An orchestration agent, scoped with you before it is priced. This agent coordinates several specialised agents. Its orchestration follows your actual workflows — which is why it is scoped with you rather than bought off the shelf. We establish the scope together, then the quotation commits it.
The journeys described below form the scope that this review refines and the quotation commits. The specialised agents it coordinates can be ordered today.
Request a quote
● Public sector — Public procurement & contracts
The AI agent for public buyers: prepare the contracts, make the analysis reliable
Drafting the documents, checking that files are complete, comparing tenders against objective criteria: these tasks take up a considerable share of public buyers' time — without being the heart of the decision. Your AI agent absorbs that repetitive work. Hosted in France — on local inference or an isolated resource — public procurement data stays under control. The public officer decides; the award remains a human decision. This page describes the public buyer's job; the product that serves it is the canonical agent “AI agent for public procurement” — same agent, same pricing identifier, a single price grid.
Hosted in France
Contract data protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Public procurement assistant · hosted in France
On the “Roadworks 2026” consultation, check the completeness of the 6 bids received before the committee opens them.
6 bids analysed. 4 complete. 2 to put right: one bidder without a tax and social security compliance certificate, another whose technical submission does not cover lot 2. No fatally missing document at the application stage.
Completeness table ready — for approval before the analysis report.
⛓ Source · consultation rules + documents submitted on the buyer portal
Completeness table ready — for approval before the analysis report.
⛓ Source · consultation rules + documents submitted on the buyer portal
Compare the admissible tenders on the price criterion and the technical merit criterion.
Comparison ready: 4 tenders scored on the grid in the rules (price 40%, technical merit 60%). The best-value tender is not the cheapest. 1 potentially abnormally low tender (−32% below the estimate): a request for justification should be made before any ranking.
I am preparing a draft analysis report for your review — ranking and award rest with the committee.
✎ Action · analysis report ready for review — the buyer and the committee decide
I am preparing a draft analysis report for your review — ranking and award rest with the committee.
✎ Action · analysis report ready for review — the buyer and the committee decide
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 awards alone
✦ In brief
In a purchasing or public procurement department, a Blue Lemon Agent agent helps draft specifications, technical specifications and consultation rules, checks the completeness of applications and tenders and compares bids against objective criteria for human analysis. It runs on local inference or is hosted in France: public procurement data is never exposed to a foreign service, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. The time saved is redirected towards steering and the legal soundness of procedures. The award remains a decision of the committee and the buyer. Live within a few weeks. This page is the ROLE page for the public buyer: the canonical product that serves it is the AI agent for public procurement, whose pricing identifier it shares. It opens neither a second price grid nor a second SKU.
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
8
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 public buyers — and why they hesitate
Public procurement is getting more complex: shifting thresholds, transparency requirements, innovation purchasing, environmental and social clauses. But available time is mechanically absorbed by drafting the documents and checking the bids — and the data involved bears on equal treatment between bidders.
! The issue
The purchasing department is caught between increasingly regulated procedures (transparency, competitive tendering, reasoned decisions) and an ever-growing production workload (drafting documents, completeness checks, tender analysis). Yet most consumer AI solutions amount to entrusting bidders' tenders, prices, technical submissions and the authority's data to a third party, often hosted outside Europe and subject to the Cloud Act.
✓ Our answer
AI is only of interest to a public buyer if it is sovereign and confidential by design. Local inference or an isolated resource hosted in France, systematic human oversight, decisions reserved to the buyer and the committee: the time saved on preparation is never paid for in lost confidentiality or compromised equal treatment. The aim is not to replace the buyer, but to make their procedures more reliable and give them time back.
The decisive point
Public procurement data: sovereignty & compliance
A contracts department handles confidential tenders, prices and data bearing on equal treatment. Here is how the architecture of our agents protects it, consultation by consultation.
Local inference
The agent can run on a machine belonging to the authority: no bid leaves the network, nothing passes through a cloud.
Hosting in France
Otherwise, a dedicated and isolated resource, hosted in France under French law — your data: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
The tenders: architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
One isolated resource per entity
No pooling of tenders: an environment strictly dedicated to your purchasing department.
Encryption & controlled access
Encryption in transit and at rest, role-based access (RBAC), strong authentication and logging.
AI Act: governed deployment
An agent strictly in support; no automated award; 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 options are available depending on your requirements. A single architecture is designed to answer both the GDPR and extraterritorial exposure. Designed for deployment in line with the GDPR and the AI Act, after the processing, roles and context-specific risks have been assessed.
Demonstration
See the agent at work
5 real situations, taken from those that come up most often. Pick one: the exchange unfolds as it would in your organisation.
A scripted demonstration. These exchanges show how the agent behaves — its sources, its refusals, what it leaves to your teams. Nothing is sent from this page, no model is queried here, and the matters named are fictional. That is precisely what we promise your data.
The behaviours shown here — monitoring, automation rules, routing and reminders — are configured with you during deployment, from your tools, your rules and your thresholds.
The architecture points named in these exchanges — location, local execution, isolation, encryption, role-based access, logging — are not a guarantee attached to the demonstration: they are those of the architecture set out in your quotation, and verified before commissioning.
The public body in this demonstration
Fictional public bodyRoche-Vallon Urban Community — 18 municipalities, 92,000 residents
- Sector
- Urban community — water and wastewater, waste collection, community roads, sports facilities, urban transport; a procurement service shared by the 18 member municipalities
- Headcount
- 640 staff, of whom a procurement service of 4: 2 buyers, a procedures officer and a head of service; no post dedicated to analysing tenders
- Public served
- 92,000 residents, 18 member municipalities and 26 buying departments that express their needs
- Order of magnitude
- 118 procurements a year — 74 adapted procedures, 31 open procedures, 13 framework agreements with purchase orders; 512 tenders received; 214 contracts in performance; €38m of public spending committed
- Tools in place
- Buyer profile platform for publication and tender submission, financial management software, document management system and 9 years of procurement files filed by year — the agent connects in read-only, nothing is replaced or migrated
- Who decides what
- The president signs the contracts, and contracts below the threshold go out under the delegation of signature of the vice-president for finance; the tender board awards above the formal thresholds; the head of procurement settles every tender file before publication
- The points to improve
- The completeness check takes 60 % of the opening work on a procurement — 6 hours out of 10; 9 procurements last year received a single tender; 41 candidate questions went unanswered within the deadline set by the consultation rules; and 11 of the service's 34 standard documents refer to earlier wording
At Roche-Vallon, four people run 118 procurements a year for 26 buying departments and 18 municipalities. Time goes into checking documents and formatting analyses, before the first comparison of tenders is written. The agent runs in local inference on a machine of the authority, reads the nine years of procurement files without ever writing to them, and sends nothing out without the validation of the head of procurement. The exchanges below cover one year, from the rewriting of the standard documents to the review presented to the community board.
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.
Public procurement assistant · hosted in France
I reread your nine years of procurement files over the weekend — 1,040 procurements, 4,900 tenders, 34 standard documents — and I draw from them a gap that shows in your public spending, not only in your deadlines.
The gap measured, and it decides everything else: 9 of the 118 procurements of last year received a single tender. On those 9, the price accepted came out on average 12 % above your estimate; on procurements that received four tenders or more, it came out 4 % below. Sixteen points of difference, on works and routine services: that is public money competition was giving back to you, and that a narrow procurement does not.
The second gap, on your documents: 11 of your 34 standard documents refer to earlier wording or to a document title that no longer exists. A standard document is a template of the service — specification, consultation rules, technical bid framework — reused from one procurement to the next. These are not drafting mistakes: they are templates nobody has had time to reopen in nine years.
What that costs today, on the three items you can check in your own records:
· The completeness check on a procurement: 60 % of the opening work — 6 hours out of the 10 a procurement demands, from receipt of tenders to the report.
· Preparing the tender comparison: 30 % of the analysis time — 4 h 30 out of 15 hours.
· Handling candidate questions: 8 % of an officer's time during the consultation week — 3 h 12 out of 40 hours.
What I propose, and it is not an intention — it is already written: I rewrote the 11 standard documents overnight, each aligned on the wording in force today, with the replaced line and the line replacing it side by side. What that would have changed over the past year: the 74 adapted procedures would have gone out with up-to-date consultation rules, and the 31 open procedures with a document title candidates can find on your buyer profile.
What I do with the three items once the documents are settled: completeness drops from 60 % to 16 % — from 6 hours to 1 h 36; the comparison from 30 % to 10 % — from 4 h 30 to 1 h 30; questions from 8 % to 6 % — from 3 h 12 to 2 h 24. Your four officers keep what they alone can do: settle the need, judge technical merit, give reasons for the award.
The next step takes twenty minutes: tomorrow morning you read the three standard documents that weigh the most — the consultation rules, the particular administrative conditions and the technical bid framework. The head of procurement settles the version, and it serves from the next procurement on — it is that validation which makes the document the authority's own, and it is the only gesture I leave to you. contract-memory_9-years-of-procurements-indexed.pdf1,040 procurements indexed, 9 single-tender procurements, 16 points of difference
⛓ Sourced · 9 years of procurement files, 1,040 procedures, 4,900 tenders, 34 standard documents, published award notices
The gap measured, and it decides everything else: 9 of the 118 procurements of last year received a single tender. On those 9, the price accepted came out on average 12 % above your estimate; on procurements that received four tenders or more, it came out 4 % below. Sixteen points of difference, on works and routine services: that is public money competition was giving back to you, and that a narrow procurement does not.
The second gap, on your documents: 11 of your 34 standard documents refer to earlier wording or to a document title that no longer exists. A standard document is a template of the service — specification, consultation rules, technical bid framework — reused from one procurement to the next. These are not drafting mistakes: they are templates nobody has had time to reopen in nine years.
What that costs today, on the three items you can check in your own records:
· The completeness check on a procurement: 60 % of the opening work — 6 hours out of the 10 a procurement demands, from receipt of tenders to the report.
· Preparing the tender comparison: 30 % of the analysis time — 4 h 30 out of 15 hours.
· Handling candidate questions: 8 % of an officer's time during the consultation week — 3 h 12 out of 40 hours.
What I propose, and it is not an intention — it is already written: I rewrote the 11 standard documents overnight, each aligned on the wording in force today, with the replaced line and the line replacing it side by side. What that would have changed over the past year: the 74 adapted procedures would have gone out with up-to-date consultation rules, and the 31 open procedures with a document title candidates can find on your buyer profile.
What I do with the three items once the documents are settled: completeness drops from 60 % to 16 % — from 6 hours to 1 h 36; the comparison from 30 % to 10 % — from 4 h 30 to 1 h 30; questions from 8 % to 6 % — from 3 h 12 to 2 h 24. Your four officers keep what they alone can do: settle the need, judge technical merit, give reasons for the award.
The next step takes twenty minutes: tomorrow morning you read the three standard documents that weigh the most — the consultation rules, the particular administrative conditions and the technical bid framework. The head of procurement settles the version, and it serves from the next procurement on — it is that validation which makes the document the authority's own, and it is the only gesture I leave to you. contract-memory_9-years-of-procurements-indexed.pdf1,040 procurements indexed, 9 single-tender procurements, 16 points of difference
⛓ Sourced · 9 years of procurement files, 1,040 procedures, 4,900 tenders, 34 standard documents, published award notices
Nine procurements with a single tender — I sensed it without being able to put a figure on it. We do not know the companies beyond those that already bid.
I know them, and I have identified them for you: 34 companies that submitted a tender on a comparable procurement in neighbouring authorities, and that have never seen one of yours.
Where I get them, and it is verifiable line by line: published award notices — public, dated data, and every line carries the procurement, the authority, the subject and the year. I guess no company: I read them where public procurement publishes them.
What the reconciliation gives, purchasing family by purchasing family:
· Road works: 12 companies, 7 of them within 40 kilometres of Roche-Vallon.
· Green space maintenance: 9 companies, in a family where your last three procurements received one tender each.
· Intellectual services: 8 companies.
· Workshop supplies: 5 companies.
What I propose, and the letters are written: a complementary notice sent to the 34, in addition to the regulatory publication on your buyer profile — the buyer profile is the platform on which the authority publishes its procurements and receives tenders. The same notice, on the same day, to all of them: that is what makes the approach regular, and equal treatment is what protects it.
What it is worth, measured on your own history: your procurements with four tenders or more come out 4 % below the estimate, against 12 % above when the tender is unique. On the 9 single-tender procurements of last year, that is €214,000 of public spending competition could have given back to the authority.
The next step I propose: that the file of 34 companies becomes the service's supplier pool, kept up to date with every published award notice. It grows on its own, it costs your officers not one hour, and at each procurement I hand you the list of those that bid on that subject — say yes and the first procurement benefits from Thursday. supplier-pool_34-companies-sourced.pdf34 companies, 4 purchasing families, €214,000 at stake
⛓ Sourced · award notices published over 3 years, 9 single-tender procurements of the authority, estimates and prices accepted
Where I get them, and it is verifiable line by line: published award notices — public, dated data, and every line carries the procurement, the authority, the subject and the year. I guess no company: I read them where public procurement publishes them.
What the reconciliation gives, purchasing family by purchasing family:
· Road works: 12 companies, 7 of them within 40 kilometres of Roche-Vallon.
· Green space maintenance: 9 companies, in a family where your last three procurements received one tender each.
· Intellectual services: 8 companies.
· Workshop supplies: 5 companies.
What I propose, and the letters are written: a complementary notice sent to the 34, in addition to the regulatory publication on your buyer profile — the buyer profile is the platform on which the authority publishes its procurements and receives tenders. The same notice, on the same day, to all of them: that is what makes the approach regular, and equal treatment is what protects it.
What it is worth, measured on your own history: your procurements with four tenders or more come out 4 % below the estimate, against 12 % above when the tender is unique. On the 9 single-tender procurements of last year, that is €214,000 of public spending competition could have given back to the authority.
The next step I propose: that the file of 34 companies becomes the service's supplier pool, kept up to date with every published award notice. It grows on its own, it costs your officers not one hour, and at each procurement I hand you the list of those that bid on that subject — say yes and the first procurement benefits from Thursday. supplier-pool_34-companies-sourced.pdf34 companies, 4 purchasing families, €214,000 at stake
⛓ Sourced · award notices published over 3 years, 9 single-tender procurements of the authority, estimates and prices accepted
Tenders contain the companies' prices and technical bids. They are the most sensitive material we hold. Where do they go when you read them?
They go nowhere. I run in local inference on a machine of the authority, and nothing I read leaves your network.
Local inference means the model computes on your machine: the price of a tender or a candidate's technical bid crosses no external network to be processed. If the authority prefers not to host a machine, the other route is an isolated resource hosted in France, dedicated to Roche-Vallon — no pooling with another authority, which is the condition of the continuity of your service.
What that changes, point by point:
· Candidates' tenders train no model, neither ours nor a third party's.
· I work read-only on your buyer profile and your document management, and the technical account through which I read has no right to write — that is stronger than a promise, because it is checked with one command.
· Tenders stay sealed until the opening hour: I open nothing before the date and time set in the consultation rules, and the log establishes it with timestamps. It is the first thing an unsuccessful candidate will check, and you can show it to them.
· Encryption in transit and at rest, role-based access — rights follow the function: a buying department sees its own procurements, tenders open only for the four officers of the procurement service. 29 roles for the 26 departments and the 3 functions of the service, and the log shows 0 access outside role since go-live.
· Hosting in France, under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The gesture that stays with a person, and it is what gives the contract its value: the award belongs to the tender board and to the buyer, and the signature to the president or the holder of the delegation. Everything leading to it, I have already done: the documents assembled, completeness checked, tenders scored on the published grid, and the analysis report drafted with the calculation set out in plain sight under every score. It is that human, reasoned decision which makes the award defensible before an unsuccessful candidate and before the public accountant.
The figure that sums all this up: 0 tender out of the authority's network, processing in the EU targeted, over 512 tenders and 1,340 traced outputs since go-live.
What I propose: that I keep up to date the register sheet your data protection officer will ask for — hosting, data processed, retention periods, who accesses what. It is asked for once a year and takes two days to rebuild; the first version is already written and you have it as an exhibit. technical-framework_where-the-tenders-live.pdfLocal inference, read-only, sealed tenders, processing in the EU targeted
✎ Framework · deployment architecture, rights of the technical account, tender opening log, first version of the register sheet
Local inference means the model computes on your machine: the price of a tender or a candidate's technical bid crosses no external network to be processed. If the authority prefers not to host a machine, the other route is an isolated resource hosted in France, dedicated to Roche-Vallon — no pooling with another authority, which is the condition of the continuity of your service.
What that changes, point by point:
· Candidates' tenders train no model, neither ours nor a third party's.
· I work read-only on your buyer profile and your document management, and the technical account through which I read has no right to write — that is stronger than a promise, because it is checked with one command.
· Tenders stay sealed until the opening hour: I open nothing before the date and time set in the consultation rules, and the log establishes it with timestamps. It is the first thing an unsuccessful candidate will check, and you can show it to them.
· Encryption in transit and at rest, role-based access — rights follow the function: a buying department sees its own procurements, tenders open only for the four officers of the procurement service. 29 roles for the 26 departments and the 3 functions of the service, and the log shows 0 access outside role since go-live.
· Hosting in France, under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The gesture that stays with a person, and it is what gives the contract its value: the award belongs to the tender board and to the buyer, and the signature to the president or the holder of the delegation. Everything leading to it, I have already done: the documents assembled, completeness checked, tenders scored on the published grid, and the analysis report drafted with the calculation set out in plain sight under every score. It is that human, reasoned decision which makes the award defensible before an unsuccessful candidate and before the public accountant.
The figure that sums all this up: 0 tender out of the authority's network, processing in the EU targeted, over 512 tenders and 1,340 traced outputs since go-live.
What I propose: that I keep up to date the register sheet your data protection officer will ask for — hosting, data processed, retention periods, who accesses what. It is asked for once a year and takes two days to rebuild; the first version is already written and you have it as an exhibit. technical-framework_where-the-tenders-live.pdfLocal inference, read-only, sealed tenders, processing in the EU targeted
✎ Framework · deployment architecture, rights of the technical account, tender opening log, first version of the register sheet
We are launching « Road works 2026 ». The technical department sent me its need in two pages. The tender file has to be assembled.
The file is assembled, on your updated standard documents, and for every line I tell you where it comes from.
The tender file — the set of documents given to candidates — says what the authority is buying, on what terms, and how tenders will be judged.
What the file contains, and the source of each document:
· The consultation rules — the document that sets the rules of the game: deadline for submission, documents to provide, award criteria and their weighting. Built on your rewritten standard document, with the tender reception period entered in the calendar in the version in force at the date of publication — I hand it over with the text beside it, never a number of days from memory.
· The particular administrative conditions — the document governing relations between the authority and the contractor: deadlines, penalties, payment terms, termination. Taken from your 2024 road works contract, with the 4 clauses your last two performances brought into discussion, rewritten.
· The particular technical conditions — the document describing the technical need, structure by structure. Built from the technical department's two pages and your services' survey, and I flag the 6 items it described without a quantity — the 6 requests for clarification are written, ready to go out today.
· The technical bid framework, which imposes the same plan on all candidates — it is what makes tenders genuinely comparable, and your last three procurements had none.
· The price schedule and breakdown, built on the 214 items of your previous road works contract.
The time that moves: preparing a tender file goes from two days to three hours of review, and across 118 procurements a year it is the item where your service has the most to gain.
The next step I propose: that we settle the lotting before writing another line — I have costed it on your nine years, and the result is worth ten minutes of your time. tender-documents_road-works-2026.pdfConsultation rules, administrative and technical conditions, bid framework, 6 clarifications
⛓ Sourced · rewritten standard documents, 2024 road works contract, technical department survey, performance of the last two contracts
The tender file — the set of documents given to candidates — says what the authority is buying, on what terms, and how tenders will be judged.
What the file contains, and the source of each document:
· The consultation rules — the document that sets the rules of the game: deadline for submission, documents to provide, award criteria and their weighting. Built on your rewritten standard document, with the tender reception period entered in the calendar in the version in force at the date of publication — I hand it over with the text beside it, never a number of days from memory.
· The particular administrative conditions — the document governing relations between the authority and the contractor: deadlines, penalties, payment terms, termination. Taken from your 2024 road works contract, with the 4 clauses your last two performances brought into discussion, rewritten.
· The particular technical conditions — the document describing the technical need, structure by structure. Built from the technical department's two pages and your services' survey, and I flag the 6 items it described without a quantity — the 6 requests for clarification are written, ready to go out today.
· The technical bid framework, which imposes the same plan on all candidates — it is what makes tenders genuinely comparable, and your last three procurements had none.
· The price schedule and breakdown, built on the 214 items of your previous road works contract.
The time that moves: preparing a tender file goes from two days to three hours of review, and across 118 procurements a year it is the item where your service has the most to gain.
The next step I propose: that we settle the lotting before writing another line — I have costed it on your nine years, and the result is worth ten minutes of your time. tender-documents_road-works-2026.pdfConsultation rules, administrative and technical conditions, bid framework, 6 clarifications
⛓ Sourced · rewritten standard documents, 2024 road works contract, technical department survey, performance of the last two contracts
Lotting, precisely. The technical department would like a single lot; they say it is simpler to follow.
Both options are costed on your nine years, and the difference is clear: in three lots, your road works procurements received 4.2 tenders on average; in a single lot, 1.8.
Lotting is the division of a contract into several lots that companies can win separately: it is the lever that opens the procurement to companies that cannot do everything, and it is the principle public procurement adopts by default.
What your own history gives, over 27 road works procurements in nine years:
· Single lot — 11 procurements: 1.8 tenders on average, price accepted 9 % above the estimate.
· Three lots — 16 procurements: 4.2 tenders on average, price accepted 5 % below.
· The 14-point gap holds procurement by procurement, and it does not depend on the year: the two configurations alternate across the whole period.
What I propose, and the division is written: three lots — carriageways, signage, surface drainage structures —, each with its technical conditions, its price schedule and its estimate. Of the 34 companies in the pool, 12 can bid on carriageways, 9 on signage and 7 on structures: all three lots find a bidder.
And the technical department's concern, costed as well: following three lots takes 3 hours more a year than following a single lot — I counted them across your 16 three-lot procurements. That is what the automatic tracker gives back in the first year, site meetings and deadlines included: the technical department gains time, it does not lose any.
The decision belongs to the authority — and I hand it to you in ten minutes, both options costed side by side, the division ready to publish either way.
The next step I propose: that we now settle the award criteria — they must be written and published before the tenders are opened, and I have already tested them on your past procurements. lotting_1-lot-versus-3-lots-over-9-years.pdf1.8 tenders against 4.2, 14 points of difference on price
⛓ Sourced · 27 road works procurements over 9 years, tenders received, prices accepted and estimates, tracked follow-up time
Lotting is the division of a contract into several lots that companies can win separately: it is the lever that opens the procurement to companies that cannot do everything, and it is the principle public procurement adopts by default.
What your own history gives, over 27 road works procurements in nine years:
· Single lot — 11 procurements: 1.8 tenders on average, price accepted 9 % above the estimate.
· Three lots — 16 procurements: 4.2 tenders on average, price accepted 5 % below.
· The 14-point gap holds procurement by procurement, and it does not depend on the year: the two configurations alternate across the whole period.
What I propose, and the division is written: three lots — carriageways, signage, surface drainage structures —, each with its technical conditions, its price schedule and its estimate. Of the 34 companies in the pool, 12 can bid on carriageways, 9 on signage and 7 on structures: all three lots find a bidder.
And the technical department's concern, costed as well: following three lots takes 3 hours more a year than following a single lot — I counted them across your 16 three-lot procurements. That is what the automatic tracker gives back in the first year, site meetings and deadlines included: the technical department gains time, it does not lose any.
The decision belongs to the authority — and I hand it to you in ten minutes, both options costed side by side, the division ready to publish either way.
The next step I propose: that we now settle the award criteria — they must be written and published before the tenders are opened, and I have already tested them on your past procurements. lotting_1-lot-versus-3-lots-over-9-years.pdf1.8 tenders against 4.2, 14 points of difference on price
⛓ Sourced · 27 road works procurements over 9 years, tenders received, prices accepted and estimates, tracked follow-up time
For the criteria we always use the same: price 60, technical merit 40. Nobody has ever looked at whether that is the right setting.
I have looked at it — across your 22 most recent procurements — and I propose three sets of criteria, each replayed on your real tenders, with the ranking it would have produced.
Award criteria are the criteria announced in the consultation rules on which each tender will be scored; their weighting sets the weight of each.
The three sets, replayed on the 22 procurements:
· A — price 60 / technical merit 40, your current setting: across the 22, it designates the cheapest tender 22 times. In other words, technical merit has never changed a ranking at your authority — and that is measurable, not arguable.
· B — price 40 / technical merit 60: it changes the ranking on 5 of the 22. On those 5, the contractors actually selected accounted for 11 of the 19 late-delivery penalties of the period: the technical criterion would have kept out those who cost you in performance.
· C — price 40 / technical merit 40 / lead time 20: it changes the ranking on 7 of the 22, and it performs best on works contracts, where your average schedule overrun is 23 days.
What I do beyond that, and what nobody has the time to do: I propose two new technical sub-criteria, written in the form of your own documents and already costed — the qualifications of the staff assigned to the site and the arrangements for keeping traffic flowing for residents. For each I give you the sentence in plain language, the score candidates in your 22 procurements would have obtained, and the tenders the sub-criterion would have moved up or down. You choose on figures, not on intuition.
The rule I hold most firmly, and it is the one that protects the authority: a criterion, a weighting and a scoring rule are settled before publication and do not move once the tenders are opened — and that is precisely why I put them on the table costed now, fifteen days before publication, rather than after. Validation stays with the service: criteria come into force only once settled by the head of procurement, and that is what makes them enforceable against a candidate who would challenge them.
The next step I propose: that the chosen set be replayed each year on the procurements of the financial year, and that in January I hand you the setting your own results designate. The work is done, it fits on one page, and it costs the service one reading a year. award-criteria_3-sets-replayed-on-22-procurements.pdf22 times the cheapest, 5 and 7 rankings changed, 2 sub-criteria written
⛓ Sourced · 22 recent procurements, real tenders and scores, penalty register, performance schedule overruns
Award criteria are the criteria announced in the consultation rules on which each tender will be scored; their weighting sets the weight of each.
The three sets, replayed on the 22 procurements:
· A — price 60 / technical merit 40, your current setting: across the 22, it designates the cheapest tender 22 times. In other words, technical merit has never changed a ranking at your authority — and that is measurable, not arguable.
· B — price 40 / technical merit 60: it changes the ranking on 5 of the 22. On those 5, the contractors actually selected accounted for 11 of the 19 late-delivery penalties of the period: the technical criterion would have kept out those who cost you in performance.
· C — price 40 / technical merit 40 / lead time 20: it changes the ranking on 7 of the 22, and it performs best on works contracts, where your average schedule overrun is 23 days.
What I do beyond that, and what nobody has the time to do: I propose two new technical sub-criteria, written in the form of your own documents and already costed — the qualifications of the staff assigned to the site and the arrangements for keeping traffic flowing for residents. For each I give you the sentence in plain language, the score candidates in your 22 procurements would have obtained, and the tenders the sub-criterion would have moved up or down. You choose on figures, not on intuition.
The rule I hold most firmly, and it is the one that protects the authority: a criterion, a weighting and a scoring rule are settled before publication and do not move once the tenders are opened — and that is precisely why I put them on the table costed now, fifteen days before publication, rather than after. Validation stays with the service: criteria come into force only once settled by the head of procurement, and that is what makes them enforceable against a candidate who would challenge them.
The next step I propose: that the chosen set be replayed each year on the procurements of the financial year, and that in January I hand you the setting your own results designate. The work is done, it fits on one page, and it costs the service one reading a year. award-criteria_3-sets-replayed-on-22-procurements.pdf22 times the cheapest, 5 and 7 rankings changed, 2 sub-criteria written
⛓ Sourced · 22 recent procurements, real tenders and scores, penalty register, performance schedule overruns
The tenders for « Road works 2026 » have arrived. Six candidates. The board meets on Thursday.
The 6 tenders are checked: 4 complete, 2 to regularise, and the 2 requests for regularisation are written, identical in form, ready to go out today.
What the check looked at, document by document, against the list in the consultation rules:
· The candidature documents — identity of the company, authority of the signatory, sworn declaration.
· The supporting certificates — the certificates a candidate must produce: tax and social security compliance, liability insurance, and where required professional qualifications. I check that they are present, still valid, and issued in the name of the company — this is where an expired certificate most easily goes unnoticed.
· The tender documents — form of undertaking, priced schedule, technical bid following the imposed framework.
What that gives:
· 4 complete tenders.
· 2 to regularise: one candidate without a tax and social security compliance certificate, another whose technical bid does not cover lot 2. No missing document is fatal at the candidature stage: both companies can complete, and both letters go out the same day, with the same deadline, the same list of documents and the same closing time. The same treatment for both is equal treatment, and it is proved by the log.
The time that moves: the completeness check goes from 60 % to 16 % of the opening work — from 6 hours to 1 h 36. Across 118 procurements a year, that is 519 hours returned to the service and to your four officers, at unchanged headcount.
The next step I propose: that I score the 4 complete tenders today on the published grid, and that the 2 regularised ones join the ranking as soon as they come in. The completeness table is waiting for you as an exhibit: you validate it, the letters go out, and Thursday's board opens on a clean file. completeness-check_6-tenders-2-to-regularise.pdf4 complete, 2 regularisations, the same deadline for both
⛓ Sourced · published consultation rules, 6 tenders submitted on the buyer profile, certificates and validity dates
What the check looked at, document by document, against the list in the consultation rules:
· The candidature documents — identity of the company, authority of the signatory, sworn declaration.
· The supporting certificates — the certificates a candidate must produce: tax and social security compliance, liability insurance, and where required professional qualifications. I check that they are present, still valid, and issued in the name of the company — this is where an expired certificate most easily goes unnoticed.
· The tender documents — form of undertaking, priced schedule, technical bid following the imposed framework.
What that gives:
· 4 complete tenders.
· 2 to regularise: one candidate without a tax and social security compliance certificate, another whose technical bid does not cover lot 2. No missing document is fatal at the candidature stage: both companies can complete, and both letters go out the same day, with the same deadline, the same list of documents and the same closing time. The same treatment for both is equal treatment, and it is proved by the log.
The time that moves: the completeness check goes from 60 % to 16 % of the opening work — from 6 hours to 1 h 36. Across 118 procurements a year, that is 519 hours returned to the service and to your four officers, at unchanged headcount.
The next step I propose: that I score the 4 complete tenders today on the published grid, and that the 2 regularised ones join the ranking as soon as they come in. The completeness table is waiting for you as an exhibit: you validate it, the letters go out, and Thursday's board opens on a clean file. completeness-check_6-tenders-2-to-regularise.pdf4 complete, 2 regularisations, the same deadline for both
⛓ Sourced · published consultation rules, 6 tenders submitted on the buyer profile, certificates and validity dates
Go ahead with the scoring. And I want to understand every score: that is what the board will ask me.
The 4 complete tenders are scored on the published grid — price 40 %, technical merit 60 % — and the calculation is set out in plain sight under every score.
How each score is built, and it can be reproduced by anyone:
· Price: score referred to the lowest tender, with the formula written out in full in the published consultation rules, and the calculation reproduced under the score.
· Technical merit: four sub-criteria, each tied to a precise passage of the candidate's bid, with the quotation from the bid beside the score. A score with no quoted passage does not exist in my table.
The ranking that comes out: the best tender is not the cheapest. The cheapest tender comes third on technical merit — traffic continuity covered in two lines, no staff named — and its price advantage does not make up the 18 points it loses on the technical side.
What I hand over with the ranking: the tender analysis report drafted — the tender analysis report sets out, tender by tender, how each was scored and why the ranking is what it is; it is what gives reasons for the award before the board, before unsuccessful candidates and before the public accountant.
And the table is alive: change a weighting in front of the board and the ranking is recomputed in ten seconds, with the difference it makes to every tender. This is not a way of changing the rules — the published criteria remain the ones that apply — it is a way of showing the board what the chosen setting produces, with figures, before it votes.
The time that moves: preparing the comparison goes from 30 % to 10 % of the analysis time — from 4 h 30 to 1 h 30, and the analysis report is available the day after the tenders are opened, against nine days. Across 118 procurements, that is 354 hours.
The next step I propose: that I also draft the 4 reasoned rejection letters, one per unsuccessful candidate, recalling their score and the successful tenderer's scores. They are written: a candidate who understands why they lost comes back to the next procurement, and that is your competition next year. tender-analysis-report_4-tenders-scored.pdfPrice 40 / technical merit 60, calculation in plain sight under every score
⛓ Sourced · grid of the published rules, 4 admissible tenders, candidates' technical bids, price schedules
How each score is built, and it can be reproduced by anyone:
· Price: score referred to the lowest tender, with the formula written out in full in the published consultation rules, and the calculation reproduced under the score.
· Technical merit: four sub-criteria, each tied to a precise passage of the candidate's bid, with the quotation from the bid beside the score. A score with no quoted passage does not exist in my table.
The ranking that comes out: the best tender is not the cheapest. The cheapest tender comes third on technical merit — traffic continuity covered in two lines, no staff named — and its price advantage does not make up the 18 points it loses on the technical side.
What I hand over with the ranking: the tender analysis report drafted — the tender analysis report sets out, tender by tender, how each was scored and why the ranking is what it is; it is what gives reasons for the award before the board, before unsuccessful candidates and before the public accountant.
And the table is alive: change a weighting in front of the board and the ranking is recomputed in ten seconds, with the difference it makes to every tender. This is not a way of changing the rules — the published criteria remain the ones that apply — it is a way of showing the board what the chosen setting produces, with figures, before it votes.
The time that moves: preparing the comparison goes from 30 % to 10 % of the analysis time — from 4 h 30 to 1 h 30, and the analysis report is available the day after the tenders are opened, against nine days. Across 118 procurements, that is 354 hours.
The next step I propose: that I also draft the 4 reasoned rejection letters, one per unsuccessful candidate, recalling their score and the successful tenderer's scores. They are written: a candidate who understands why they lost comes back to the next procurement, and that is your competition next year. tender-analysis-report_4-tenders-scored.pdfPrice 40 / technical merit 60, calculation in plain sight under every score
⛓ Sourced · grid of the published rules, 4 admissible tenders, candidates' technical bids, price schedules
There is a very low tender in the lot. Thirty per cent below our estimate. We never know what to do with those.
It is at −32 % of your estimate, and the request for justification is written, ready to go out today: it is the route that allows this tender to be selected if it is serious, and it is the one I propose.
An abnormally low tender is a tender whose price is so far from the market that it raises doubt about the candidate's ability to perform; public procurement then requires that the candidate be questioned before any decision, and that what follows be reasoned.
What I checked before telling you, because a gap is not a proof:
· Your estimate itself: taken from the 2024 contract and updated on the public works index — it holds.
· The other three tenders: they sit within a 7 % range, the low tender is clearly isolated.
· The price breakdown, item by item: the gap is concentrated on 3 items out of 214 — temporary signage, spoil removal and site supervision — the other items are within the average. That is what makes the request for justification precise and useful.
· The company's history: it has performed two contracts for you in nine years, with no penalty. The fact is on file, it counts, and I give it to you with the rest.
The request for justification I hand you carries the three named items, the question asked for each, and the reply deadline. It is written in the form you use, and it goes out under your signature.
What the reply will allow: if the candidate justifies the three items — a spoil removal process of its own, supervision shared with a neighbouring site — the tender is selected and the authority buys 32 % cheaper. If the justification does not come, the board's decision is reasoned, written and defensible. Either way you decide on a written answer, not on an impression.
What I run across all your procurements, and it costs nothing: the same gap check on the 512 tenders of the year. It raised 7 tenders more than 25 % out in twelve months: 4 were justified and selected — €186,000 of public spending saved — and 3 were not.
The next step I propose: that the gap check fires at the opening of every procurement, without your having to ask, and that the request for justification reaches you written the same day, with the items named. abnormally-low-tender_request-for-justification.pdf−32 %, 3 items named out of 214, 7 tenders raised in 12 months
⛓ Sourced · updated estimate, price breakdown of the 4 tenders, 214 schedule items, candidate's performance history
An abnormally low tender is a tender whose price is so far from the market that it raises doubt about the candidate's ability to perform; public procurement then requires that the candidate be questioned before any decision, and that what follows be reasoned.
What I checked before telling you, because a gap is not a proof:
· Your estimate itself: taken from the 2024 contract and updated on the public works index — it holds.
· The other three tenders: they sit within a 7 % range, the low tender is clearly isolated.
· The price breakdown, item by item: the gap is concentrated on 3 items out of 214 — temporary signage, spoil removal and site supervision — the other items are within the average. That is what makes the request for justification precise and useful.
· The company's history: it has performed two contracts for you in nine years, with no penalty. The fact is on file, it counts, and I give it to you with the rest.
The request for justification I hand you carries the three named items, the question asked for each, and the reply deadline. It is written in the form you use, and it goes out under your signature.
What the reply will allow: if the candidate justifies the three items — a spoil removal process of its own, supervision shared with a neighbouring site — the tender is selected and the authority buys 32 % cheaper. If the justification does not come, the board's decision is reasoned, written and defensible. Either way you decide on a written answer, not on an impression.
What I run across all your procurements, and it costs nothing: the same gap check on the 512 tenders of the year. It raised 7 tenders more than 25 % out in twelve months: 4 were justified and selected — €186,000 of public spending saved — and 3 were not.
The next step I propose: that the gap check fires at the opening of every procurement, without your having to ask, and that the request for justification reaches you written the same day, with the items named. abnormally-low-tender_request-for-justification.pdf−32 %, 3 items named out of 214, 7 tenders raised in 12 months
⛓ Sourced · updated estimate, price breakdown of the 4 tenders, 214 schedule items, candidate's performance history
I went back over the 214 questions candidates asked on your procurements this year: 12 forms of question make up 163 of the 214, and the 12 answers are written, sourced on your own documents.
What the sorting gave:
· 163 questions of form — documents to provide, format and closing time for submission on the buyer profile, site visit, signature arrangements, division into lots, date of the board.
· 39 technical questions, which call for the technical department — and they reach it with the relevant passage of the technical conditions already located.
· 12 questions touching price or the assessment of a tender — those go through the buyer, and the draft answer is handed to them written.
What the real cost was, and it does not show in hours: 41 questions last year went unanswered within the deadline set by your consultation rules. An unanswered question means candidates who give up: of the 9 single-tender procurements, 7 had a question left unanswered. The link fits on one line, and it explains a good part of your €214,000 gap.
What I propose, and it is already drafted: a written mandate, capped, dated and withdrawable with a word.
· Scope: the 12 forms of question, listed by name, and nothing else.
· Distribution: every answer goes to ALL the candidates of the procurement at the same time, published on the buyer profile — that is what keeps a fast answer a regular one, and equal treatment comes out reinforced, not weakened.
· Cap: no question touching price, the assessment of a tender or technical content enters the mandate — those reach you drafted, within four hours, and it is you who sends them.
· A one-page daily record, review at three months, and without an explicit decision at the review the mandate stops — renewal is what requires a signature, not stopping. Withdrawal: one word, and direct sending ceases within the minute.
What that gives, in figures: handling questions goes from 8 % to 6 % of an officer's time during the consultation week — from 3 h 12 to 2 h 24, that is 94 hours over the year — and above all 41 questions out of time brought down to 0.
The next step I propose: that the 12 answers be read tomorrow, one by one — twenty minutes. As soon as they are validated, the candidates of the current procurement get their answer the same day, and the first publication goes out on Thursday. candidate-reply-mandate_12-question-types-capped.pdf163 questions covered, distribution to all, review at 3 months
✎ Framework · 214 candidate questions over 12 months, consultation rules, buyer profile publication log, drafted mandate
What the sorting gave:
· 163 questions of form — documents to provide, format and closing time for submission on the buyer profile, site visit, signature arrangements, division into lots, date of the board.
· 39 technical questions, which call for the technical department — and they reach it with the relevant passage of the technical conditions already located.
· 12 questions touching price or the assessment of a tender — those go through the buyer, and the draft answer is handed to them written.
What the real cost was, and it does not show in hours: 41 questions last year went unanswered within the deadline set by your consultation rules. An unanswered question means candidates who give up: of the 9 single-tender procurements, 7 had a question left unanswered. The link fits on one line, and it explains a good part of your €214,000 gap.
What I propose, and it is already drafted: a written mandate, capped, dated and withdrawable with a word.
· Scope: the 12 forms of question, listed by name, and nothing else.
· Distribution: every answer goes to ALL the candidates of the procurement at the same time, published on the buyer profile — that is what keeps a fast answer a regular one, and equal treatment comes out reinforced, not weakened.
· Cap: no question touching price, the assessment of a tender or technical content enters the mandate — those reach you drafted, within four hours, and it is you who sends them.
· A one-page daily record, review at three months, and without an explicit decision at the review the mandate stops — renewal is what requires a signature, not stopping. Withdrawal: one word, and direct sending ceases within the minute.
What that gives, in figures: handling questions goes from 8 % to 6 % of an officer's time during the consultation week — from 3 h 12 to 2 h 24, that is 94 hours over the year — and above all 41 questions out of time brought down to 0.
The next step I propose: that the 12 answers be read tomorrow, one by one — twenty minutes. As soon as they are validated, the candidates of the current procurement get their answer the same day, and the first publication goes out on Thursday. candidate-reply-mandate_12-question-types-capped.pdf163 questions covered, distribution to all, review at 3 months
✎ Framework · 214 candidate questions over 12 months, consultation rules, buyer profile publication log, drafted mandate
That leaves performance. Two hundred and fourteen contracts running, and that is where things slip: tacit renewals we discover afterwards, contracts that expire without anything being relaunched.
The tracker for the 214 contracts is built, and it has already surfaced 31 renewals and 4 expiry dates nobody had time to reconcile.
What the tracker carries, contract by contract:
· Subject, contractor, buying department and amount committed.
· The expiry date and the renewal mechanism — renewal is the extension of the contract for a further period; when tacit, it takes effect without anyone having to decide, which means it has to be seen coming.
· The amount consumed against the maximum, for framework agreements with purchase orders — a framework agreement with purchase orders sets prices and terms, and the authority orders as its needs arise within a ceiling. 3 of your 13 framework agreements have consumed more than 80 % of their ceiling with seven months to run: the 3 replacement procurements are scheduled.
· Penalties applied and performance incidents, taken from your site reports.
What the review gave:
· 31 contracts tacitly renewed last year with no review at all. For all 31 I hand you the review sheet: amount consumed, incidents, contract price compared with comparable procurements in neighbouring authorities. Of the 31, 6 would gain from being put back out to competition and 25 from being renewed — and I tell you so with the figure under every line.
· 4 contracts expired with no procurement launched. The 4 tender files are assembled, and the backward schedule for each is set.
And the point that protects public spending right to the end: before every payment order I check that the certification of service rendered is on file — service rendered is the written attestation that a service has actually been performed: without it, the public accountant does not pay. Over the quarter: 0 payment order returned for a missing document, against 23 last year, and your overall supplier payment time goes from 41 to 24 days. An authority that pays on time keeps its companies, and they are the ones that bid on the next procurement.
The next step I propose: that every expiry be flagged to you six months ahead, with the review sheet already written and the tender file assembled if putting it back out to competition is chosen. Across the 214 contracts, that means zero expiry discovered too late — and a procurement calendar spread over the year instead of three peaks. contract-tracker_214-contracts-deadlines-and-renewals.pdf31 renewals reviewed, 4 expiries caught, 23 payment orders → 0
⛓ Sourced · 214 contracts in performance, expiry dates and renewal clauses, framework agreement consumption, payment order log and payment times
What the tracker carries, contract by contract:
· Subject, contractor, buying department and amount committed.
· The expiry date and the renewal mechanism — renewal is the extension of the contract for a further period; when tacit, it takes effect without anyone having to decide, which means it has to be seen coming.
· The amount consumed against the maximum, for framework agreements with purchase orders — a framework agreement with purchase orders sets prices and terms, and the authority orders as its needs arise within a ceiling. 3 of your 13 framework agreements have consumed more than 80 % of their ceiling with seven months to run: the 3 replacement procurements are scheduled.
· Penalties applied and performance incidents, taken from your site reports.
What the review gave:
· 31 contracts tacitly renewed last year with no review at all. For all 31 I hand you the review sheet: amount consumed, incidents, contract price compared with comparable procurements in neighbouring authorities. Of the 31, 6 would gain from being put back out to competition and 25 from being renewed — and I tell you so with the figure under every line.
· 4 contracts expired with no procurement launched. The 4 tender files are assembled, and the backward schedule for each is set.
And the point that protects public spending right to the end: before every payment order I check that the certification of service rendered is on file — service rendered is the written attestation that a service has actually been performed: without it, the public accountant does not pay. Over the quarter: 0 payment order returned for a missing document, against 23 last year, and your overall supplier payment time goes from 41 to 24 days. An authority that pays on time keeps its companies, and they are the ones that bid on the next procurement.
The next step I propose: that every expiry be flagged to you six months ahead, with the review sheet already written and the tender file assembled if putting it back out to competition is chosen. Across the 214 contracts, that means zero expiry discovered too late — and a procurement calendar spread over the year instead of three peaks. contract-tracker_214-contracts-deadlines-and-renewals.pdf31 renewals reviewed, 4 expiries caught, 23 payment orders → 0
⛓ Sourced · 214 contracts in performance, expiry dates and renewal clauses, framework agreement consumption, payment order log and payment times
And the 26 buying departments? Each does more or less as it pleases, and we are the ones who put it right at the end.
Each now receives what it was missing: an up-to-date sheet for its purchasing family, and a need form that comes back complete — 84 % of need statements arrive complete against 39 % a year ago.
What each buying department receives:
· The sheet for its purchasing family — works, supplies, services, intellectual services — with the procedure applicable to the estimated amount, the documents required and the lead time to allow. The amount below which a contract may be awarded without prior advertising or competition appears in the version in force, with the text beside it and its date of entry into force — never a figure from memory.
· The need statement form, which asks the six questions whose absence sent files back: quantities, term, whether variants are allowed, site constraints, estimated budget, and which department will follow performance.
· A realistic calendar for its procurement, built backwards from the date on which the department needs the service.
What that changed, measured across 118 procurements:
· Need statements complete first time: 39 % → 84 %.
· Exchanges between the buying department and procurement: 3.2 per procurement → 0.7.
· Procurements launched late against the real need: 22 → 4.
And one effect I did not expect, which your logs show: the 26 departments ask 41 % fewer questions of the procurement service — not because they hold back, but because the sheet answers before they call.
The next step I propose: that the purchasing family sheet go to the 26 departments on the first Monday of each month, and that the annual procurement calendar be published internally from January. It is built on the 118 procurements of the year and the 214 expiry dates: departments know when their need must be stated, and your four officers stop discovering an urgent contract on the day it is already late. purchasing-family-sheets_26-buying-departments.pdf39 % → 84 % complete needs, 3.2 → 0.7 exchanges
⛓ Sourced · 118 need statements of the year, log of exchanges, versions in force of procurement texts, expiry dates of the 214 contracts
What each buying department receives:
· The sheet for its purchasing family — works, supplies, services, intellectual services — with the procedure applicable to the estimated amount, the documents required and the lead time to allow. The amount below which a contract may be awarded without prior advertising or competition appears in the version in force, with the text beside it and its date of entry into force — never a figure from memory.
· The need statement form, which asks the six questions whose absence sent files back: quantities, term, whether variants are allowed, site constraints, estimated budget, and which department will follow performance.
· A realistic calendar for its procurement, built backwards from the date on which the department needs the service.
What that changed, measured across 118 procurements:
· Need statements complete first time: 39 % → 84 %.
· Exchanges between the buying department and procurement: 3.2 per procurement → 0.7.
· Procurements launched late against the real need: 22 → 4.
And one effect I did not expect, which your logs show: the 26 departments ask 41 % fewer questions of the procurement service — not because they hold back, but because the sheet answers before they call.
The next step I propose: that the purchasing family sheet go to the 26 departments on the first Monday of each month, and that the annual procurement calendar be published internally from January. It is built on the 118 procurements of the year and the 214 expiry dates: departments know when their need must be stated, and your four officers stop discovering an urgent contract on the day it is already late. purchasing-family-sheets_26-buying-departments.pdf39 % → 84 % complete needs, 3.2 → 0.7 exchanges
⛓ Sourced · 118 need statements of the year, log of exchanges, versions in force of procurement texts, expiry dates of the 214 contracts
The community board will ask me what this has given us. I need a figure I can defend in front of elected members.
967 hours returned to the service over the year — seven months of work of a full-time officer — and the calculation is yours, it fits in three lines.
The calculation, item by item, so you can redo it:
· Completeness check: 118 procurements, 6 hours brought down to 1 h 36 — 60 % → 16 % of the opening work — that is 519 hours.
· Tender comparison: 118 procurements, 4 h 30 brought down to 1 h 30 — 30 % → 10 % — that is 354 hours.
· Candidate questions: 118 consultation weeks, 3 h 12 brought down to 2 h 24 — 8 % → 6 % — that is 94 hours.
What these hours are, and it is what defends best in front of elected members: staff time returned to the service, at unchanged headcount — no post cut, no post created. It is checked against your own procurement logs.
What these hours and the decisions that came with them produced:
· Procurements receiving a single tender: 9 → 2, and the last two are in a family where three companies exist in the whole country.
· Price accepted compared with the estimate, all procurements: 3 % above → 2 % below — on €38m of purchases, that is the measure your elected members will look at first.
· Analysis report available: 9 days after opening → the next day.
· Candidate questions answered within the deadline of the rules: 173 of 214 → 214 of 214.
· Payment orders returned by the public accountant for a missing document: 23 → 0, and overall payment time: 41 days → 24.
· Tacit renewals passed with no review: 31 → 0.
· Abnormally low tenders justified then selected: 4, that is €186,000 of public spending saved.
And the framework measures: 0 award pronounced without the board or the buyer, 0 output sent without a person's validation, 0 tender out of the network, over 1,340 traced outputs.
What I propose for the meeting: the calculation page is written and fits on one side — three lines of calculation, seven measures, three framework measures. Give it to the elected members with the agenda: a figure read the day before is discussed better than a figure discovered in the room. year-review_967-hours-returned-to-the-service.pdf60→16, 30→10, 8→6, and the calculation redoable on one side
⛓ Sourced · procurement logs, estimates and prices accepted, payment order log, register of candidate questions
The calculation, item by item, so you can redo it:
· Completeness check: 118 procurements, 6 hours brought down to 1 h 36 — 60 % → 16 % of the opening work — that is 519 hours.
· Tender comparison: 118 procurements, 4 h 30 brought down to 1 h 30 — 30 % → 10 % — that is 354 hours.
· Candidate questions: 118 consultation weeks, 3 h 12 brought down to 2 h 24 — 8 % → 6 % — that is 94 hours.
What these hours are, and it is what defends best in front of elected members: staff time returned to the service, at unchanged headcount — no post cut, no post created. It is checked against your own procurement logs.
What these hours and the decisions that came with them produced:
· Procurements receiving a single tender: 9 → 2, and the last two are in a family where three companies exist in the whole country.
· Price accepted compared with the estimate, all procurements: 3 % above → 2 % below — on €38m of purchases, that is the measure your elected members will look at first.
· Analysis report available: 9 days after opening → the next day.
· Candidate questions answered within the deadline of the rules: 173 of 214 → 214 of 214.
· Payment orders returned by the public accountant for a missing document: 23 → 0, and overall payment time: 41 days → 24.
· Tacit renewals passed with no review: 31 → 0.
· Abnormally low tenders justified then selected: 4, that is €186,000 of public spending saved.
And the framework measures: 0 award pronounced without the board or the buyer, 0 output sent without a person's validation, 0 tender out of the network, over 1,340 traced outputs.
What I propose for the meeting: the calculation page is written and fits on one side — three lines of calculation, seven measures, three framework measures. Give it to the elected members with the agenda: a figure read the day before is discussed better than a figure discovered in the room. year-review_967-hours-returned-to-the-service.pdf60→16, 30→10, 8→6, and the calculation redoable on one side
⛓ Sourced · procurement logs, estimates and prices accepted, payment order log, register of candidate questions
And when you get it wrong? A candidate wrongly flagged on a document turns into an appeal here.
It happened, I counted it and I publish it: of the 256 missing-document flags of the first quarter, 11 were unfounded — 4.3 %. That is the figure that does not flatter me, and it is better that it comes from me than from a candidate.
The cause, measured and not supposed:
· 7 of the 11 came from a certificate filed under an unusual title — the insurance certificate tucked at the end of the technical bid, tax and social security compliance provided as an annual statement. The document was there; I was looking for it under the name your rules give it.
· 4 came from professional qualifications presented in a form equivalent to the one the standard document named.
What I did with it, and it is measured: I now read every document by its content and not by its title; 9 further forms of certificate were added to the check; and every flag carries the file and the page where I looked, which allows it to be lifted in thirty seconds.
The following quarter: 1 unfounded flag out of 243 — 0.4 %.
What you should know about the 11, because that is the real question: no candidate was excluded. A missing-document flag opens a request for regularisation, never a rejection — the 11 companies were informed the same day, all 11 replied, and all 11 stayed in the procurement. The check cost eleven letters; it cost no tender, and that is exactly why the circuit is built this way.
And the rule that holds the rest: a document I have not found, I ask for — I never declare it absent without saying where I looked. Over the quarter: 243 flags, 243 files and pages cited, 232 regularisations obtained within 48 hours. That is what makes 0.4 % a figure you can rely on.
What I propose now: that the 9 added forms of certificate be written into your standard consultation rules, in full. Candidates will know what is accepted before they file, and the number of regularisations will fall for everyone — the page is written and you have it tonight. unfounded-flags_11-then-1.pdf4.3 % → 0.4 %, measured cause, 0 candidate excluded
⛓ Sourced · completeness flag log over two quarters, documents filed, regularisation letters and replies
The cause, measured and not supposed:
· 7 of the 11 came from a certificate filed under an unusual title — the insurance certificate tucked at the end of the technical bid, tax and social security compliance provided as an annual statement. The document was there; I was looking for it under the name your rules give it.
· 4 came from professional qualifications presented in a form equivalent to the one the standard document named.
What I did with it, and it is measured: I now read every document by its content and not by its title; 9 further forms of certificate were added to the check; and every flag carries the file and the page where I looked, which allows it to be lifted in thirty seconds.
The following quarter: 1 unfounded flag out of 243 — 0.4 %.
What you should know about the 11, because that is the real question: no candidate was excluded. A missing-document flag opens a request for regularisation, never a rejection — the 11 companies were informed the same day, all 11 replied, and all 11 stayed in the procurement. The check cost eleven letters; it cost no tender, and that is exactly why the circuit is built this way.
And the rule that holds the rest: a document I have not found, I ask for — I never declare it absent without saying where I looked. Over the quarter: 243 flags, 243 files and pages cited, 232 regularisations obtained within 48 hours. That is what makes 0.4 % a figure you can rely on.
What I propose now: that the 9 added forms of certificate be written into your standard consultation rules, in full. Candidates will know what is accepted before they file, and the number of regularisations will fall for everyone — the page is written and you have it tonight. unfounded-flags_11-then-1.pdf4.3 % → 0.4 %, measured cause, 0 candidate excluded
⛓ Sourced · completeness flag log over two quarters, documents filed, regularisation letters and replies
Last question, and it is the one that will decide the board: what do you do on your own, and if we stop in a year, what is left?
Three gestures, not one more, each withdrawable with a word — and what stays with the authority is everything this year has produced.
The three gestures I make on my own:
· I check the completeness of every tender as soon as it is filed and hand the table to the buyer. And the reverse is true as well: a tender withdrawn by its candidate before the closing time leaves the table the same minute — I keep no trace of a tender the company has withdrawn as it is entitled to do.
· I flag every contract expiry six months ahead, with the review sheet written. That is the gesture that took renewals passed with no review from 31 to 0.
· I hand the head of procurement the weekly record every Monday: procurements under way, tenders expected, candidate questions, expiry dates, payment orders awaiting certification of service rendered. It is the only thing I send on my own, and it goes to them alone.
What stays with a person, because it is exactly what gives it its value: the award belongs to the tender board and the buyer; the signature of the contract to the president or the holder of the delegation; the assessment of technical merit and the reasons for the ranking to the service. Across 1,340 outputs, those gestures were made 1,340 times by a person.
And the exit, since the board will ask:
· The index of the 1,040 procurements is deleted, and it contained none of your files — only what is needed to find them. Your nine years of archives have not moved by a single byte, and your buyer profile stays yours.
· The 34 updated standard documents, the 12 answers to candidates, the 4 purchasing family sheets, the need statement form, the tracker for the 214 contracts and the analysis grid stay with the authority, in its office format, readable without us. That is the asset this year will have created, and it would be wrong for it to stay with us.
· No migration on the way in, therefore none on the way out: your buyer profile, your financial software and your document management are not replaced.
What I propose so that this does not stay a sentence: a dry-run exit at the end of the first quarter, half a day: we switch off, we check that the service runs a procurement exactly as before, we switch back on. The protocol is written, it fits on one page, and the least costly date is the first Friday of August — no procurement has a deadline that day. The board will know what the promise is worth before a second year is committed. technical-framework_where-the-tenders-live.pdfReversibility: 0 migration in, 0 migration out contract-tracker_214-contracts-deadlines-and-renewals.pdfOwned by the authority, readable without us
✎ Framework · settings of the automatic gestures, sending log, export formats of the standard documents and the tracker, dry-run exit protocol
The three gestures I make on my own:
· I check the completeness of every tender as soon as it is filed and hand the table to the buyer. And the reverse is true as well: a tender withdrawn by its candidate before the closing time leaves the table the same minute — I keep no trace of a tender the company has withdrawn as it is entitled to do.
· I flag every contract expiry six months ahead, with the review sheet written. That is the gesture that took renewals passed with no review from 31 to 0.
· I hand the head of procurement the weekly record every Monday: procurements under way, tenders expected, candidate questions, expiry dates, payment orders awaiting certification of service rendered. It is the only thing I send on my own, and it goes to them alone.
What stays with a person, because it is exactly what gives it its value: the award belongs to the tender board and the buyer; the signature of the contract to the president or the holder of the delegation; the assessment of technical merit and the reasons for the ranking to the service. Across 1,340 outputs, those gestures were made 1,340 times by a person.
And the exit, since the board will ask:
· The index of the 1,040 procurements is deleted, and it contained none of your files — only what is needed to find them. Your nine years of archives have not moved by a single byte, and your buyer profile stays yours.
· The 34 updated standard documents, the 12 answers to candidates, the 4 purchasing family sheets, the need statement form, the tracker for the 214 contracts and the analysis grid stay with the authority, in its office format, readable without us. That is the asset this year will have created, and it would be wrong for it to stay with us.
· No migration on the way in, therefore none on the way out: your buyer profile, your financial software and your document management are not replaced.
What I propose so that this does not stay a sentence: a dry-run exit at the end of the first quarter, half a day: we switch off, we check that the service runs a procurement exactly as before, we switch back on. The protocol is written, it fits on one page, and the least costly date is the first Friday of August — no procurement has a deadline that day. The board will know what the promise is worth before a second year is committed. technical-framework_where-the-tenders-live.pdfReversibility: 0 migration in, 0 migration out contract-tracker_214-contracts-deadlines-and-renewals.pdfOwned by the authority, readable without us
✎ Framework · settings of the automatic gestures, sending log, export formats of the standard documents and the tracker, dry-run exit protocol
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 public purchasing department
Each use corresponds to an agent we deploy. All of them work in support, subject to your approval — the award remaining a human decision.
Included in your agent
The 6 capabilities essential to this promise are included, at no extra cost.
Drafting the contract documents
Help drafting the specifications, technical specifications and consultation rules, from your templates and the nature of the need.
Completeness checks on the bids
Checking applications and tenders: administrative documents, certificates, guarantees — flagged so they can be put right.
Comparing the tenders
Scoring the admissible tenders on the grid in the rules, for human analysis and a draft analysis report.
Checking supporting documents
Extraction and checking of the tax, social security and insurance certificates submitted by bidders.
Anomaly detection & abnormally low tenders
Spotting inconsistencies, duplicates and potentially abnormally low tenders, so justification can be sought.
Answering bidders' questions
Answering companies' recurring questions about a consultation: documents, deadlines, how to submit.
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
Record validations, conflicts of interest, criteria and decisions
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.
Link enquiry, order, goods receipt and invoice without double entry
Other needs our agents cover
Each card says where the matching agent stands: available, on quote, or still being architected.
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.
Memory of the contracts
Instantly find a precedent, a document template or a clause in the contracts and internal procedures.
Automated regulatory and legal watch from 825 € incl. VAT / month Information & monitoring officer →Contract follow-up & steering
Tracking tables for consultations, deadlines, renewals and performance — the fuel for purchasing management.
Management control from 785 € incl. VAT / month Management controller →Purchasing & supply steering agent
ERP, forecasts.
Purchasing & supply steering agent from 474 € incl. VAT / month Discover the agent →
Does your need fall outside this?
Book the free audit →
Build your agent
In 15 minutes we identify the agent that will give your staff the most time back — without oversizing the project.
The gain
How much time can a purchasing department win back?
By automating completeness checks and the preparation of comparisons, a contracts department can aim for a noticeable reduction in production time on every consultation — reinvested in legal soundness and in steering.
Completeness check on a consultation
Preparing a comparison of tenders
Answering a question from a bidder
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
One offer, one agent
A public procurement agent (drafting documents, completeness checks, comparing tenders), installed and operated for you. Prices adapted to the public sector — subscription, the time it takes for the gains to settle in.
This agent is priced with you, not online. We are adjusting its scope at the moment, and online subscription stays closed while we do. Tell us what you need: we will come back to you with a price. Request a quote
Our commitment
Four guarantees that matter to a public purchasing department
Tenders never leave the departmentLocal inference or an isolated resource hosted in France; no bid entrusted to a foreign third party.
Data in France, under French lawFor the tenders: hosting under French law, native minimisation, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The award stays a human decisionThe agent prepares the analysis; ranking and award rest with the committee and the buyer.
Human oversight & traceabilityDrafting the contract documents: monitoring, updates and logging, compliant with the AI Act and with equal treatment between bidders.
Frequently asked questions
Your questions, our answers
Does the agent award the contract?
No: it prepares the analysis — completeness checks, scoring on the grid in the rules, a draft report. Ranking and award rest with the committee and the buyer. The agent assists, the public officer decides.
Does it know the 2026 thresholds?
Yes: exemption thresholds, innovation purchasing and formal procedures are built in and kept up to date. The agent applies the framework in force, under the buyer's control.
Is it compatible with a buyer portal?
Yes: the agent prepares the documents and the analysis, while submission and publication stay on the buyer's own platform. We adapt the integration to your environment.
Is bidders' data protected?
Yes. Tenders, prices and technical submissions are confidential and bear on equal treatment: our agents run locally within the authority or are hosted in France, with the deployment objective of processing and access operated within the European Union and an architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
How does the agent guarantee equal treatment between bidders?
The agent applies the same criteria, taken from the consultation rules, to every bid, and records each operation. It flags anomalies and potentially abnormally low tenders, but neither rejects nor ranks any tender on its own: the decision and its reasoning stay human.
Do we have to change our buyer portal or our tools?
No. The agent connects to your existing tools (buyer portal, document management, document templates) and complements them, without imposing a migration.
How long does it take to deploy an agent?
A few weeks as a rule, after a free audit that identifies the use case with the best return, then a phase of design, integration and testing before going live and training your officers.
Related roles
Other public sector roles
Let's talk
Let us estimate the potential in your purchasing department
15 minutes to identify the use case with the best return — hosted in France, supervised, with no commitment.