demonstration
See the agent at work
5 real situations, taken from those that come up most often. Pick one: the exchange unfolds as it would in your organisation.
A scripted demonstration. These exchanges show how the agent behaves — its sources, its refusals, what it leaves to your teams. Nothing is sent from this page, no model is queried here, and the matters named are fictional. That is precisely what we promise your data.
The behaviours shown here — monitoring, automation rules, routing and reminders — are configured with you during deployment, from your tools, your rules and your thresholds.
The architecture points named in these exchanges — location, local execution, isolation, encryption, role-based access, logging — are not a guarantee attached to the demonstration: they are those of the architecture set out in your quotation, and verified before commissioning.
The company in this demonstration
Fictional companyVertuone — small domestic appliances and home tools, sold online and in store
- Sector
- Retail of small domestic appliances and home tools — online store, 4 shops, 2 marketplaces
- Headcount
- 68 employees, 6 of them in after-sales: 4 advisers, 1 manager, 1 workshop technician
- Customers
- Consumers in France and Belgium — 210,000 customers, 31 % of whom buy through a marketplace
- Volume
- 4,200 catalogue references, €148 average basket, 620 return requests a month
- Tools already in place
- E-commerce platform, in-store till system, ticketing tool, both marketplace accounts, carrier portal — the agent plugs into them, nothing is replaced
- Who decides
- The after-sales manager grants cover and goodwill credits; the sales director rules on anything above €300
- Room for improvement
- Finding the original order accounts for 50 % of handling a return; a request waits 2.4 days for its first reply; 1 file in 5 goes back and forth for want of a missing document
Vertuone handles its returns well and wants to handle them fast: most of its advisers' time currently goes into finding the original order and the applicable warranty term, and that is precisely the time the agent gives back. It runs on local inference on a machine belonging to the company and plugs into the e-commerce platform, the in-store tills, the ticketing tool, both marketplace accounts and the carrier portal: it gathers, it quantifies, it prepares — the after-sales team decides. The exchanges below cover one quarter, from the review of three years of files to the closing report.
This company, its figures and the exchanges that follow were invented for the demonstration. They illustrate a common situation; they describe no real client.
After-sales & warranties · hosted in France
I read your last three years of after-sales overnight — 21,400 return files — and one single variable separates your fast files from the ones that get stuck.
The gap, and it is a sharp one: across your 5,200 files closed in under 24 hours, the original order was identified on the first message in 94 % of cases; across your 3,100 files that ran past eight days, in 11 %. The reasons for return, meanwhile, are spread almost identically between the two groups — 3 points of difference at most. So it is not the difficulty of the fault that stretches a file, it is identifying the order, and your own time records give it as the first cost item: 50 % of handling a return.
What I have already done with that, without asking: I reconciled your four sales sources and rebuilt a single index of 610,000 order lines over three years — reference, channel, date, and serial number where one exists. 384,000 lines come from the online store, 135,000 from the two marketplaces, 91,000 from the tills of your 4 shops.
What that gives you this morning: of the 47 requests that arrived since last night, 45 have their original order attached, with purchase date, channel and exact reference. The other 2 carry the note “order not found”, never “probably this one”: the right order gives the right term, and a right term quoted first time saves you writing twice to the same customer.
What I suggest: that the after-sales manager reviews the index over one day's worth of requests, then that I apply it as each one arrives. Your advisers will start a file with a decision rather than with a search. three-years-of-returns_what-the-gap-shows.pdf5,200 fast files against 3,100 stuck ones, one single difference
⛓ Sourced · 21,400 after-sales files over three years, ticketing tool log, sales history across all 4 channels
The gap, and it is a sharp one: across your 5,200 files closed in under 24 hours, the original order was identified on the first message in 94 % of cases; across your 3,100 files that ran past eight days, in 11 %. The reasons for return, meanwhile, are spread almost identically between the two groups — 3 points of difference at most. So it is not the difficulty of the fault that stretches a file, it is identifying the order, and your own time records give it as the first cost item: 50 % of handling a return.
What I have already done with that, without asking: I reconciled your four sales sources and rebuilt a single index of 610,000 order lines over three years — reference, channel, date, and serial number where one exists. 384,000 lines come from the online store, 135,000 from the two marketplaces, 91,000 from the tills of your 4 shops.
What that gives you this morning: of the 47 requests that arrived since last night, 45 have their original order attached, with purchase date, channel and exact reference. The other 2 carry the note “order not found”, never “probably this one”: the right order gives the right term, and a right term quoted first time saves you writing twice to the same customer.
What I suggest: that the after-sales manager reviews the index over one day's worth of requests, then that I apply it as each one arrives. Your advisers will start a file with a decision rather than with a search. three-years-of-returns_what-the-gap-shows.pdf5,200 fast files against 3,100 stuck ones, one single difference
⛓ Sourced · 21,400 after-sales files over three years, ticketing tool log, sales history across all 4 channels
Our marketplace buyers never have our order number. That is exactly where our days go.
They do have a number, it simply is not your numbering. I use five identification routes, from the most certain to the broadest, and every file states which one was used.
A marketplace is a third-party site on which you sell under your own name: the order exists on their side, with their own numbering, and that is the only one the customer holds.
· The marketplace order number, reconciled with yours through their daily export table — certain route, 29 % of matches.
· The serial number on the rating plate, which designates one unit and therefore one unique line — certain as well, 24 %.
· The buyer account e-mail address, cross-checked with the reference and a date window — a single candidate in 9 cases out of 10, 31 % of matches; when there are two, I put both in front of you with what separates them.
· The parcel tracking number — the identifier the carrier assigns to the shipment and that the customer finds in the delivery e-mail — pulled from the carrier portal, 11 %.
· The in-store till receipt, found by date, amount and reference — 5 %.
The result over your last full month: 601 requests out of 620 matched to their original order, that is 97 %, of which 84 % on the customer's first message. On marketplace purchases, your largest room for progress, the rate reaches 95 %: the export table already existed, it only had to be cross-checked.
What that shifts, in minutes: finding the original order accounted for 50 % of handling a return, it now accounts for 7 %. Across 620 requests a month, this is the item that gives you back the most hours, and it calls for no decision on your side.
The next step I propose: recording the serial number at dispatch rather than at return. 1,900 of your 4,200 references carry one; captured at the outset, it makes identification certain on nearly half the catalogue. order-identification_5-routes.pdf601 of 620 matched, and by which route
⛓ Sourced · exports from both marketplaces, carrier portal, in-store tills, 620 requests from the last month
A marketplace is a third-party site on which you sell under your own name: the order exists on their side, with their own numbering, and that is the only one the customer holds.
· The marketplace order number, reconciled with yours through their daily export table — certain route, 29 % of matches.
· The serial number on the rating plate, which designates one unit and therefore one unique line — certain as well, 24 %.
· The buyer account e-mail address, cross-checked with the reference and a date window — a single candidate in 9 cases out of 10, 31 % of matches; when there are two, I put both in front of you with what separates them.
· The parcel tracking number — the identifier the carrier assigns to the shipment and that the customer finds in the delivery e-mail — pulled from the carrier portal, 11 %.
· The in-store till receipt, found by date, amount and reference — 5 %.
The result over your last full month: 601 requests out of 620 matched to their original order, that is 97 %, of which 84 % on the customer's first message. On marketplace purchases, your largest room for progress, the rate reaches 95 %: the export table already existed, it only had to be cross-checked.
What that shifts, in minutes: finding the original order accounted for 50 % of handling a return, it now accounts for 7 %. Across 620 requests a month, this is the item that gives you back the most hours, and it calls for no decision on your side.
The next step I propose: recording the serial number at dispatch rather than at return. 1,900 of your 4,200 references carry one; captured at the outset, it makes identification certain on nearly half the catalogue. order-identification_5-routes.pdf601 of 620 matched, and by which route
⛓ Sourced · exports from both marketplaces, carrier portal, in-store tills, 620 requests from the last month
And the 19 you do not find? That is where an adviser spends an hour for nothing.
They move forward by another route: all 19 go out the same day, instead of waiting for an adviser to find an hour to search.
What I do for each one:
· I write the customer the shortest message that will do — two items asked for, never five: a photo of the rating plate and the e-mail address used at purchase, with a visual example of the photo expected. 13 of the 19 came back identified in under 24 hours.
· I open a provisional file with what the customer supplied and the note “order not found” at the top. The adviser sees at a glance what is settled and what is missing — they do not repeat the search I have just run.
· I escalate the remaining 6 files to the manager, grouped once a day, with the five routes tried and their result: ten minutes of review, against nineteen searches one by one.
The rule that holds the 97 %, and the one I ask you to keep: every match carries the name of the route that produced it and how certain that route is, and when two orders remain candidates, I put both in front of you with what separates them. The adviser therefore always knows whether they are reading a fact or a reconciliation, and they decide on that basis in seconds.
The overall gain: your first reply used to go out in 2.4 days on average, it goes out in 3 hours — including on those 19 files. A customer who gets an answer the same day does not call back the next day, and your advisers get back the call they no longer have to take.
What I suggest testing this week: a personal tracking link sent with the first reply, where the customer sees the state of their file without writing. Over the trial month it cut follow-up messages by 41 % — people chase when they do not know.
What I do for each one:
· I write the customer the shortest message that will do — two items asked for, never five: a photo of the rating plate and the e-mail address used at purchase, with a visual example of the photo expected. 13 of the 19 came back identified in under 24 hours.
· I open a provisional file with what the customer supplied and the note “order not found” at the top. The adviser sees at a glance what is settled and what is missing — they do not repeat the search I have just run.
· I escalate the remaining 6 files to the manager, grouped once a day, with the five routes tried and their result: ten minutes of review, against nineteen searches one by one.
The rule that holds the 97 %, and the one I ask you to keep: every match carries the name of the route that produced it and how certain that route is, and when two orders remain candidates, I put both in front of you with what separates them. The adviser therefore always knows whether they are reading a fact or a reconciliation, and they decide on that basis in seconds.
The overall gain: your first reply used to go out in 2.4 days on average, it goes out in 3 hours — including on those 19 files. A customer who gets an answer the same day does not call back the next day, and your advisers get back the call they no longer have to take.
What I suggest testing this week: a personal tracking link sent with the first reply, where the customer sees the state of their file without writing. Over the trial month it cut follow-up messages by 41 % — people chase when they do not know.
We have 4,200 references and as many different sets of terms. Nobody here knows them all, and the customer notices.
They now sit in a single matrix, reference by reference, and every file cites the term used — not “the warranty”, that precise term, with its date.
Three regimes overlap on each of your references, and the matrix keeps them apart, line by line:
· The statutory warranty — the one the law imposes on the seller, at no extra cost, and from which a contract cannot depart. It covers your 4,200 references, across the four channels.
· The commercial warranty — the additional undertaking you or the manufacturer offer, on the terms you write. You have 14 versions of it depending on the brand, 3 of which changed duration mid-year: your 1,240 kitchen appliance references went from 2 to 3 years on 1 March.
· Supplier recourse — what your supplier owes you, when the product fails. It is the only one of the three that brings money in, and I bring you the count in a moment.
What the matrix carries for each reference: duration and starting point of each regime — delivery, first use or supplier invoice, they are not the same date —, faults covered and faults excluded, documents required, contact and recourse window, and the date the term was last updated. That date is what protects you the day a customer disputes.
What it shifts in time: checking the cover accounted for 35 % of handling a return, it now accounts for 6 %. An adviser no longer works through fourteen sets of terms to retain one: they read the term used, its source and its date, in one line.
What I suggest: that the manager approves the matrix brand by brand — 14 ten-minute readings, once only — and that I keep it current every time a supplier announces a change. You correct a line with a word: it is your material, I have set it out clearly. warranty-matrix_4200-references.pdfThree overlapping regimes, one term cited per file
⛓ Sourced · terms of sale, 14 brand commercial warranties, supplier contracts, catalogue of 4,200 references
Three regimes overlap on each of your references, and the matrix keeps them apart, line by line:
· The statutory warranty — the one the law imposes on the seller, at no extra cost, and from which a contract cannot depart. It covers your 4,200 references, across the four channels.
· The commercial warranty — the additional undertaking you or the manufacturer offer, on the terms you write. You have 14 versions of it depending on the brand, 3 of which changed duration mid-year: your 1,240 kitchen appliance references went from 2 to 3 years on 1 March.
· Supplier recourse — what your supplier owes you, when the product fails. It is the only one of the three that brings money in, and I bring you the count in a moment.
What the matrix carries for each reference: duration and starting point of each regime — delivery, first use or supplier invoice, they are not the same date —, faults covered and faults excluded, documents required, contact and recourse window, and the date the term was last updated. That date is what protects you the day a customer disputes.
What it shifts in time: checking the cover accounted for 35 % of handling a return, it now accounts for 6 %. An adviser no longer works through fourteen sets of terms to retain one: they read the term used, its source and its date, in one line.
What I suggest: that the manager approves the matrix brand by brand — 14 ten-minute readings, once only — and that I keep it current every time a supplier announces a change. You correct a line with a word: it is your material, I have set it out clearly. warranty-matrix_4200-references.pdfThree overlapping regimes, one term cited per file
⛓ Sourced · terms of sale, 14 brand commercial warranties, supplier contracts, catalogue of 4,200 references
And when the fault is not covered? That is where we fall out with the customer, and often lose them.
I put the file down with the applicable term cited and the three outcomes already quantified: a file that arrives with its three outcomes keeps the customer, and a customer kept is worth far more than a spare part.
What the adviser has in front of them, in one page: the term that excludes the fault and its date, the original order, how long the customer has been with you and what they have bought from you since. That last line changes the decision in one case out of three, and it is now there before the adviser opens the file.
· Workshop repair — part cost and technician slot. Possible in 1 case out of 3 in the families concerned, and often cheaper than a goodwill credit.
· A paid repair quote, sent with the real lead time. 34 % accept it when the amount goes out in the first reply, only 9 % when it arrives a week later.
· A goodwill credit, when the relationship warrants it: I give the amount I recommend and the calculation behind it, never a bare figure.
And here is what I found while building that page — not a saving, a receivable: over three years, 1,340 returns concerned references still covered by supplier recourse and were never charged back. At the average cost of €46.60 per unit, that is €62,400 you carried and that your contracts placed on your suppliers.
What I did with it: 214 files are still within the window, sorted by supplier with the supporting document each of them requires — €18,200 recoverable, and nothing for you to write; the first batch, 74 files, is worth €6,900 and still runs for 4 months. For what follows, I flag the recourse as soon as the file opens: the return is handled and the supplier is claimed against in the same movement.
What I suggest: that the manager reviews the 214 files in supplier batches — 7 batches, one hour in total — and I go out on the ones she approves. supplier-recourse_62400-euros.pdf1,340 returns never charged back, 214 still in time
⛓ Sourced · warranty matrix, supplier contracts, 21,400 return files, log of repair quotes
What the adviser has in front of them, in one page: the term that excludes the fault and its date, the original order, how long the customer has been with you and what they have bought from you since. That last line changes the decision in one case out of three, and it is now there before the adviser opens the file.
· Workshop repair — part cost and technician slot. Possible in 1 case out of 3 in the families concerned, and often cheaper than a goodwill credit.
· A paid repair quote, sent with the real lead time. 34 % accept it when the amount goes out in the first reply, only 9 % when it arrives a week later.
· A goodwill credit, when the relationship warrants it: I give the amount I recommend and the calculation behind it, never a bare figure.
And here is what I found while building that page — not a saving, a receivable: over three years, 1,340 returns concerned references still covered by supplier recourse and were never charged back. At the average cost of €46.60 per unit, that is €62,400 you carried and that your contracts placed on your suppliers.
What I did with it: 214 files are still within the window, sorted by supplier with the supporting document each of them requires — €18,200 recoverable, and nothing for you to write; the first batch, 74 files, is worth €6,900 and still runs for 4 months. For what follows, I flag the recourse as soon as the file opens: the return is handled and the supplier is claimed against in the same movement.
What I suggest: that the manager reviews the 214 files in supplier batches — 7 batches, one hour in total — and I go out on the ones she approves. supplier-recourse_62400-euros.pdf1,340 returns never charged back, 214 still in time
⛓ Sourced · warranty matrix, supplier contracts, 21,400 return files, log of repair quotes
Hold on. You would send the supplier claims yourself, and grant credits to our customers?
Yes, and not vaguely: on a written mandate — a dated framework stating what I carry out, within which limits, and which is withdrawn with a word. I have drafted both, all you have to do is read them.
Claiming from a supplier and granting a credit commit your money and your signature. That is not a reason to leave them waiting on a desk: it is a reason to frame them.
Mandate no. 1 — supplier recourse. Scope: the 7 suppliers under framework contract, to the exclusion of any other. Cap: none, since I only claim what the contract already owes you. Cumulative conditions: original order found by a certain or strong route, supporting document attached, recourse window still open. Term: 3 months, renewable in writing. Withdrawal: one word, immediate effect on all future sendings.
Mandate no. 2 — goodwill credit. Credits up to €60 per file, on a product whose order has been found and whose fault is established. Monthly cap: €3,000, automatic stop once reached. Above €60, the manager; above €300, the sales director — exactly as today. Term: 3 months. Withdrawal: one word.
What that changes in minutes, and this is the real point: a €24 credit waited an average of 2.1 days for the manager to be free; under mandate, it goes out in 4 minutes. The customer who receives their credit on the day they complain does not talk about you the way the one who waits two days does.
What always reaches you, built, quantified and with the term cited: product safety, a file already in dispute, anything above the cap, and the customer who asks for a person.
What I suggest: start with mandate no. 1, which only brings in what you are owed. If the €18,200 comes in as expected, the second will sign itself. mandates_recourse-and-goodwill.pdfCapped, dated, withdrawn with a word — both templates
✎ Framework · mandate templates, framework contracts with all 7 suppliers, approval delays measured over the quarter
Claiming from a supplier and granting a credit commit your money and your signature. That is not a reason to leave them waiting on a desk: it is a reason to frame them.
Mandate no. 1 — supplier recourse. Scope: the 7 suppliers under framework contract, to the exclusion of any other. Cap: none, since I only claim what the contract already owes you. Cumulative conditions: original order found by a certain or strong route, supporting document attached, recourse window still open. Term: 3 months, renewable in writing. Withdrawal: one word, immediate effect on all future sendings.
Mandate no. 2 — goodwill credit. Credits up to €60 per file, on a product whose order has been found and whose fault is established. Monthly cap: €3,000, automatic stop once reached. Above €60, the manager; above €300, the sales director — exactly as today. Term: 3 months. Withdrawal: one word.
What that changes in minutes, and this is the real point: a €24 credit waited an average of 2.1 days for the manager to be free; under mandate, it goes out in 4 minutes. The customer who receives their credit on the day they complain does not talk about you the way the one who waits two days does.
What always reaches you, built, quantified and with the term cited: product safety, a file already in dispute, anything above the cap, and the customer who asks for a person.
What I suggest: start with mandate no. 1, which only brings in what you are owed. If the €18,200 comes in as expected, the second will sign itself. mandates_recourse-and-goodwill.pdfCapped, dated, withdrawn with a word — both templates
✎ Framework · mandate templates, framework contracts with all 7 suppliers, approval delays measured over the quarter
Our files keep going backwards: a photo, an invoice or a serial number is missing.
It is measured and it is fixed: one file in five made a back-and-forth, now it is one in twenty-five.
A back-and-forth is a file sent back to the customer because a document is missing: two more days for them, a second contact for you, and often one extra parcel.
What was missing was not the care, it was the list: it changes with the reason, and nobody had ever written it down. I rebuilt it, reason by reason, from your 21,400 files — not what the terms of sale list, but what actually proved enough to close a file without reopening it. Every document I ask for has therefore proved that it serves.
· Failure on first use — rating plate, display or indicator light, date of first use. Three documents, and the file closes in 1 day.
· Damage on unpacking — sealed parcel if it exists, product, and inner packaging: that one carries the carrier claim, it weighed 62 % of the reopenings, and it is now asked for from the outset.
· Failure after several months — serial number, symptom, and date it appeared, which determines the applicable regime.
· Wrong reference delivered — the parcel label. One single document, file closed in 12 minutes.
The message goes out pre-filled, with a visual example of each photo expected. The example is what produced the gain, not the list: people happily send the right photo when they know what it looks like.
What it gives you back, in euros: 124 back-and-forths a month became 25; the 99 avoided each carried a second shipment at €9.80, that is €970 a month and €11,640 a year — and above all 20 adviser hours a month spent asking again for documents.
What I suggest: that I install the same list, visuals included, on the return form of your website. The documents then arrive with the request, and the first back-and-forth disappears. documents-expected_by-reason-for-return.pdfWhat actually proved enough to close, reason by reason
⛓ Sourced · 21,400 return files, reopening reasons, carrier rates under the current contract
A back-and-forth is a file sent back to the customer because a document is missing: two more days for them, a second contact for you, and often one extra parcel.
What was missing was not the care, it was the list: it changes with the reason, and nobody had ever written it down. I rebuilt it, reason by reason, from your 21,400 files — not what the terms of sale list, but what actually proved enough to close a file without reopening it. Every document I ask for has therefore proved that it serves.
· Failure on first use — rating plate, display or indicator light, date of first use. Three documents, and the file closes in 1 day.
· Damage on unpacking — sealed parcel if it exists, product, and inner packaging: that one carries the carrier claim, it weighed 62 % of the reopenings, and it is now asked for from the outset.
· Failure after several months — serial number, symptom, and date it appeared, which determines the applicable regime.
· Wrong reference delivered — the parcel label. One single document, file closed in 12 minutes.
The message goes out pre-filled, with a visual example of each photo expected. The example is what produced the gain, not the list: people happily send the right photo when they know what it looks like.
What it gives you back, in euros: 124 back-and-forths a month became 25; the 99 avoided each carried a second shipment at €9.80, that is €970 a month and €11,640 a year — and above all 20 adviser hours a month spent asking again for documents.
What I suggest: that I install the same list, visuals included, on the return form of your website. The documents then arrive with the request, and the first back-and-forth disappears. documents-expected_by-reason-for-return.pdfWhat actually proved enough to close, reason by reason
⛓ Sourced · 21,400 return files, reopening reasons, carrier rates under the current contract
Show me what a file you put on my desk actually looks like.
One page, eight blocks, the decision at the end. Here is one, in at 8:12 and ready at 8:16.
· The customer and her history — Mathilde Ancel, 4 years with you, 11 orders, 1 single return before this one. That is the line that steers the goodwill decision: it is at the top, not in an appendix.
· The original order — ref. VRT-4180, 12/04/2026, online store, €219, found by the serial number: certain route.
· The product — serial number 4180-K92418, technical sheet, 6 spare parts still available: repair is possible, and that is known before it is discussed.
· The reason for return I propose — failure after 4 months of use, symptom described by the customer, consistent with 23 identical cases on this reference.
· The warranty term used — commercial warranty 2 years, from delivery, updated on 1 March 2026. Fault covered.
· The supplier recourse — applicable, 24 months, 8 months of window left, €52 chargeable back, supporting document attached.
· The documents supplied by the customer — 3 expected, 3 present, 3 legible, checked one by one.
· The quantified outcomes — exchange within 48 h at €0 net after recourse, workshop repair in 6 days for €34, refund €219. And the decision box, empty.
What is ready behind it and goes out on a click: the prepaid return label, the workshop slot provisionally booked, the message to the customer in your house style, and the supplier claim. Nothing goes out while the box is unticked.
The time shifted: building a file accounted for 30 % of handling a return, it now accounts for 8 %; end to end, 24 minutes of preparation become 4 — and those 4 minutes are your reading, the part that decides.
What I suggest: a trial on a single family, domestic vacuum cleaning, 62 files a month. If the format suits you, I extend it to the 4,200 references in a day. return-file_the-full-template.pdfEight blocks, four actions ready, one empty box
⛓ Sourced · the 8:12 file, customer history, warranty matrix, workshop availability and carrier portal
· The customer and her history — Mathilde Ancel, 4 years with you, 11 orders, 1 single return before this one. That is the line that steers the goodwill decision: it is at the top, not in an appendix.
· The original order — ref. VRT-4180, 12/04/2026, online store, €219, found by the serial number: certain route.
· The product — serial number 4180-K92418, technical sheet, 6 spare parts still available: repair is possible, and that is known before it is discussed.
· The reason for return I propose — failure after 4 months of use, symptom described by the customer, consistent with 23 identical cases on this reference.
· The warranty term used — commercial warranty 2 years, from delivery, updated on 1 March 2026. Fault covered.
· The supplier recourse — applicable, 24 months, 8 months of window left, €52 chargeable back, supporting document attached.
· The documents supplied by the customer — 3 expected, 3 present, 3 legible, checked one by one.
· The quantified outcomes — exchange within 48 h at €0 net after recourse, workshop repair in 6 days for €34, refund €219. And the decision box, empty.
What is ready behind it and goes out on a click: the prepaid return label, the workshop slot provisionally booked, the message to the customer in your house style, and the supplier claim. Nothing goes out while the box is unticked.
The time shifted: building a file accounted for 30 % of handling a return, it now accounts for 8 %; end to end, 24 minutes of preparation become 4 — and those 4 minutes are your reading, the part that decides.
What I suggest: a trial on a single family, domestic vacuum cleaning, 62 files a month. If the format suits you, I extend it to the 4,200 references in a day. return-file_the-full-template.pdfEight blocks, four actions ready, one empty box
⛓ Sourced · the 8:12 file, customer history, warranty matrix, workshop availability and carrier portal
There are files where I want a person from the first minute. How do you spot them?
They are spotted on arrival and they reach you built — so faster than the others, not slower.
The five families I always escalate, with their monthly volume:
· Product safety — burn, fire, injury, electric shock. 2 a month, to the manager within the minute, with the list of customers who bought the same batch.
· A file already in dispute, or where the customer mentions an association, a lawyer or mediation — 4 a month, within the hour, dated history attached.
· An out-of-warranty request from a long-standing customer — 11 a month: the decision there is commercial, not contractual, and I attach their purchase value over 3 years.
· An amount above the mandate, so above €60, and in every case above €300 — 9 a month, with the three quantified outcomes.
· A customer who asks for a person — 5 a month, switched within 10 minutes, with everything already gathered: they do not tell their story twice.
What that represents: 31 files a month out of 620, that is 5 % — and the gain is here: those 31 files used to arrive at the end of the day, after the 95 % mechanical ones. They now come first, the file already built, at the hour when attention is whole.
What the sorting rests on, and it is checkable line by line: what the customer asks for, the reason, the amount, the seniority and the reference — facts written in their message or in your own base, each of them arguable. Sorting built on a fact can be reread, challenged and corrected in one line: that is what lets you justify an escalation to the manager as well as to a customer, and the moment a word says they want a person, they get one.
What I suggest: a monthly review of those 31 files with the manager. Over the quarter, 3 should have been escalated sooner — I bring the tightened version of the rule to each review, already run over the past month, and you approve the one that suits you. reserved-files_the-human-5-percent.pdfFive families, 31 files a month, built in advance
✎ Framework · escalation rules approved by the after-sales manager, log of escalations over the quarter
The five families I always escalate, with their monthly volume:
· Product safety — burn, fire, injury, electric shock. 2 a month, to the manager within the minute, with the list of customers who bought the same batch.
· A file already in dispute, or where the customer mentions an association, a lawyer or mediation — 4 a month, within the hour, dated history attached.
· An out-of-warranty request from a long-standing customer — 11 a month: the decision there is commercial, not contractual, and I attach their purchase value over 3 years.
· An amount above the mandate, so above €60, and in every case above €300 — 9 a month, with the three quantified outcomes.
· A customer who asks for a person — 5 a month, switched within 10 minutes, with everything already gathered: they do not tell their story twice.
What that represents: 31 files a month out of 620, that is 5 % — and the gain is here: those 31 files used to arrive at the end of the day, after the 95 % mechanical ones. They now come first, the file already built, at the hour when attention is whole.
What the sorting rests on, and it is checkable line by line: what the customer asks for, the reason, the amount, the seniority and the reference — facts written in their message or in your own base, each of them arguable. Sorting built on a fact can be reread, challenged and corrected in one line: that is what lets you justify an escalation to the manager as well as to a customer, and the moment a word says they want a person, they get one.
What I suggest: a monthly review of those 31 files with the manager. Over the quarter, 3 should have been escalated sooner — I bring the tightened version of the rule to each review, already run over the past month, and you approve the one that suits you. reserved-files_the-human-5-percent.pdfFive families, 31 files a month, built in advance
✎ Framework · escalation rules approved by the after-sales manager, log of escalations over the quarter
You read our orders, our customers' contact details, ten years of their purchases. Where does all that go?
Nowhere: I run on local inference on a Vertuone machine, and your customer base is not copied to any outside server.
Local inference means the model computes on your machine: a customer's name, address and purchases cross no external network. The other route, if you would rather not host a machine: an isolated resource in France, dedicated to your company, with no pooling with another retailer.
· Hosted in France, under French law, 0 data outside the European Union — architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity, including against a US operator hosting in Europe.
· Nothing I read trains a model, neither yours nor a third party's: what I learn from your returns serves your returns.
· Encryption in transit and at rest, and role-based access — rights follow the job: an adviser opens the files in their own queue, the technician sees the reference without the contact details, the manager sees everything.
· A full log: who asked what, when, and what was approved.
The figure that makes this commercial: you keep the address, phone number and ten years of purchases of 210,000 consumers, and receive 40 rights requests a month. Each one cost 25 minutes of searching across 4 systems; it costs 3, because the index knows where they appear: 16.7 hours a month become 2, that is 15 hours given back every month.
And you change no tool: I plug into the e-commerce platform, the tills, the ticketing tool, both marketplaces and the carrier portal: no migration, no extra subscription, and a few weeks to go live.
What I suggest: that I keep current the processing record asked for every year — sources, retention periods, access. And for your disputed files, reinforced hosting, SecNumCloud option included, sits on the same architecture. technical-framework_where-your-customers-live.pdfLocal inference, role-based access, processing in the EU targeted
✎ Framework · deployment architecture, access log, register of rights requests
Local inference means the model computes on your machine: a customer's name, address and purchases cross no external network. The other route, if you would rather not host a machine: an isolated resource in France, dedicated to your company, with no pooling with another retailer.
· Hosted in France, under French law, 0 data outside the European Union — architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity, including against a US operator hosting in Europe.
· Nothing I read trains a model, neither yours nor a third party's: what I learn from your returns serves your returns.
· Encryption in transit and at rest, and role-based access — rights follow the job: an adviser opens the files in their own queue, the technician sees the reference without the contact details, the manager sees everything.
· A full log: who asked what, when, and what was approved.
The figure that makes this commercial: you keep the address, phone number and ten years of purchases of 210,000 consumers, and receive 40 rights requests a month. Each one cost 25 minutes of searching across 4 systems; it costs 3, because the index knows where they appear: 16.7 hours a month become 2, that is 15 hours given back every month.
And you change no tool: I plug into the e-commerce platform, the tills, the ticketing tool, both marketplaces and the carrier portal: no migration, no extra subscription, and a few weeks to go live.
What I suggest: that I keep current the processing record asked for every year — sources, retention periods, access. And for your disputed files, reinforced hosting, SecNumCloud option included, sits on the same architecture. technical-framework_where-your-customers-live.pdfLocal inference, role-based access, processing in the EU targeted
✎ Framework · deployment architecture, access log, register of rights requests
And the marketplaces? Their deadlines are not ours, and their penalties come fast.
Four channels, four clocks, one single matrix — and the 3 penalties of the previous quarter have gone.
· Your online store — your terms, your withdrawal period, your carrier, acknowledgement within 24 hours.
· Your 4 shops — return at the counter, till receipt as proof, and I push the applicable term to the assistant's till, in three lines.
· Marketplace A — acknowledgement within 24 hours, on pain of a penalty and a drop in your seller rating — the indicator the marketplace computes on your response times, and on which your visibility in its results depends.
· Marketplace B — 48 hours, and a reason to be taken from their list of 7, where you have 11. I have written the mapping table once and for all: every file goes out with the right reason, and the return stays on the party your contract provides for.
The result over the quarter: 0 late acknowledgements on 574 marketplace requests, against 17 the previous quarter on 531 — including 3 penalties, for €340. Your seller rating on marketplace A went from 4.2 to 4.6: it is the figure in this conversation that shows up in your sales.
The map of product faults, you have it: I built it on your 21,400 files and your own customer reviews. It was amply enough: 3 references exceed a 9 % return rate against 2.1 % on average, the first at 11.4 %, for €9,800 of returns a year. Those three are worth a supplier conversation, not a discount. And it is the route that stands up to a lawyer: database law reserves the extraction of a substantial part of a protected base, and your own data spares you that entirely.
What I suggest: that I watch the return rate per reference every week, at the alert threshold you set. A reference that slips shows up in ten days instead of a quarter, and that is when a supplier agrees to take a batch back. four-channels_one-single-rule.pdf0 late acknowledgements on 574, seller rating 4.2 → 4.6
⛓ Sourced · both marketplace accounts, acknowledgement log, return rate per reference over 3 years
· Your online store — your terms, your withdrawal period, your carrier, acknowledgement within 24 hours.
· Your 4 shops — return at the counter, till receipt as proof, and I push the applicable term to the assistant's till, in three lines.
· Marketplace A — acknowledgement within 24 hours, on pain of a penalty and a drop in your seller rating — the indicator the marketplace computes on your response times, and on which your visibility in its results depends.
· Marketplace B — 48 hours, and a reason to be taken from their list of 7, where you have 11. I have written the mapping table once and for all: every file goes out with the right reason, and the return stays on the party your contract provides for.
The result over the quarter: 0 late acknowledgements on 574 marketplace requests, against 17 the previous quarter on 531 — including 3 penalties, for €340. Your seller rating on marketplace A went from 4.2 to 4.6: it is the figure in this conversation that shows up in your sales.
The map of product faults, you have it: I built it on your 21,400 files and your own customer reviews. It was amply enough: 3 references exceed a 9 % return rate against 2.1 % on average, the first at 11.4 %, for €9,800 of returns a year. Those three are worth a supplier conversation, not a discount. And it is the route that stands up to a lawyer: database law reserves the extraction of a substantial part of a protected base, and your own data spares you that entirely.
What I suggest: that I watch the return rate per reference every week, at the alert threshold you set. A reference that slips shows up in ten days instead of a quarter, and that is when a supplier agrees to take a batch back. four-channels_one-single-rule.pdf0 late acknowledgements on 574, seller rating 4.2 → 4.6
⛓ Sourced · both marketplace accounts, acknowledgement log, return rate per reference over 3 years
One last thing: does the customer know they are writing to a machine?
Yes, from the first message, and it is not a box to tick in a settings panel.
Since 2 August 2026, Article 50(1) of the European regulation on artificial intelligence requires that anyone interacting with an AI system be informed, unless this is obvious. The notice is built into the welcome message, in your customer's language — French or Dutch for your Belgian buyers — and they can ask for a human adviser at any moment.
What that produces in practice: across the 1,860 files of the quarter, 29 customers asked for a human, that is 1.6 %, and all 29 were switched in under ten minutes, their file already gathered. Announcing the machine costs nothing: what did cost was the waiting, and it has gone.
· 0 decision taken without human approval across those 1,860 files: cover, exchange, refund, goodwill credit — each carries the name of the person who granted it and the time. That is how I am wired in, not a setting.
· One trace per file: what I read, the term I cited, what I proposed, what was decided — enough to answer a customer, a supplier or a mediator with dated facts.
· An indicator per adviser, I know how to produce it and I hand it to you within a day: the law allows it under three conditions — proportionality, prior information of the people concerned, consultation of your staff representatives —, and I bring you all three met, with the file ready to present. The decision is yours, and it is lawful.
What I recommend, figures in hand: start with the table of files — time, reason, reference, reopening rate —, because it points straight at the useful lever: your 148 reworks this quarter came from three out-of-date sets of terms, and re-entering them gave back 5.6 points. Tell me which of the two tables you want, I will build it this week.
✎ Framework · approval log for the quarter, multilingual welcome message, register of requests for a human adviser
Since 2 August 2026, Article 50(1) of the European regulation on artificial intelligence requires that anyone interacting with an AI system be informed, unless this is obvious. The notice is built into the welcome message, in your customer's language — French or Dutch for your Belgian buyers — and they can ask for a human adviser at any moment.
What that produces in practice: across the 1,860 files of the quarter, 29 customers asked for a human, that is 1.6 %, and all 29 were switched in under ten minutes, their file already gathered. Announcing the machine costs nothing: what did cost was the waiting, and it has gone.
· 0 decision taken without human approval across those 1,860 files: cover, exchange, refund, goodwill credit — each carries the name of the person who granted it and the time. That is how I am wired in, not a setting.
· One trace per file: what I read, the term I cited, what I proposed, what was decided — enough to answer a customer, a supplier or a mediator with dated facts.
· An indicator per adviser, I know how to produce it and I hand it to you within a day: the law allows it under three conditions — proportionality, prior information of the people concerned, consultation of your staff representatives —, and I bring you all three met, with the file ready to present. The decision is yours, and it is lawful.
What I recommend, figures in hand: start with the table of files — time, reason, reference, reopening rate —, because it points straight at the useful lever: your 148 reworks this quarter came from three out-of-date sets of terms, and re-entering them gave back 5.6 points. Tell me which of the two tables you want, I will build it this week.
✎ Framework · approval log for the quarter, multilingual welcome message, register of requests for a human adviser
After a quarter, what has this given us back, precisely?
207 hours a month given back to your six people, and they went to the phone and to the hard files — not to doing more files.
On the three items you measure: finding the original order, 50 % → 7 % · checking the cover, 35 % → 6 % · building the file, 30 % → 8 %. End to end, 24 minutes of preparation per file become 4; across 620 files a month, that is 207 hours.
What those hours produced, from your records: first reply in 3 hours instead of 2.4 days · back-and-forths from 1 in 5 to 1 in 25, that is €11,640 of postage a year · 0 late acknowledgements on the marketplaces and seller rating from 4.2 to 4.6 · €18,200 of supplier recourse already claimed out of the €62,400 identified. The last two lines are not time savings: they are receivables.
The figure that does not flatter me, and that I give myself: of the 1,860 files prepared this quarter, 148 were reworked by an adviser because the term I had picked was wrong — 8 %.
The cause: 111 of those 148 concern three suppliers. Two had changed the duration of their commercial warranty mid-year without your internal table being updated; the third only supplies its terms as a scanned document, two lines of which I had misread. It was a source problem, not a judgement problem.
What I did about it, and the result: the three sets were re-entered and reviewed by the manager, and I now flag any term whose source is more than six months old. Over the last six weeks the rework rate has gone from 8 % to 2.4 %: the remaining 2.4 % are one-off cases, routed on arrival to the person who should handle them.
What I suggest next: asking your 7 suppliers in writing for the dated version of their terms, every six months. Two of them have never sent it unprompted. quarterly-report_207-hours-and-8-percent.pdfWhat came back, what was reworked, what was corrected
⛓ Sourced · 1,860 files of the quarter, departmental time records, rework log, carrier rates, marketplace accounts
On the three items you measure: finding the original order, 50 % → 7 % · checking the cover, 35 % → 6 % · building the file, 30 % → 8 %. End to end, 24 minutes of preparation per file become 4; across 620 files a month, that is 207 hours.
What those hours produced, from your records: first reply in 3 hours instead of 2.4 days · back-and-forths from 1 in 5 to 1 in 25, that is €11,640 of postage a year · 0 late acknowledgements on the marketplaces and seller rating from 4.2 to 4.6 · €18,200 of supplier recourse already claimed out of the €62,400 identified. The last two lines are not time savings: they are receivables.
The figure that does not flatter me, and that I give myself: of the 1,860 files prepared this quarter, 148 were reworked by an adviser because the term I had picked was wrong — 8 %.
The cause: 111 of those 148 concern three suppliers. Two had changed the duration of their commercial warranty mid-year without your internal table being updated; the third only supplies its terms as a scanned document, two lines of which I had misread. It was a source problem, not a judgement problem.
What I did about it, and the result: the three sets were re-entered and reviewed by the manager, and I now flag any term whose source is more than six months old. Over the last six weeks the rework rate has gone from 8 % to 2.4 %: the remaining 2.4 % are one-off cases, routed on arrival to the person who should handle them.
What I suggest next: asking your 7 suppliers in writing for the dated version of their terms, every six months. Two of them have never sent it unprompted. quarterly-report_207-hours-and-8-percent.pdfWhat came back, what was reworked, what was corrected
⛓ Sourced · 1,860 files of the quarter, departmental time records, rework log, carrier rates, marketplace accounts
What do you do without asking us? I want the exact list.
Three actions, not one more, and each undoes as fast as it does.
· I acknowledge a request within the hour, with what is already identified and what remains to be supplied — an acknowledgement within the hour beats a perfect reply two days later. And the reverse holds too: if the customer replies that they would rather deal with a person, the request goes back to manual and stays there, without anyone having to re-enter it.
· I provisionally book a workshop slot when repair is the most likely outcome. It falls away by itself if the decision goes elsewhere: the slot fits the file, never the file the slot.
· I alert the manager within the minute on a product safety report, with the list of customers from the same batch. That one stays active in all circumstances: it is the only action where waiting costs more than being wrong.
Six acts wait for a decision, and each carries a person's name: granting cover, issuing a credit above the mandate, approving a refund, committing to a paid repair, writing to a supplier outside the mandate, closing a disputed file. What I bring to each is constant: the term cited and dated, the three quantified outcomes, the full history of exchanges.
What changed for your advisers, and it is the only figure that counts for them: they used to handle 18 files a day, 11 of which required no decision at all. They handle as many, but the 11 now arrive decided at a glance and the time went to the other 7: your records show 34 % more time on the files that matter, and 0 overtime in the department.
What I suggest: a review of this list every quarter. An automatic action that has served no purpose in three months comes off the list — it is its brevity that makes it credible, to your teams as well as to your customers. who-decides-what_three-actions-six-decisions.pdfThree reversible actions, six reserved decisions
✎ Framework · list of automatic actions approved by the after-sales manager, departmental activity records
· I acknowledge a request within the hour, with what is already identified and what remains to be supplied — an acknowledgement within the hour beats a perfect reply two days later. And the reverse holds too: if the customer replies that they would rather deal with a person, the request goes back to manual and stays there, without anyone having to re-enter it.
· I provisionally book a workshop slot when repair is the most likely outcome. It falls away by itself if the decision goes elsewhere: the slot fits the file, never the file the slot.
· I alert the manager within the minute on a product safety report, with the list of customers from the same batch. That one stays active in all circumstances: it is the only action where waiting costs more than being wrong.
Six acts wait for a decision, and each carries a person's name: granting cover, issuing a credit above the mandate, approving a refund, committing to a paid repair, writing to a supplier outside the mandate, closing a disputed file. What I bring to each is constant: the term cited and dated, the three quantified outcomes, the full history of exchanges.
What changed for your advisers, and it is the only figure that counts for them: they used to handle 18 files a day, 11 of which required no decision at all. They handle as many, but the 11 now arrive decided at a glance and the time went to the other 7: your records show 34 % more time on the files that matter, and 0 overtime in the department.
What I suggest: a review of this list every quarter. An automatic action that has served no purpose in three months comes off the list — it is its brevity that makes it credible, to your teams as well as to your customers. who-decides-what_three-actions-six-decisions.pdfThree reversible actions, six reserved decisions
✎ Framework · list of automatic actions approved by the after-sales manager, departmental activity records
A lot of people write to us before they even have a return to make. You do not handle that.
I already handle it, and I have quantified it: it is your biggest remaining seam — 34 % of what reaches after-sales is not a return.
Of the 940 incoming messages of the month, 320 required no file at all: “where is my delivery”, “how do I descale this”, “does this part fit that model”, “when will it be back in stock”. Your advisers handle them because they land in the same inbox, and each costs 6 minutes: 32 hours a month.
What I do today: I identify them on arrival and set them aside, with the reply already drafted when your product documentation holds it. Your advisers approve them in a batch instead of writing them one by one: 32 hours become 4.
What I recommend to finish the job, and I quantify it:
· A customer support agent upstream, on the website and in the general inbox, taking those 320 messages before they reach after-sales. The remaining 4 hours fall to zero, and the after-sales inbox holds nothing but returns.
· A voice agent on the switchboard: you receive 410 calls a month in the department, 6 in 10 of which end with “I will send you an e-mail”. That one qualifies, identifies the order during the call and hands me the file already open — the same journey, one step fewer.
And the argument to keep for the end of your own discussions: all of this lives in France, on your machine, with the deployment objective of processing and access operated within the European Union. The contact details of 210,000 consumers and ten years of their purchases do not leave your premises, and every file carries the term cited, its date and the name of whoever decided. For a retailer that is not a legal argument: it is the written answer you will have the day a customer raises the question.
What I suggest to start: customer support upstream, live in a few weeks, giving you back the 32 hours from the first month.
⛓ Sourced · 940 incoming messages of the month, switchboard log, time records by request type
Of the 940 incoming messages of the month, 320 required no file at all: “where is my delivery”, “how do I descale this”, “does this part fit that model”, “when will it be back in stock”. Your advisers handle them because they land in the same inbox, and each costs 6 minutes: 32 hours a month.
What I do today: I identify them on arrival and set them aside, with the reply already drafted when your product documentation holds it. Your advisers approve them in a batch instead of writing them one by one: 32 hours become 4.
What I recommend to finish the job, and I quantify it:
· A customer support agent upstream, on the website and in the general inbox, taking those 320 messages before they reach after-sales. The remaining 4 hours fall to zero, and the after-sales inbox holds nothing but returns.
· A voice agent on the switchboard: you receive 410 calls a month in the department, 6 in 10 of which end with “I will send you an e-mail”. That one qualifies, identifies the order during the call and hands me the file already open — the same journey, one step fewer.
And the argument to keep for the end of your own discussions: all of this lives in France, on your machine, with the deployment objective of processing and access operated within the European Union. The contact details of 210,000 consumers and ten years of their purchases do not leave your premises, and every file carries the term cited, its date and the name of whoever decided. For a retailer that is not a legal argument: it is the written answer you will have the day a customer raises the question.
What I suggest to start: customer support upstream, live in a few weeks, giving you back the 32 hours from the first month.
⛓ Sourced · 940 incoming messages of the month, switchboard log, time records by request type
Local inference · no data outside the EU
Your case is not here? That is exactly what a 15-minute conversation is for. Book the free audit →
Use cases
What does the agent actually do?
One agent, the whole journey of a return. 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 522 € excl. VAT / month
Qualifying the return
Identifies the product, order and sales channel from the request.
Checking the warranty
Applies your terms and cites the one that was used.
Preparing the file
Gathers the documents expected from the customer and proposes the reason for return. The steps that call for action from the customer are announced to them by text message.
Controls and safeguards
These 4 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
Resolve common requests through to a verifiable closure
Escalate quickly to a human, with the reason and the history
What the agent must be connected to
This connection is required for the agent to work. It concerns your information system and is scoped during the audit.
Carry the full context across channels and into the ticket/CRM
Other needs our agents cover
Each card says where the matching agent stands: available, on quote, or still being architected.
Customer support
For everyday questions upstream, a dedicated support agent takes them on.
On quote View the agent page →Need to go further?
These agents handle a different business process, with their own owner and their own price. They are added to this one.
E-commerce
For the online retail sector as a whole, see our dedicated page.
Product FAQ agent for e-commerce from 527 € excl. VAT / month E-commerce →Switchboard
For the department's incoming calls, a dedicated voice agent completes the picture.
Voice agent (AI switchboard) from 856 € excl. VAT / month Switchboard →
Does your need fall outside this?
Book the free audit →
Build your agent
In 15 minutes we identify the most relevant agent — without oversizing the project.
The gain
How many returns can a department handle without a wait?
By taking on the search and the building of the file, the effort shifts towards the decision and the contact with the customer. How large the gain is depends on your volume and remains to be confirmed by a pilot.
Finding the original order
Checking the cover
Building the return file
Indicative figures, not contractual, to be confirmed by a pilot on your volume of returns and the size of your catalogue. Acceptance, exchange, refund or a goodwill gesture commit the company: those decisions belong to the after-sales team.
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 after-sales agent (qualification, warranties, files), installed and operated for you.
Agility
Setup + controlled subscription
4,630 € excl. VAT setup
then 522 € 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
782 € 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
7,965 € excl. VAT setup
then 719 € excl. VAT/month · + hardware from 1,058 € (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 after-sales service
Your customers' history stays with youLocal inference or an isolated resource hosted in France; no contact details and no purchase history entrusted to a third party, no data used to train a model.
Data in France, under French lawYour customers' contact details and their purchase history: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The after-sales team keeps the decisionThe agent produces qualified return requests, which can be checked and altered; no approval is automated.
Human oversight & traceabilityOn your volume of returns and the size of your catalogue: systematic logging and tracking, in line with the AI Act.
Going further
Related resources
Frequently asked questions
Your questions, our answers
Does the agent grant the claims?
No. It applies your terms, cites the one used and prepares the file. Granting a claim or a goodwill gesture commits the company: the decision falls to the department.
How does it find the original order?
In your sales systems, from the details the customer provides. Where the order cannot be found, that is stated as such rather than assumed.
Can it handle several sales channels?
Yes: online shop, physical network, marketplaces, depending on the systems it is connected to at the design stage.
What does it do with a fault that is not covered?
It presents the file with the applicable term cited and the material the customer supplied, so the department can decide quickly what follows.
Which tools can our customers use to talk to the agent?
The ones they already use: WhatsApp Business, your website chat and email. Your customers have no account to create and no app to install. Your own teams reach the same agent from Microsoft Teams, Slack or their email. These connections rest on open standards, including the MCP protocol, and are included in every plan, at no extra cost, within the number of connections your level includes; the catalogue grows without a surcharge. Only the fees charged by third-party platforms — WhatsApp Business bills per conversation, and the gateway bills the text messages sent — are passed on to you at cost, without margin, outside the subscription. You keep oversight from a web dashboard.
Is our clients' data protected?
Yes. The agent is hosted in France, on local inference or an isolated resource, with the deployment objective of processing and access operated within the European Union and an architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. That data is not used to train a third-party model.
Does the agent state that it is an artificial intelligence?
Yes, from the very first interaction, and this is not a configuration option: since 2 August 2026, Article 50(1) of the European AI Regulation requires that any person interacting with an AI system be informed, unless this is obvious. The announcement is built into the greeting, in the other party’s language, and they can ask for a human at any time.
How long does it take to deploy this agent?
A few weeks as a rule, depending on the volume of returns and the integration with your sales systems, after a free audit then a phase of design, integration and testing.
Can the agent notify my customers by text message?
Yes, as an option. The agent acknowledges a return by text message, flags the missing document and announces that the parcel has reached the workshop. A text message is an outbound notification channel, not a conversation channel: no one writes back to the agent there — the conversation stays on WhatsApp Business, your website chat or email. The message announces without detailing, it carries your registered sender name, and it never contains a payment link, a login or a password: the recipient reaches your account area the way they usually do. Delivery goes through a gateway established in the European Union; the messages sent are passed on to you at actual cost, with no margin, and the channel itself is an optional module, set out on the pricing page.
Related use cases
Other agents for customer relations
Let's talk
Let's size up the potential in your after-sales service
15 minutes to frame your returns and your warranty terms — hosted in France, supervised, with no commitment.