● Public sector — Waste management & the waste charge
AI agent for managing waste and the waste charge
An assistant that answers residents, prepares the billing of the charge and tracks the requests — the officer keeps control of the files. Hosted in France, on a resource isolated for each authority: residents' data never leaves the country. The AI agent assists, the public officer decides.
Hosted in France
The public's data protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Waste & charge assistant · hosted in France
A resident reports that their recycling bin was not collected on rue des Tilleuls, and asks for a bigger bin for a household of five.
Request logged and qualified. The missed collection has been passed to the operations team (Tuesday round, sector 3). On the bin size: your allocation table provides for a 240 L bin for a household of five — the request form has been pre-filled.
The reply to the resident is ready to send.
⛓ Source · your collection rules + the bin allocation table
The reply to the resident is ready to send.
⛓ Source · your collection rules + the bin allocation table
Prepare the half-year billing of the pay-as-you-throw charge for households in the north sector.
Preparation complete. 1,240 households: fixed element + lifts counted against your pricing parameters. 18 anomalies detected (lifts with no subscription, duplicate chips) and set aside for checking.
Billing file and covering letters ready for review.
✎ Action · billing prepared — the public officer approves before issue
Billing file and covering letters ready for review.
✎ Action · billing prepared — the public officer approves before issue
Local inference · no data outside the EU
Data hosted in France
Sovereign by designLocal inference or hosting in France
GDPR & AI Act: governed deploymentTraceability & human oversight
TurnkeyDesigned, installed and operated for you
The public officer decidesThe agent assists, never approves alone
✦ In brief
In a waste department, a Blue Lemon Agent agent absorbs the repetitive work — replies to residents, billing of the waste charge, tracking bin requests and reports — and passes on what needs to go to the operations team. It runs on local inference or is hosted in France on a resource dedicated to and isolated for each authority: addresses, household composition and billing data are never exposed to a foreign service, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. The public officer approves every invoice and every letter; the time won back goes to the service given to residents. Live within a few weeks. Your public-sector staff write to it from Microsoft Teams, Slack or their email, and your users reach it on WhatsApp Business, the website chat or email — with no account to create and nothing to install. Reaching the administration from the tool people already have means less non-take-up of rights and equal access to the service. These connections are included in every plan, at no extra cost, within the number of connections your level includes.
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
7
uses ready to deploy on this scope
0
decision taken without human approval
Reference points describing our offer, not results measured at a client. The scale of the gain is confirmed by a pilot on your own scope.
The context
Why AI matters to waste departments — and why they hesitate
Between collection, recycling centres, the pay-as-you-throw charge and complaints, the waste department is one of the busiest in the authority. Requests arrive on every channel, and the billing demands flawless rigour on personal data: addresses, household composition, bin lifts.
! The issue
The teams are caught between residents who expect quick answers — a bin not collected, a bin size, the calendar, the rate — and a pay-as-you-throw charge whose billing is ever finer (lifts, drop-offs, anomalies). Yet most consumer AI tools amount to entrusting addresses, household composition and billing data to a third party, often hosted outside Europe and subject to the Cloud Act.
✓ Our answer
For public data, AI is only of interest if it is sovereign and confidential by design. Local inference or an isolated resource hosted in France, systematic human oversight, decisions reserved to the public officer: the pricing parameters stay set by the authority, and every invoice and every letter is approved before issue. The aim is not to replace the teams, but to give them back time for operations and for residents.
The decisive point
Protecting the public's data: sovereignty & compliance
A waste department handles personal data on a large scale: addresses, households, billing. Here is how the architecture of our agents protects it, authority by authority.
Local inference
The agent can run on a machine belonging to the authority: no data leaves the network, nothing passes through a cloud.
Hosting in France
Otherwise, a dedicated and isolated resource, hosted in France under French law — the public's data: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
For residents' data, the architecture aims to reduce exposure to the Cloud Act and FISA 702; being located in France or in the European Union does not, on its own, guarantee immunity.
One isolated resource per authority
No pooling of data: an environment strictly dedicated to your authority, guaranteeing the continuity of the public service.
Encryption & controlled access
Encryption in transit and at rest, role-based access (RBAC), strong authentication and logging.
AI Act: governed deployment
The agent is strictly in support; no invoice is issued 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.
- The encryption mechanisms in transit and at rest, their components and key management are those documented for the architecture chosen.
- Roles and permissions are configured and accepted for the identities and systems actually connected.
- The events logged, their content, their retention period and who may access them are defined for the deployment chosen.
For the most sensitive data, SecNumCloud and HDS options are available depending on your requirements. A single architecture is designed to answer both the GDPR and extraterritorial exposure. Designed for deployment in line with the GDPR and the AI Act, after the processing, roles and context-specific risks have been assessed.
Demonstration
See the agent at work
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.
The public body in this demonstration
Fictional public bodyTrois-Vallons Community of Communes — waste authority covering 17 municipalities (fictional authority)
- Sector
- In-house collection and charge authority for a community of communes of 26,800 inhabitants — kerbside collection of residual waste and packaging, 2 civic amenity sites, 46 bring banks, pay-as-you-throw household waste charge since 1 January 2024
- Headcount
- 34 staff in the authority, 8 of them on resident relations and charging — 6 case officers, one head of service and one apprentice; the other 26 are on collection rounds and civic amenity sites, and the agent does not touch that side
- Public served
- 12,400 households across 17 municipalities and 5 collection sectors — 21,600 enquiries a year: 12,900 phone calls, 5,400 emails, 2,100 counter visits, 1,200 website forms
- Order of magnitude
- 2 billing campaigns a year, 24,800 invoices issued; 3,940 collection complaints, 2,180 bin allocation or exchange requests, 1,460 civic amenity site enquiries, 6,300 charge letters; 3,620 tonnes of residual household waste collected in 2025, i.e. 135 kg per inhabitant since the move to pay-as-you-throw
- Tools already in place
- Waste management software (charge-payer file, bin chips, lift log), billing software, published collection calendar, service mailbox and Microsoft Teams — the agent reads them, nothing is replaced and nothing is migrated
- Who decides what
- The head of the waste service approves every letter and every billing file before it goes out; tariffs and the bin allocation grid are set by resolution of the community council; ex gratia waivers rest with the chair
- Room for improvement
- 4,128 calls out of 12,900 ring unanswered in the week following each billing run — 32%; an email waits 9 days on average; preparing a campaign ties up all 6 case officers for 11 working days; and the charge-payer file had never been reconciled against the lift log — 18 anomalies were sitting undisturbed in the north sector alone
At Trois-Vallons, eight staff carry resident relations and the waste charge for 12,400 households, with two annual peaks around the billing campaigns. The most repetitive questions — bin not collected, bin size, site opening hours, calendar — take up three quarters of the flow, and it is the cases that deserve a human that queue up behind them. The exchanges below cover a full year, from the first reconciliation of the file to the report presented to the community council. The agent is hosted in France on an isolated resource, reads the waste management software and the lift log without ever writing to them, and no invoice goes out without the head of service's approval.
This public body, its figures and the exchanges that follow were invented for the demonstration. They illustrate a common situation; they describe no real service.
Waste & charge assistant · hosted in France
I have reread your 21,600 enquiries for the year and reconciled, household by household, your charge-payer file against the lift log — the reconciliation the campaign never leaves time for.
What the count says: 12 topics cover 16,740 enquiries out of 21,600, more than three in four. Bin not collected, bin size and exchange, collection calendar, opening hours and access to the two civic amenity sites, what goes in the yellow bin, bulky waste, composter, moving house, change of occupier, amount of the charge, payment methods, site access card. The 12 reference answers are written, sourced on your collection by-law and on the resolution of 27 November 2025, and dated — they are waiting for you.
What the three items you can check against your own logs are worth:
· A collection complaint takes 12 minutes end to end, 60% of which goes on finding the sector, the round day and on drafting — 7 minutes 12.
· A charge letter takes 25 minutes, 40% of it on the drafting itself — 10 minutes.
· Preparing a billing campaign is 100% manual work: 11 working days for your 6 case officers, i.e. 462 hours per campaign.
And the gap the reconciliation brought out, in the north sector alone — 1,240 households: 18 anomalies, never seen because nobody ever had the 11 days to look. 7 lifts with no active contract, 5 duplicate chips, 3 unreadable chips, 3 standing charges billed to an occupier who had left. In euros: €612.00 you did not bill and €966.00 billed wrongly or to the wrong payer.
The next step, and it takes twenty minutes: you reread the 18 lines and tell me, for each one, the action you want. The head of service settles the version, and the 18 come out of the file before the next campaign. waste-service-flow_21600-enquiries-12-topics.pdf16,740 enquiries across 12 topics, the three items costed, 3,371 hours identified
⛓ Sourced · 12 months of enquiry log, charge-payer file, lift log, tariff resolution of 27 November 2025
What the count says: 12 topics cover 16,740 enquiries out of 21,600, more than three in four. Bin not collected, bin size and exchange, collection calendar, opening hours and access to the two civic amenity sites, what goes in the yellow bin, bulky waste, composter, moving house, change of occupier, amount of the charge, payment methods, site access card. The 12 reference answers are written, sourced on your collection by-law and on the resolution of 27 November 2025, and dated — they are waiting for you.
What the three items you can check against your own logs are worth:
· A collection complaint takes 12 minutes end to end, 60% of which goes on finding the sector, the round day and on drafting — 7 minutes 12.
· A charge letter takes 25 minutes, 40% of it on the drafting itself — 10 minutes.
· Preparing a billing campaign is 100% manual work: 11 working days for your 6 case officers, i.e. 462 hours per campaign.
And the gap the reconciliation brought out, in the north sector alone — 1,240 households: 18 anomalies, never seen because nobody ever had the 11 days to look. 7 lifts with no active contract, 5 duplicate chips, 3 unreadable chips, 3 standing charges billed to an occupier who had left. In euros: €612.00 you did not bill and €966.00 billed wrongly or to the wrong payer.
The next step, and it takes twenty minutes: you reread the 18 lines and tell me, for each one, the action you want. The head of service settles the version, and the 18 come out of the file before the next campaign. waste-service-flow_21600-enquiries-12-topics.pdf16,740 enquiries across 12 topics, the three items costed, 3,371 hours identified
⛓ Sourced · 12 months of enquiry log, charge-payer file, lift log, tariff resolution of 27 November 2025
18 anomalies in a single sector — what does that mean across 12,400 households? And above all: how do we avoid doing it all again next year?
Across the 12,400 households, the same reconciliation rule brings out 180 per half-year — 360 over the year. In euros: roughly €6,120 not billed and €9,660 billed wrongly per campaign, i.e. €12,240 and €19,320 over the year. The amount is not the main point: each of those 360 lines is a resident complaint waiting to happen, and a complaint costs 12 minutes plus the trust.
What I did on top, and what the campaign never leaves time for: I wrote four control rules, in the language of your own by-law, and ran them over your last 24 months of lifts. For each one I give you the number of alerts it would have produced, the share confirmed after examination, and what it would have let through — you choose on figures, not on a hunch:
· Rule A — a chip read when no contract is active at the address on the day of the lift: 412 alerts over 24 months, 389 confirmed — 94.4%. The other 23 are contracts entered late, which is fixed at data entry.
· Rule B — the same chip read twice less than two hours apart on two different rounds: 196 alerts, 183 confirmed — 93.3%.
· Rule C — a standing charge billed to a payer whose departure the service has recorded: 148 alerts, 148 confirmed — 100%.
Total: 756 alerts over 24 months, 720 confirmed — 360 a year, exactly the figure from the reconciliation.
And here is what those three rules let through, because you need to know it before you sign: 68 lifts recorded at the flat rate for want of a readable chip fall under none of the three. So I wrote rule D — a lift recorded with no readable bin identifier — and ran it over the same 24 months: 68 alerts, 61 confirmed. The four rules together cover 781 of the 824 alerts raised over the period — 94.8%.
Bringing a rule into force stays with the service: a rule only goes live once approved by the head of service, and that is precisely what makes it defensible against a resident who disputes their invoice. I save you the writing and the measuring; the decision takes twenty minutes instead of a committee. charge-anomalies_4-rules-tested-over-24-months.pdfThe 18 north-sector anomalies in detail, 4 rules tested, 781 of 824 alerts confirmed
⛓ Sourced · 24 months of lift log (1,187,400 lifts), charge-payer file, collection by-law
What I did on top, and what the campaign never leaves time for: I wrote four control rules, in the language of your own by-law, and ran them over your last 24 months of lifts. For each one I give you the number of alerts it would have produced, the share confirmed after examination, and what it would have let through — you choose on figures, not on a hunch:
· Rule A — a chip read when no contract is active at the address on the day of the lift: 412 alerts over 24 months, 389 confirmed — 94.4%. The other 23 are contracts entered late, which is fixed at data entry.
· Rule B — the same chip read twice less than two hours apart on two different rounds: 196 alerts, 183 confirmed — 93.3%.
· Rule C — a standing charge billed to a payer whose departure the service has recorded: 148 alerts, 148 confirmed — 100%.
Total: 756 alerts over 24 months, 720 confirmed — 360 a year, exactly the figure from the reconciliation.
And here is what those three rules let through, because you need to know it before you sign: 68 lifts recorded at the flat rate for want of a readable chip fall under none of the three. So I wrote rule D — a lift recorded with no readable bin identifier — and ran it over the same 24 months: 68 alerts, 61 confirmed. The four rules together cover 781 of the 824 alerts raised over the period — 94.8%.
Bringing a rule into force stays with the service: a rule only goes live once approved by the head of service, and that is precisely what makes it defensible against a resident who disputes their invoice. I save you the writing and the measuring; the decision takes twenty minutes instead of a committee. charge-anomalies_4-rules-tested-over-24-months.pdfThe 18 north-sector anomalies in detail, 4 rules tested, 781 of 824 alerts confirmed
⛓ Sourced · 24 months of lift log (1,187,400 lifts), charge-payer file, collection by-law
We bill 12,400 households. Addresses, household composition, bank details: where does that data go?
Nowhere outside your walls, and that is verifiable rather than promised.
Where I run: on an isolated resource hosted in France, under French law, dedicated to Trois-Vallons and to nobody else — or on a machine in your own depot if you prefer local inference, in which case nothing even leaves the network. No resident data is passed to a foreign service: exposure to the Cloud Act and FISA 702 is reduced accordingly, without location alone ruling it out — it is documented for the configuration chosen.
What I read, and what I cannot reach: I have read-only access to the charge-payer file, the lift log and the collection calendar. I have no write access to the billing software: I produce files and draft letters, your case officer imports them. Bank details are not open to me at all — I have no use for them in preparing an invoice, and article 5.1.c GDPR asks precisely that only what serves the purpose be exposed. Access is withdrawn with a word, and withdrawal takes effect within the second.
What you can show your data protection officer: every read is logged with its date, its reason and the line concerned; every output carries the sources it was drawn from; and the record of processing activities is already written, legal basis by legal basis — public interest task for the charge, article 6.1.e.
The next step I propose: a thirty-minute quarterly review in which I set out the access rights actually used and those that served no purpose — an unused right of access is a right to close, and I bring you the list rather than asking you for it. sovereign-framework_hosting-access-traceability.pdfWhere the data lives, who reads what, what is withdrawn with a word
✎ Framework · hosting architecture, access log, record of processing activities
Where I run: on an isolated resource hosted in France, under French law, dedicated to Trois-Vallons and to nobody else — or on a machine in your own depot if you prefer local inference, in which case nothing even leaves the network. No resident data is passed to a foreign service: exposure to the Cloud Act and FISA 702 is reduced accordingly, without location alone ruling it out — it is documented for the configuration chosen.
What I read, and what I cannot reach: I have read-only access to the charge-payer file, the lift log and the collection calendar. I have no write access to the billing software: I produce files and draft letters, your case officer imports them. Bank details are not open to me at all — I have no use for them in preparing an invoice, and article 5.1.c GDPR asks precisely that only what serves the purpose be exposed. Access is withdrawn with a word, and withdrawal takes effect within the second.
What you can show your data protection officer: every read is logged with its date, its reason and the line concerned; every output carries the sources it was drawn from; and the record of processing activities is already written, legal basis by legal basis — public interest task for the charge, article 6.1.e.
The next step I propose: a thirty-minute quarterly review in which I set out the access rights actually used and those that served no purpose — an unused right of access is a right to close, and I bring you the list rather than asking you for it. sovereign-framework_hosting-access-traceability.pdfWhere the data lives, who reads what, what is withdrawn with a word
✎ Framework · hosting architecture, access log, record of processing activities
A resident reports that their yellow bin was not collected on rue des Tilleuls, and asks for a larger bin for a household of five.
Request logged and qualified, both parts handled in 90 seconds.
The collection complaint: rue des Tilleuls, in Marnaville, is in sector 3, whose packaging round runs on Tuesdays. The round of Tuesday the 12th did take place — the log shows 486 lifts across the sector — but this bin's chip was not read. Two possible causes, and I give you both: bin put out after the lorry passed, or a faulty chip. The complaint goes to the technical services with the sector, the round, the time the lorry passed the address and the history of this bin's last 6 lifts — that last point is the one that settles it, and nobody has time to go and find it by hand.
The allocation: your grid, annex 2 to the resolution of 27 November 2025, provides 120 L up to 2 people, 180 L for 3 to 4, and 240 L from 5. For a household of five, that is a 240 L bin — the exchange request form is pre-filled with the identity, the collection point address, the current chip and the size requested. One document is missing, and I have already put it in the letter: proof of household composition.
Reply to the resident ready to send, in plain French, with the date of the next round and the lead time for the new bin.
What this is worth over the year: your 3,940 collection complaints go from 7 minutes 12 of searching and drafting to 1 minute 12 — 60% of the time on a request brought down to 10%, i.e. 394 hours given back to the service.
The next step I propose: that I flag to you every Monday the bins whose chip has gone unread three rounds running. Over the last 24 months, 214 bins are in that position — 187 were dead chips, and the resident noticed before you did in 141 cases. Spotting it before the resident does is 141 complaints that never get sent. complaint-and-allocation_rue-des-tilleuls-file.pdfThe qualified complaint, the pre-filled 240 L bin request, the reply to the resident
⛓ Sourced · collection by-law, resolution of 27 November 2025 (annex 2 — allocation grid), sector 3 lift log
The collection complaint: rue des Tilleuls, in Marnaville, is in sector 3, whose packaging round runs on Tuesdays. The round of Tuesday the 12th did take place — the log shows 486 lifts across the sector — but this bin's chip was not read. Two possible causes, and I give you both: bin put out after the lorry passed, or a faulty chip. The complaint goes to the technical services with the sector, the round, the time the lorry passed the address and the history of this bin's last 6 lifts — that last point is the one that settles it, and nobody has time to go and find it by hand.
The allocation: your grid, annex 2 to the resolution of 27 November 2025, provides 120 L up to 2 people, 180 L for 3 to 4, and 240 L from 5. For a household of five, that is a 240 L bin — the exchange request form is pre-filled with the identity, the collection point address, the current chip and the size requested. One document is missing, and I have already put it in the letter: proof of household composition.
Reply to the resident ready to send, in plain French, with the date of the next round and the lead time for the new bin.
What this is worth over the year: your 3,940 collection complaints go from 7 minutes 12 of searching and drafting to 1 minute 12 — 60% of the time on a request brought down to 10%, i.e. 394 hours given back to the service.
The next step I propose: that I flag to you every Monday the bins whose chip has gone unread three rounds running. Over the last 24 months, 214 bins are in that position — 187 were dead chips, and the resident noticed before you did in 141 cases. Spotting it before the resident does is 141 complaints that never get sent. complaint-and-allocation_rue-des-tilleuls-file.pdfThe qualified complaint, the pre-filled 240 L bin request, the reply to the resident
⛓ Sourced · collection by-law, resolution of 27 November 2025 (annex 2 — allocation grid), sector 3 lift log
The real problem is the phone. In the week after billing, we simply stop picking up. One resident in three gets nothing.
4,128 calls out of 12,900 ring unanswered — 32%, and 2,970 of them fall in the five days after the invoices go out. This is not a quality-of-service problem, it is a timing problem: your 6 case officers face eight calls at once between 9 and 11 in the morning.
What I take, and what I leave you: I pick up every call, I state in the first sentence that I am an artificial intelligence — article 50(1) of the EU AI Act has required this since 2 August 2026, and the resident can ask for a human at any moment —, and I handle end to end the 12 topics whose answers are approved. Anything outside those 12 topics goes to a case officer with the file already assembled: the household, the lift history, the last letter sent, the question asked word for word. Your case officer picks up knowing, instead of picking up and searching.
The arithmetic, and it is yours: the 12 topics account for 77% of the flow; 4,128 unanswered calls fall to 258, i.e. 3,870 more residents who get their answer — 2,806 of them at night, at the weekend or over lunch, when your office is shut.
And I reach them on the channels they already use: WhatsApp Business, the website chat and email, with no account to create and nothing to install. Your case officers talk to me from Microsoft Teams or their mailbox, without changing tools. Email, which today waits 9 days, goes back to 1 day.
The figure that does not flatter me, and I publish it: of the 985 complaints in the first quarter, I routed 74 to the wrong sector — 7.5%. All of them concerned the 3 municipalities that changed collection day on 1 February: I was reading the calendar as it stands today, not as it stood on the day of the lift. That is fixed: I now read the calendar as it was on the day the lorry passed. In the second quarter, 6 routing errors out of 985 — 0.6%. I propose to keep publishing that rate every quarter: an agent whose errors are not measured is an agent that cannot be corrected. resident-channels_multichannel-desk-sms-plain-language.pdfThe 12 topics handled end to end, how the rest is routed, the published error rate
⛓ Sourced · 12 months of switchboard log, agent routing log, published collection calendar
What I take, and what I leave you: I pick up every call, I state in the first sentence that I am an artificial intelligence — article 50(1) of the EU AI Act has required this since 2 August 2026, and the resident can ask for a human at any moment —, and I handle end to end the 12 topics whose answers are approved. Anything outside those 12 topics goes to a case officer with the file already assembled: the household, the lift history, the last letter sent, the question asked word for word. Your case officer picks up knowing, instead of picking up and searching.
The arithmetic, and it is yours: the 12 topics account for 77% of the flow; 4,128 unanswered calls fall to 258, i.e. 3,870 more residents who get their answer — 2,806 of them at night, at the weekend or over lunch, when your office is shut.
And I reach them on the channels they already use: WhatsApp Business, the website chat and email, with no account to create and nothing to install. Your case officers talk to me from Microsoft Teams or their mailbox, without changing tools. Email, which today waits 9 days, goes back to 1 day.
The figure that does not flatter me, and I publish it: of the 985 complaints in the first quarter, I routed 74 to the wrong sector — 7.5%. All of them concerned the 3 municipalities that changed collection day on 1 February: I was reading the calendar as it stands today, not as it stood on the day of the lift. That is fixed: I now read the calendar as it was on the day the lorry passed. In the second quarter, 6 routing errors out of 985 — 0.6%. I propose to keep publishing that rate every quarter: an agent whose errors are not measured is an agent that cannot be corrected. resident-channels_multichannel-desk-sms-plain-language.pdfThe 12 topics handled end to end, how the rest is routed, the published error rate
⛓ Sourced · 12 months of switchboard log, agent routing log, published collection calendar
And the site closed on a bank holiday, or the collection moved: people never see it. We get an avalanche of complaints the next day.
612 of your 3,940 collection complaints fall in the 12 days following your 4 rescheduled collections — 15.5% of the annual flow on 12 days of the year. It is the most concentrated item in your service, and the one that is most simply solved.
What I propose, costed: a text message the day before, to the households of the affected sectors only, based on the resolved calendar. 4 reschedules a year, 2 sectors affected each time, 2,480 households per sector: 19,840 messages over the year, rebilled at the gateway's actual cost, with no margin. A text arrives on an ordinary handset: no email address, no account, no app and no data plan — it is the channel that reaches the households an online form never reaches, and that is as much about avoiding non-take-up as about avoiding complaints.
What the message contains, and nothing else: the municipality, the sector, the old day, the new day. Never an amount, never a payment link, never an identifier. It is an outbound notification: the resident does not reply to it, they reach me on WhatsApp, the chat or email.
What it is worth, on your own figures: the 3 authorities I have as references for this arrangement are not you, so I give you the measurement you will be able to check yourself: over the 12 days concerned, 612 complaints today. I commit to the protocol, not to the result: we measure the first reschedule without texts and the second with them, on the same sectors, and you decide on the gap.
And while we are here, the other end of the subject: your characterisation of 14 March 2026 shows 22.4% rejects on the packaging stream. I have rewritten the 6 sorting instructions that explain 71% of the rejects recorded — in plain language, short sentences, one idea per sentence, following easy-to-read-and-understand practice —, and each carries its current wording alongside, so that the approval is given on the gap and not on my word. The plain-language draft is put to you for approval, as the rule requires: a plain-language version is approved, it is not inferred. It serves the 1,460 site enquiries, the chat, the counter and the bin sticker — the same dated version on all four. resident-channels_multichannel-desk-sms-plain-language.pdfThe text message word for word, the measurement protocol, the 6 instructions rewritten in plain language
⛓ Sourced · resolved collection calendar, sorting-centre characterisation of 14 March 2026, 5,400 emails over the year
What I propose, costed: a text message the day before, to the households of the affected sectors only, based on the resolved calendar. 4 reschedules a year, 2 sectors affected each time, 2,480 households per sector: 19,840 messages over the year, rebilled at the gateway's actual cost, with no margin. A text arrives on an ordinary handset: no email address, no account, no app and no data plan — it is the channel that reaches the households an online form never reaches, and that is as much about avoiding non-take-up as about avoiding complaints.
What the message contains, and nothing else: the municipality, the sector, the old day, the new day. Never an amount, never a payment link, never an identifier. It is an outbound notification: the resident does not reply to it, they reach me on WhatsApp, the chat or email.
What it is worth, on your own figures: the 3 authorities I have as references for this arrangement are not you, so I give you the measurement you will be able to check yourself: over the 12 days concerned, 612 complaints today. I commit to the protocol, not to the result: we measure the first reschedule without texts and the second with them, on the same sectors, and you decide on the gap.
And while we are here, the other end of the subject: your characterisation of 14 March 2026 shows 22.4% rejects on the packaging stream. I have rewritten the 6 sorting instructions that explain 71% of the rejects recorded — in plain language, short sentences, one idea per sentence, following easy-to-read-and-understand practice —, and each carries its current wording alongside, so that the approval is given on the gap and not on my word. The plain-language draft is put to you for approval, as the rule requires: a plain-language version is approved, it is not inferred. It serves the 1,460 site enquiries, the chat, the counter and the bin sticker — the same dated version on all four. resident-channels_multichannel-desk-sms-plain-language.pdfThe text message word for word, the measurement protocol, the 6 instructions rewritten in plain language
⛓ Sourced · resolved collection calendar, sorting-centre characterisation of 14 March 2026, 5,400 emails over the year
Prepare this half-year's pay-as-you-throw billing for the households in the north sector.
Preparation complete. 1,240 households: standing charge and lifts counted according to your tariff parameters — €148 annual standing charge, i.e. €74 for the half-year, plus €2.40 per lift beyond the 12 lifts included in the year, i.e. 6 for the half-year, as set by the resolution of 27 November 2025. 18 anomalies detected and set aside for checking — they do not go out on invoices, they go to control.
What every line carries, and this is the rule I propose you hold to: every billed line traces back to a time-stamped lift and to a resolved tariff. A line missing either of the two does not go out. That is what lets you produce, in front of a resident who disputes, the date, the time, the chip and the article of the resolution — instead of a total.
The campaign in detail, north sector: 1,222 invoices ready, 18 lines in checking, total billed €118,934.40, a 1.2% variance against the previous half-year, entirely explained by 27 move-ins and 19 departures — I give you the 46 movements, you will not have to hunt for where the variance came from.
What this is worth over the full campaign: 12,400 households, 11 working days for six today — 462 hours. Afterwards: 92 hours of checking and approval, i.e. 20% of the current time, a little over two working days for six. 370 hours given back per campaign, 739 hours over the year.
The covering letters are ready too — 3 templates: first invoice after moving in, invoice up more than 30%, invoice with exceptional lifts. A charge letter used to take 25 minutes, 10 of them drafting; it now takes 2 minutes 30. Across 6,300 letters, 787 hours given back.
Nothing is issued: the billing file and the 1,222 letters are waiting for the head of service to approve them, and I can undo the lot with a word for as long as nothing has gone out.
The next step I propose: open the full campaign across the 5 sectors, with the 4 control rules live. You recover half the campaign time on this one, and all of it on the next. billing-campaign_north-sector-1240-households.pdf1,222 invoices ready, 18 lines in checking, the 3 letter templates
✎ Action · billing prepared — the head of service approves before issue · resolution of 27 November 2025
What every line carries, and this is the rule I propose you hold to: every billed line traces back to a time-stamped lift and to a resolved tariff. A line missing either of the two does not go out. That is what lets you produce, in front of a resident who disputes, the date, the time, the chip and the article of the resolution — instead of a total.
The campaign in detail, north sector: 1,222 invoices ready, 18 lines in checking, total billed €118,934.40, a 1.2% variance against the previous half-year, entirely explained by 27 move-ins and 19 departures — I give you the 46 movements, you will not have to hunt for where the variance came from.
What this is worth over the full campaign: 12,400 households, 11 working days for six today — 462 hours. Afterwards: 92 hours of checking and approval, i.e. 20% of the current time, a little over two working days for six. 370 hours given back per campaign, 739 hours over the year.
The covering letters are ready too — 3 templates: first invoice after moving in, invoice up more than 30%, invoice with exceptional lifts. A charge letter used to take 25 minutes, 10 of them drafting; it now takes 2 minutes 30. Across 6,300 letters, 787 hours given back.
Nothing is issued: the billing file and the 1,222 letters are waiting for the head of service to approve them, and I can undo the lot with a word for as long as nothing has gone out.
The next step I propose: open the full campaign across the 5 sectors, with the 4 control rules live. You recover half the campaign time on this one, and all of it on the next. billing-campaign_north-sector-1240-households.pdf1,222 invoices ready, 18 lines in checking, the 3 letter templates
✎ Action · billing prepared — the head of service approves before issue · resolution of 27 November 2025
Bin requests and exemptions are what eats our dead time: we chase documents for three weeks.
2,520 files a year — 2,180 bin allocation or exchange requests and 340 exemption requests — and today the documents are checked at the assessment stage, that is to say three weeks too late.
What I do at submission, within the minute: I read the proof of address and the household composition statement attached to the request, and I say what is missing before the file joins the queue. Across last year's 2,520 files run back through this rule: 418 were incomplete — 16.6%. Document absent in 212 cases, proof of address older than three months in 121 cases, and in 85 cases a household composition that does not match the bin size requested — those 85 are worth their weight: they are 85 240 L bins that would have gone to two-person households, and 85 exchanges to redo.
What I read and what I keep: I take the name, the address, the date of the document and the number of people. The rest of the document is of no interest to me and is not kept — a tax notice carries far more than a bin allocation needs, and article 5.1.c GDPR asks precisely that only the useful part be retained. That is an argument your data protection officer can reuse as it stands.
What it produces: the lead time for a bin goes from 21 days to 6, because today's three weeks are three weeks of documents going back and forth.
And here is where you need to give me a mandate, because it commits you: issuing an invoice, granting an exemption or ordering a bin from the supplier are acts that bind the authority. These are not acts I rule out — they are acts I carry out as soon as the mandate is written: named scope, costed ceiling, review date, and withdrawal with a word. The model mandate is drafted, with three scopes to choose from — from the most cautious (nothing goes out without line-by-line approval) to the widest (allocations that match the grid go out on their own, the rest wait). You choose where the dial sits, and you can move it both ways at any time. document-checks_2520-files-and-model-mandate.pdf418 incomplete files caught at submission, what is read and what is kept, the three-position mandate
⛓ Sourced · 2,520 files from the year run back through the control rule, collection by-law, resolution of 27 November 2025
What I do at submission, within the minute: I read the proof of address and the household composition statement attached to the request, and I say what is missing before the file joins the queue. Across last year's 2,520 files run back through this rule: 418 were incomplete — 16.6%. Document absent in 212 cases, proof of address older than three months in 121 cases, and in 85 cases a household composition that does not match the bin size requested — those 85 are worth their weight: they are 85 240 L bins that would have gone to two-person households, and 85 exchanges to redo.
What I read and what I keep: I take the name, the address, the date of the document and the number of people. The rest of the document is of no interest to me and is not kept — a tax notice carries far more than a bin allocation needs, and article 5.1.c GDPR asks precisely that only the useful part be retained. That is an argument your data protection officer can reuse as it stands.
What it produces: the lead time for a bin goes from 21 days to 6, because today's three weeks are three weeks of documents going back and forth.
And here is where you need to give me a mandate, because it commits you: issuing an invoice, granting an exemption or ordering a bin from the supplier are acts that bind the authority. These are not acts I rule out — they are acts I carry out as soon as the mandate is written: named scope, costed ceiling, review date, and withdrawal with a word. The model mandate is drafted, with three scopes to choose from — from the most cautious (nothing goes out without line-by-line approval) to the widest (allocations that match the grid go out on their own, the rest wait). You choose where the dial sits, and you can move it both ways at any time. document-checks_2520-files-and-model-mandate.pdf418 incomplete files caught at submission, what is read and what is kept, the three-position mandate
⛓ Sourced · 2,520 files from the year run back through the control rule, collection by-law, resolution of 27 November 2025
How do we bring your four rules into service? And what if one of them starts flagging nonsense?
One at a time, each with its entry figure, and none starts without your signature.
The order I recommend, and the arithmetic behind it: rule C first — 148 alerts over 24 months, 100% confirmed, so no examination time wasted; then A — 412 alerts, 94.4% confirmed, the largest volume of unbilled income; then B and D. I advise against switching all four on the same day, and I cost it: the four together would produce 378 alerts on the first campaign, i.e. around 63 hours of examination for your 6 case officers, in the middle of a campaign. One per campaign is 15 hours each time, and you keep control of what you look at.
What I watch for you, without being asked: the confirmed share of each rule, campaign after campaign. A rule that drops below 80% confirmation is a rule that makes you examine households for nothing: I warn you from the first campaign concerned, with the tightened version already written and already run over the history, as I did for rule B — its first drafting produced 341 alerts for 183 confirmed (53.7%); the version tightened to a two-hour interval produces 196 for the same 183 confirmed, all 183 genuine cases kept, 145 households spared a pointless look. It is the tightened version I am putting to you, not the first.
And withdrawal is as simple as bringing it in: a rule stops with a word, within the second, and the lines it had put into checking go back into billing with the reason and the date of the withdrawal. Nothing is blocked for good, nothing disappears. charge-anomalies_4-rules-tested-over-24-months.pdfThe go-live schedule, the tightened version of rule B, the withdrawal threshold
⛓ Sourced · 24 months of lift log, comparative measurement of the two draftings of rule B
The order I recommend, and the arithmetic behind it: rule C first — 148 alerts over 24 months, 100% confirmed, so no examination time wasted; then A — 412 alerts, 94.4% confirmed, the largest volume of unbilled income; then B and D. I advise against switching all four on the same day, and I cost it: the four together would produce 378 alerts on the first campaign, i.e. around 63 hours of examination for your 6 case officers, in the middle of a campaign. One per campaign is 15 hours each time, and you keep control of what you look at.
What I watch for you, without being asked: the confirmed share of each rule, campaign after campaign. A rule that drops below 80% confirmation is a rule that makes you examine households for nothing: I warn you from the first campaign concerned, with the tightened version already written and already run over the history, as I did for rule B — its first drafting produced 341 alerts for 183 confirmed (53.7%); the version tightened to a two-hour interval produces 196 for the same 183 confirmed, all 183 genuine cases kept, 145 households spared a pointless look. It is the tightened version I am putting to you, not the first.
And withdrawal is as simple as bringing it in: a rule stops with a word, within the second, and the lines it had put into checking go back into billing with the reason and the date of the withdrawal. Nothing is blocked for good, nothing disappears. charge-anomalies_4-rules-tested-over-24-months.pdfThe go-live schedule, the tightened version of rule B, the withdrawal threshold
⛓ Sourced · 24 months of lift log, comparative measurement of the two draftings of rule B
The community council wants a twelve-month report from me. What do I put in front of them?
Four figures, and each one checks out against your own logs — that is the only way to hold up in front of a council.
· 3,371 hours given back to the service over the year, item by item: 394 h on collection complaints, 787 h on charge letters, 739 h on the two billing campaigns, 1,451 h on routine enquiries. At 1,607 hours to a working year, that is more than two working years given back to your eight staff. No post cut, no post created: it is time given back to the ground, to the civic amenity sites and to the residents who need a person in front of them.
· 4,128 unanswered calls down to 258, and email from 9 days to 1 day.
· 360 billing anomalies taken out of the file over the year, where the reconciliation was never done — €12,240 of income restored and €19,320 that did not go out wrongly.
· 0 invoices issued without a human approval, 0 tariffs changed, 0 waivers granted.
What I bring you on top, and what will be waiting for you every year: your annual report on the price and quality of the service — the one required by article L. 2224-17-1 of the general local authorities code — draws in part on data I already keep up to date. The 9 resident-relations indicators are extracted, dated, with the query that produces each of them, so that a councillor who asks where a figure comes from gets the answer in the chamber. The report itself stays written and presented by you: it is an act of the chair, not a machine output.
The next step I propose for the year ahead: the most expensive subject in your service remains sorting rejects — 22.4% as at 14 March 2026. I propose we measure, at the next characterisation, the gap between the 4 municipalities that will have received the plain-language instructions and the other 13. It is a protocol, not a promise: if the gap is nil, you will hear it from me. year-in-review_3371-hours-item-by-item.pdfThe 3,371 hours in detail, the lead times held, the 9 annual-report indicators
⛓ Sourced · 12 months of service logs, charge-payer file, lift log, characterisation of 14 March 2026
· 3,371 hours given back to the service over the year, item by item: 394 h on collection complaints, 787 h on charge letters, 739 h on the two billing campaigns, 1,451 h on routine enquiries. At 1,607 hours to a working year, that is more than two working years given back to your eight staff. No post cut, no post created: it is time given back to the ground, to the civic amenity sites and to the residents who need a person in front of them.
· 4,128 unanswered calls down to 258, and email from 9 days to 1 day.
· 360 billing anomalies taken out of the file over the year, where the reconciliation was never done — €12,240 of income restored and €19,320 that did not go out wrongly.
· 0 invoices issued without a human approval, 0 tariffs changed, 0 waivers granted.
What I bring you on top, and what will be waiting for you every year: your annual report on the price and quality of the service — the one required by article L. 2224-17-1 of the general local authorities code — draws in part on data I already keep up to date. The 9 resident-relations indicators are extracted, dated, with the query that produces each of them, so that a councillor who asks where a figure comes from gets the answer in the chamber. The report itself stays written and presented by you: it is an act of the chair, not a machine output.
The next step I propose for the year ahead: the most expensive subject in your service remains sorting rejects — 22.4% as at 14 March 2026. I propose we measure, at the next characterisation, the gap between the 4 municipalities that will have received the plain-language instructions and the other 13. It is a protocol, not a promise: if the gap is nil, you will hear it from me. year-in-review_3371-hours-item-by-item.pdfThe 3,371 hours in detail, the lead times held, the 9 annual-report indicators
⛓ Sourced · 12 months of service logs, charge-payer file, lift log, characterisation of 14 March 2026
In practice, what does the agent do without asking us? That is the question the council will ask.
Three actions, not one more, and all three can be undone.
· I acknowledge any collection complaint within the minute, with the sector, the round day and the date of the next pass. And the reverse is true too: if the resident writes to say the bin was collected after all, the complaint closes without any letter going out, and it stays in the statistics — because it is that statistic which will tell you if a round is slipping.
· I put into checking any billing line that triggers one of the live rules. And the reverse is true too: the case officer sets it aside with a word, the line goes back into billing, with the reason and the date it was set aside — which gives you, at the next campaign, the list of the most frequent reasons, and therefore the rules to tighten.
· I send the rescheduled-collection text to the sectors concerned, the day before, based on the resolved calendar. It can be pulled with a word for as long as it has not gone out, and a household that asks to stop receiving it is removed within the second — just as it is put back within the second if it asks again. Being removed is not a door closed for life.
Everything else waits for a named decision: no invoice is issued, no tariff is touched, no exemption is granted, no bin is ordered, no ex gratia waiver is proposed.
A word on a measurement I can produce and would advise you not to install as it stands: the number of files handled per case officer. I know how to calculate it, and I will produce it if you ask — a public employer's power to monitor is recognised, subject to three cumulative conditions, as the CNIL sets them out in its guidance on monitoring the activity of employed persons: proportionality of the indicator to the aim pursued, prior information of staff, and consultation of the local social committee before implementation. My reservation is not legal, it is mechanical: a per-officer counter becomes a target, the target distorts the triage, and the long files — a duplicate chip, an old dispute — get pushed back in favour of the short ones. You would lose the instrument at the same time as the measurement. What I propose instead, and it is already prepared: measurement by QUEUE — lead time, backlog, rework rate — which tells you where it jams without telling you who. One exception, and it is not really one: whoever approves an invoice is named and dated. A signature is not a counter. automatic-actions_three-acts-and-their-withdrawal.pdfThe 3 actions, how each is undone, what stays with a human decision
✎ Framework · settings for the automatic actions, withdrawal log, conditions for implementing an individual indicator
· I acknowledge any collection complaint within the minute, with the sector, the round day and the date of the next pass. And the reverse is true too: if the resident writes to say the bin was collected after all, the complaint closes without any letter going out, and it stays in the statistics — because it is that statistic which will tell you if a round is slipping.
· I put into checking any billing line that triggers one of the live rules. And the reverse is true too: the case officer sets it aside with a word, the line goes back into billing, with the reason and the date it was set aside — which gives you, at the next campaign, the list of the most frequent reasons, and therefore the rules to tighten.
· I send the rescheduled-collection text to the sectors concerned, the day before, based on the resolved calendar. It can be pulled with a word for as long as it has not gone out, and a household that asks to stop receiving it is removed within the second — just as it is put back within the second if it asks again. Being removed is not a door closed for life.
Everything else waits for a named decision: no invoice is issued, no tariff is touched, no exemption is granted, no bin is ordered, no ex gratia waiver is proposed.
A word on a measurement I can produce and would advise you not to install as it stands: the number of files handled per case officer. I know how to calculate it, and I will produce it if you ask — a public employer's power to monitor is recognised, subject to three cumulative conditions, as the CNIL sets them out in its guidance on monitoring the activity of employed persons: proportionality of the indicator to the aim pursued, prior information of staff, and consultation of the local social committee before implementation. My reservation is not legal, it is mechanical: a per-officer counter becomes a target, the target distorts the triage, and the long files — a duplicate chip, an old dispute — get pushed back in favour of the short ones. You would lose the instrument at the same time as the measurement. What I propose instead, and it is already prepared: measurement by QUEUE — lead time, backlog, rework rate — which tells you where it jams without telling you who. One exception, and it is not really one: whoever approves an invoice is named and dated. A signature is not a counter. automatic-actions_three-acts-and-their-withdrawal.pdfThe 3 actions, how each is undone, what stays with a human decision
✎ Framework · settings for the automatic actions, withdrawal log, conditions for implementing an individual indicator
And the day a resident disputes their invoice saying the machine decided it?
That day, you produce three items and the dispute settles on facts.
One — the line itself: the date, the time and the chip identifier of every billed lift, plus the article of the resolution of 27 November 2025 that sets the tariff applied. Every billed line traces back to a time-stamped lift and to a resolved tariff: that is the rule we have held since the first campaign, and it is what makes the invoice defensible.
Two — who decided: the name of the officer who approved the file, the date and the time of approval. No invoice was issued by the machine: I prepare, a human settles. Over the year, 0 invoices issued without a named approval — and the log shows it line by line.
Three — what the resident had been told, and when: I stated that I am an artificial intelligence in the first sentence of every exchange, in line with article 50(1) of the EU AI Act, applicable since 2 August 2026, and the resident could ask for a human at any moment — 1,184 did so over the year, and all 1,184 got one.
And if the resident wants to see what you hold on them: the access file is produced in a minute — household, contract, lifts, letters, and the log of the reads I made of their file with the reason for each. That is the right of access under article 15 GDPR, and the usual difficulty is gathering the pieces: they are already gathered.
The next step I propose: an annual thirty-minute review with your data protection officer, where I bring you the access rights that served no purpose over the year and the retention periods that have run out. You decide what to close and what to erase; I bring you the list rather than waiting to be asked for it. sovereign-framework_hosting-access-traceability.pdfWhat you produce on the day of a dispute: the line, the approver, the disclosure
✎ Framework · approval log, AI disclosure log, record of processing activities, resolution of 27 November 2025
One — the line itself: the date, the time and the chip identifier of every billed lift, plus the article of the resolution of 27 November 2025 that sets the tariff applied. Every billed line traces back to a time-stamped lift and to a resolved tariff: that is the rule we have held since the first campaign, and it is what makes the invoice defensible.
Two — who decided: the name of the officer who approved the file, the date and the time of approval. No invoice was issued by the machine: I prepare, a human settles. Over the year, 0 invoices issued without a named approval — and the log shows it line by line.
Three — what the resident had been told, and when: I stated that I am an artificial intelligence in the first sentence of every exchange, in line with article 50(1) of the EU AI Act, applicable since 2 August 2026, and the resident could ask for a human at any moment — 1,184 did so over the year, and all 1,184 got one.
And if the resident wants to see what you hold on them: the access file is produced in a minute — household, contract, lifts, letters, and the log of the reads I made of their file with the reason for each. That is the right of access under article 15 GDPR, and the usual difficulty is gathering the pieces: they are already gathered.
The next step I propose: an annual thirty-minute review with your data protection officer, where I bring you the access rights that served no purpose over the year and the retention periods that have run out. You decide what to close and what to erase; I bring you the list rather than waiting to be asked for it. sovereign-framework_hosting-access-traceability.pdfWhat you produce on the day of a dispute: the line, the approver, the disclosure
✎ Framework · approval log, AI disclosure log, record of processing activities, resolution of 27 November 2025
Local inference · no data outside the EU
Your case is not here? That is exactly what a 15-minute conversation is for. Book the free audit →
Use cases
The uses of AI in a waste department
Each use corresponds to an agent we deploy. All of them work in support, subject to approval by the public officer.
Included in your agent
The 6 capabilities essential to this promise are included, at no extra cost.
From 900 € incl. VAT / month
Replies to residents, across channels
Collection, bins, recycling centre, calendar: answer the everyday questions on every channel, 24/7. Dated changes — a collection moved, an exceptional closure — are announced by text message.
Billing of the waste charge
Prepare the billing of the pay-as-you-throw charge — lifts, drop-offs, fixed element — and the covering letters.
Bin allocation & reports
Tracks bin requests and collection-related reports, and routes them to the technical services team.
Check the documents behind the waste charge and bin requests
Check the proof of address and household composition attached to requests for a bin or an exemption.
Detecting billing anomalies
Spot lifts with no subscription, duplicate chips and unusual variances before the invoices are issued.
Plain language & easy-read draft
Preparation of a plain-language version and a draft easy-read (FALC) transcription, subject to human validation.
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
Sources, access rights and handling of questions with no answer
Maintain validation, transparency, audit trail and recourse
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 authorised tools and the existing sovereign foundation
Other needs our agents cover
Each card says where the matching agent stands: available, on quote, or still being architected.
Multichannel administrative information
RAG on procedures; web, phone and front desk; 24/7; multilingual. Sourced answers.
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.
Does your need fall outside this?
Book the free audit →
Build your agent
In 15 minutes we identify the agent that will give your staff the most time back — without oversizing the project.
The gain
How much time can a waste department win back?
By automating the replies to residents and the preparation of the billing, the authority can aim for a clear reduction in time spent on repetitive tasks — reinvested in operations and in the quality of the service.
Answering a collection complaint
A charge letter to a resident
Preparing the half-year billing
Qualitative, non-contractual comparison: the proportions shown illustrate the shift of the work towards review, they represent no measurement. Every output of the agent is reviewed and approved by a competent person.
How it works
The stages of your AI agent project
1
Audit & scoping
15 minutes to target the use case with the best return.
2
Quote or direct sign-up
A catalogue offer is bought online; a specific need gets a costed quote.
3
Design
We design the agent and its guardrails.
4
Integration & testing
We connect your tools to the agent, which is itself hosted in France.
5
Rollout
Going live and training your team.
6
Operation
Continuous supervision and improvement.
Pricing
Three options, one agent
A waste & charge agent (resident relations, billing, tracking requests), installed and operated for you. Choose according to how you are organised. Prices exclude VAT — available by direct award below the public procurement thresholds.
Agility
Setup + controlled subscription
10,660 € incl. VAT setup
then 900 € incl. VAT/month — you invest at installation and pay a reduced subscription. Ideal for keeping the cost under control over time.
- Installation, configuration and training for your teams
- Operation, human oversight, updates and support
- Sovereign hosting in France, a dedicated and isolated resource
The simplest
Serenity
All inclusive, no setup fee
1,495 € incl. VAT /month
all inclusive, immediate start. No upfront investment: a single subscription. Ideal for starting quickly and simply.
- Setup included (installation, configuration, training)
- Operation, human oversight, updates and support
- Sovereign hosting in France, managed end to end
100% Sovereign
On site, you own it
15,041 € incl. VAT setup
then 1,140 € incl. VAT/month · + hardware from 2,989 € (one-off purchase, in addition) — a sovereign computer installed on your premises, maintained remotely. Models run locally, your data returned at the end of the contract. 36-month commitment.
- Hardware installed on your premises (you own it)
- French / European AI models run locally
- Secure remote maintenance (Pro support included)
Not included in the packages: AI consumption (model tokens), re-invoiced at real cost with no margin, and tracked in real time in your client area. Maintenance and supervision subscription for an initial term of 12 months for the Agility package, 24 months for the Serenity package and 36 months for the 100% Sovereign package, renewable; support levels (SLA 72 h / 24 h / 4 h) optional. Bespoke development, additional integrations or exceptional volumes are quoted separately. Support Monday to Friday, 9am to 6pm. Prices include VAT at 20%: as a public body that is not VAT-registered, you cannot reclaim it.
AI model:
none of the AI models offered currently carries a fixed surcharge. When the selected model carries a cost, that cost is shown when you choose it, before you order, and re-invoiced at the cost incurred, with no mark-up; usage is billed at the publisher's price. Publishers' prices are published in US dollars: the amount re-invoiced is the amount in euros actually borne by Blue Lemon Agent on the publisher's invoice, at that invoice's exchange rate, with no commission or mark-up.
Included components and additional components Components included in the base offer: the Blue Lemon Agent software foundation, the AI models listed in the order journey, the standard channels (Microsoft Teams, Slack, WhatsApp Business, email, website chat, calendars, Microsoft 365 / Google Workspace, file storage, market VoIP telephony, professional social-media pages and accounts, Google Business Profile), hosting in France for the package chosen, backups, supervision, updates and support. If adapting the AI agent to your constraints, your needs or your requests requires other paid components — a third-party publisher's software licence, paid API access to one of your applications, hosting of health data, for which French law requires an HDS-certified host (art. L. 1111-8 of the French Public Health Code), SecNumCloud-qualified hosting, a speech synthesis service, particular hardware —, they are offered to you as an option or on quotation and re-invoiced at the cost incurred; nothing is committed without your written agreement. Where the artificial intelligence model you choose entails an additional cost, that cost is shown to you before you order and re-invoiced to you at the cost incurred, with no margin.
What to expect
Go-live
2 to 3 weeks
Agent designed, channels connected, team trained.
Steady state
4 to 7 weeks
After a few weeks of real use, once the agent's behaviour matches what you expect.
Indicative estimate, adjusted to the options you keep. It is not a delivery commitment.
Our commitment
Four guarantees that matter to a local authority
Residents' data never leaves the authorityLocal inference or an isolated resource hosted in France; no data entrusted to a foreign third party.
Data in France, under French lawExposure of residents' data to the Cloud Act and FISA 702 is reduced by design, that data being located and minimised in France; location alone does not guarantee immunity.
The public officer keeps the decisionThe agent prepares invoices, letters and tracking that can be checked; the pricing parameters stay set by the authority.
Human oversight & traceabilityMonitoring and logging frame the replies to residents across channels; compliant with the requirements of the AI Act.
Frequently asked questions
Your questions, our answers
Does the agent manage collection on the ground?
No: strictly administrative (resident relations, billing, tracking requests). Operations stay run by the teams.
Where is the data processed?
In France, on a resource isolated for each authority. GDPR: governed deployment.
Does the agent handle the pay-as-you-throw charge?
Yes: it tracks lifts and drop-offs and prepares the billing, but the pricing parameters stay set by the authority.
Does it integrate with our waste management software?
Yes: through secure integrations defined at the design stage.
Does the agent state that it is an artificial intelligence?
Yes, from the very first interaction, and this is not a configuration option: since 2 August 2026, Article 50(1) of the European AI Regulation requires that any person interacting with an AI system be informed, unless this is obvious. The announcement is built into the greeting, in the other party’s language, and they can ask for a human at any time.
How long does it take to deploy an agent?
A few weeks as a rule, after a free audit that identifies the most useful use case, then a phase of design, integration and testing before going live and handing over to the teams.
Do we need a technical team in-house?
No. The agent is designed, installed and operated by us: supervision, updates and maintenance are included. Your teams concentrate on their work.
Do we have to change software?
No. The agent complements your existing tools (waste management, billing), with no migration imposed. We adapt the integration to your authority's environment.
Which tools can users use to reach the agent?
The ones they already have. The agent answers on WhatsApp Business, the website chat and email: users have no account to create and no application to install. This is a lever for access to the service — WhatsApp and the telephone reach people an online form never does, which reduces the non-take-up of rights and serves equal access. Internally, your public-sector staff talk to the agent from Microsoft Teams, Slack or their email, without switching tools. Oversight runs from a web dashboard. These connectors rely on open standards, including the MCP protocol; they are included in every plan, at no extra cost, within the number of connections your level includes. Only the fees charged by the platforms themselves — WhatsApp Business bills per conversation, and the gateway bills the text messages sent — are passed on at actual cost, with no margin, outside the subscription.
Can the agent notify users by text message?
Yes, as an option — and it is the channel that reaches the people the others never do. A text message arrives on an ordinary phone: no email address, no account, no app, no data plan and no smartphone. Announcing by text message a collection moved, a recycling centre closed or a bin to put out the evening before a public holiday means fewer entitlements going unclaimed and more equal access to the service. It is an outbound notification channel, not a conversation channel: users do not write to the agent there — they reach you on WhatsApp Business, the website chat or email. The message announces without detailing — never the reason, never the detail of the situation — and never contains a payment link, a login or a password: users reach the service the way they usually do. Delivery goes through a gateway established in the European Union; the messages sent are passed on at actual cost, with no margin, and the channel itself is an optional module, set out on the pricing page.
Related roles
Other roles in local public service
Let's talk
Let us estimate the potential in your authority
A few minutes to identify the most useful use case — hosted in France, supervised, with no commitment.