● B2B offer — Industry & maintenance
Predictive maintenance: the intervention proposed at the right moment
Stepping in too early costs parts and availability; stepping in too late costs a stoppage. Your agent draws on your sensor data and your intervention history to flag the equipment whose signals are moving, with the measurements and the precedents that justify it. Hosted in France: your industrial data stays with you. The maintenance manager schedules and decides on any intervention.
Hosted in France
Industrial data protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Predictive maintenance · hosted in France
Which equipment needs our attention?
Four items of equipment whose signals are moving away from their usual behaviour.
For each: the successive measurements, the comparison period and the past interventions on similar signals.
The operating ranges you have set are shown.
🔗 Sourced · sensors and intervention history
For each: the successive measurements, the comparison period and the past interventions on similar signals.
The operating ranges you have set are shown.
🔗 Sourced · sensors and intervention history
Should we schedule a shutdown?
I provide the measurements, the precedents and the production windows available.
Scheduling a shutdown commits production and safety: the decision belongs to the maintenance manager, who knows the site's constraints.
✎ Support · material gathered, human scheduling
Scheduling a shutdown commits production and safety: the decision belongs to the maintenance manager, who knows the site's constraints.
✎ Support · material gathered, human scheduling
Local inference · no data outside the EU
Sensor 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 maintenance manager decidesThe agent prepares, never rules
✦ In brief
A Blue Lemon Agent predictive maintenance agent draws on your sensors and your intervention history to flag the equipment whose signals are moving, with the successive measurements, the comparison period and comparable precedents. Scheduling a shutdown remains a maintenance decision. It runs on local inference or is hosted in France: your industrial data stays with you, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
6
maintenance uses ready to deploy
0
decision taken without human approval
These figures describe our offer, not results measured at a client. How large the gain is on your number of items tracked and the density of your sensors is confirmed by a pilot.
The context
What does an AI agent bring to your maintenance?
An intervention placed at the right moment beats one carried out as a matter of routine, or a repair made in a hurry.
! The issue
Placing an intervention at the right moment means cross-checking how the signals are moving, the equipment's history and the production windows available. That cross-check is continuous and covers a lot of equipment. The agent keeps it up and documents every flag with its measurements and its precedents.
✓ Our answer
The maintenance manager has documented flags across the whole fleet, with comparable precedents and the production windows. Scheduling a shutdown commits production and safety: that decision remains theirs. Local inference or an isolated resource hosted in France: your industrial data and your failure history, which describe your production equipment, do not leave the company.
The decisive point
Your industrial data and your failure history: sovereignty & compliance
Your sensor data and your intervention history describe your industrial equipment precisely. Here is how they are protected.
Local inference
The agent can run on a machine belonging to your organisation: no sensor data and no intervention history leaves the network.
Hosting in France
Otherwise, a dedicated and isolated resource hosted in France, under French law — your equipment and your sensor data: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
For your industrial data and your failure history, the architecture aims to reduce exposure to the Cloud Act and FISA 702; being located in France or in the European Union does not, on its own, guarantee immunity.
Isolated resource
No pooling: an environment strictly dedicated to your industrial site and its fleet of equipment.
Measurements and precedents provided
Every flag keeps its measurements, its comparison period and the comparable past interventions; encryption, role-based access and logging.
AI Act: governed deployment
The agent is strictly in support; no shutdown is scheduled and no intervention is triggered automatically; traceability and human oversight from end to end.
What depends on the architecture chosen These points are not general guarantees: they are settled deployment by deployment, in the quotation.
- The applicable location is that of the architecture set out in the quotation and verified before commissioning.
- Local execution is announced only for the configuration explicitly described and accepted in the quotation.
- The applicable isolation depends on the deployment mode set out in the quotation; no dedicated isolation is presumed.
- Roles and permissions are configured and accepted for the identities and systems actually connected.
- The events logged, their content, their retention period and who may access them are defined for the deployment chosen.
For equipment covered by a patent or a manufacturing secret, SecNumCloud and reinforced hosting are options depending on your requirements. A single architecture is designed to answer both the GDPR and extraterritorial exposure. Designed for deployment in line with the GDPR and the AI Act, after the processing, roles and context-specific risks have been assessed.
Demonstration
See the agent at work
5 real situations, taken from those that come up most often. Pick one: the exchange unfolds as it would in your organisation.
A scripted demonstration. These exchanges show how the agent behaves — its sources, its refusals, what it leaves to your teams. Nothing is sent from this page, no model is queried here, and the matters named are fictional. That is precisely what we promise your data.
The behaviours shown here — monitoring, automation rules, routing and reminders — are configured with you during deployment, from your tools, your rules and your thresholds.
The architecture points named in these exchanges — location, local execution, isolation, encryption, role-based access, logging — are not a guarantee attached to the demonstration: they are those of the architecture set out in your quotation, and verified before commissioning.
The company in this demonstration
Fictional companySertane Industries — technical plastic injection moulding, subcontractor to the automotive and appliance industries
- Sector
- Technical plastic injection moulding — automotive and appliance subcontracting, single 11,000 m² site
- Headcount
- 240 employees, 11 of them in maintenance: 8 technicians on three shifts, 1 methods engineer, 1 parts storekeeper, 1 manager
- Customers
- 9 clients on multi-year contracts, 3 of which account for 61 % of revenue — with contractual late-delivery penalties
- Fleet and rhythm
- 212 pieces of equipment: 34 injection presses, 6 chillers, 4 compressors, 21 conveyors, 147 ancillaries — three shifts, 6 days a week
- Tools already in place
- CMMS with 6 years of history, press controllers, energy monitoring, 1,340 sensors already fitted, production ERP — the agent plugs into them, nothing is replaced
- Who decides
- The maintenance manager plans every shutdown; the production manager releases the windows; the site director rules on anything above €15,000
- Room for improvement
- 38 pieces of equipment out of 212 are genuinely tracked closely; tracking the signals takes 65 % of the methods engineer's time; 14 unplanned shutdowns last year, that is 96 hours of production lost
Sertane Industries has the skills and the sensors: 1,340 are already fitted, and the time to read them all is exactly what the agent brings. It runs on local inference on a machine at the site and plugs into the CMMS, the press controllers, the energy monitoring and the production ERP: it tracks the whole fleet, documents what moves, pulls out the precedents and proposes the windows — the maintenance manager plans and decides. The exchanges below cover six months, from the review of the fleet to the half-year report.
This company, its figures and the exchanges that follow were invented for the demonstration. They illustrate a common situation; they describe no real client.
Predictive maintenance · hosted in France
I went back over your 1,340 sensors and six years of CMMS this weekend — the software in which your technicians record every job — and the whole fleet has been under watch since this morning: 212 machines instead of 38.
What the review showed: of your 14 unplanned shutdowns last year, 9 were on equipment absent from your 38. Your list is a sound one, it is simply short: it is chosen on the value of the machine, and a €4,000 conveyor stops the same press as a €40,000 ram.
What those 14 shutdowns weighed: 96 hours of production, that is €110,400 at €1,150 of hourly margin, and 2 late-delivery penalties. That is the prize, and it is entirely within reach of the sensors already fitted.
What is live now, without one extra sensor: all 212 machines are tracked continuously, each on the quantities its own sensors already report. Nothing bought, nothing wired, nothing stopped — and the six maintenance uses are delivered together, from signal tracking to the proposed shutdown window.
What it gives you back: tracking the signals took 65 % of your methods engineer's time; it now takes 7 %. He moves from the man who watches the curves to the man who decides what to do about them, which is the post you created.
What I suggest: that the maintenance manager reads my list of 212 family by family — half an hour. Every line adjusts with a word, and each time I tell you what the machine will go on reporting by itself. fleet-review_212-machines.pdfWhat each machine already says, and what nobody was listening to
⛓ Sourced · the site's 1,340 sensors, 6 years of CMMS, production shutdown log, hourly costs from management accounting
What the review showed: of your 14 unplanned shutdowns last year, 9 were on equipment absent from your 38. Your list is a sound one, it is simply short: it is chosen on the value of the machine, and a €4,000 conveyor stops the same press as a €40,000 ram.
What those 14 shutdowns weighed: 96 hours of production, that is €110,400 at €1,150 of hourly margin, and 2 late-delivery penalties. That is the prize, and it is entirely within reach of the sensors already fitted.
What is live now, without one extra sensor: all 212 machines are tracked continuously, each on the quantities its own sensors already report. Nothing bought, nothing wired, nothing stopped — and the six maintenance uses are delivered together, from signal tracking to the proposed shutdown window.
What it gives you back: tracking the signals took 65 % of your methods engineer's time; it now takes 7 %. He moves from the man who watches the curves to the man who decides what to do about them, which is the post you created.
What I suggest: that the maintenance manager reads my list of 212 family by family — half an hour. Every line adjusts with a word, and each time I tell you what the machine will go on reporting by itself. fleet-review_212-machines.pdfWhat each machine already says, and what nobody was listening to
⛓ Sourced · the site's 1,340 sensors, 6 years of CMMS, production shutdown log, hourly costs from management accounting
We are told over and over that we must add instrumentation before hoping for anything. You are saying the opposite?
I am saying use what is fitted first: of your 1,340 sensors, 1,006 were already reporting a useful quantity, and it is now read every hour.
What they are enough to see, family by family:
· The 34 presses — oil temperature, injection pressure, cycle time, consumption, screw position. Cycle time alone catches 6 drifts out of 10, and your ERP has always recorded it.
· The 4 compressors — loaded running time rising 8 % at constant output announces an air leak or a tired valve six weeks before the failure.
· The 6 chillers — setpoint-to-actual gap and recovery time: four weeks of warning.
· The 21 conveyors — motor current, the most ordinary sensor on site and the most revealing: a bearing shows before it is heard.
The other 334 already work for you differently: 210 are on/off safety sensors doing their job, and 124 are due for repair. 31 of them sit on equipment I want to see: €4,900 in total, and it is the only purchase I recommend.
What I save you straight away: vibration signature monitoring on all 147 ancillaries — the fine measurement of a machine's vibrations, which reveals a worn bearing or gear before any other sign — costs €96,000 and 3 weeks of installation for equipment that accounts for 1 shutdown out of 14. Fit it instead on the 4 compressors and the 6 chillers: €11,200, and they account for 5 shutdowns out of 14 and 39 hours.
The order I suggest: the €4,900 now, the €11,200 revisited in three months on the records I will have built up. You will decide on your own measurements rather than on a quote. sensors_what-is-already-fitted.pdf1,006 usable sensors, 124 to repair, 0 mandatory
⛓ Sourced · inventory of the 1,340 sensors, controller logs, 6 years of CMMS, current instrumentation quotes
What they are enough to see, family by family:
· The 34 presses — oil temperature, injection pressure, cycle time, consumption, screw position. Cycle time alone catches 6 drifts out of 10, and your ERP has always recorded it.
· The 4 compressors — loaded running time rising 8 % at constant output announces an air leak or a tired valve six weeks before the failure.
· The 6 chillers — setpoint-to-actual gap and recovery time: four weeks of warning.
· The 21 conveyors — motor current, the most ordinary sensor on site and the most revealing: a bearing shows before it is heard.
The other 334 already work for you differently: 210 are on/off safety sensors doing their job, and 124 are due for repair. 31 of them sit on equipment I want to see: €4,900 in total, and it is the only purchase I recommend.
What I save you straight away: vibration signature monitoring on all 147 ancillaries — the fine measurement of a machine's vibrations, which reveals a worn bearing or gear before any other sign — costs €96,000 and 3 weeks of installation for equipment that accounts for 1 shutdown out of 14. Fit it instead on the 4 compressors and the 6 chillers: €11,200, and they account for 5 shutdowns out of 14 and 39 hours.
The order I suggest: the €4,900 now, the €11,200 revisited in three months on the records I will have built up. You will decide on your own measurements rather than on a quote. sensors_what-is-already-fitted.pdf1,006 usable sensors, 124 to repair, 0 mandatory
⛓ Sourced · inventory of the 1,340 sensors, controller logs, 6 years of CMMS, current instrumentation quotes
Our press curves and breakdown histories describe our production tooling better than a drawing does. Where does all that live?
With you, and nowhere else: I run on local inference on a machine at your site, inside your network.
Local inference means the model computes on your machine: a pressure curve or a breakdown report is processed without crossing any external network. For a site with no technical room, the other route gives the same guarantee: an isolated resource hosted in France, dedicated to your plant, with no pooling with another moulder.
You are right on the substance, and I deal with it first: your cycle times, your pressure curves and your settings are your know-how, and they remain your exclusive property.
The guarantees, line by line:
· Hosted in France, under French law, 0 data outside the European Union — architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity, including against a US operator hosting in Europe.
· What I learn from your presses stays with your presses: no data trains a model, yours or a third party's.
· Encryption in transit and at rest, and role-based access — rights follow the job: a shift technician sees his machines, the methods engineer sees the fleet, the manager sees everything, a contractor sees only the machine he is working on.
· A full log: who asked what, when, and what was approved.
How we got here: a one-day audit, free of charge, then design, integration and testing — three months, without stopping a press.
And the commercial argument, with dates: 2 of your 3 main customers already require hosting inside the European Union in their specification, and your last quality audit raised an observation on the traceability of your digital tools. You answer both in writing, architecture diagram in hand.
What I suggest: that I keep that architecture sheet current, continuously. It used to cost you three days a year; it will now cost you one read-through. technical-framework_where-your-curves-live.pdfLocal inference, role-based access, processing in the EU targeted
✎ Framework · deployment architecture, access log, supplier specifications of the 3 main customers
Local inference means the model computes on your machine: a pressure curve or a breakdown report is processed without crossing any external network. For a site with no technical room, the other route gives the same guarantee: an isolated resource hosted in France, dedicated to your plant, with no pooling with another moulder.
You are right on the substance, and I deal with it first: your cycle times, your pressure curves and your settings are your know-how, and they remain your exclusive property.
The guarantees, line by line:
· Hosted in France, under French law, 0 data outside the European Union — architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity, including against a US operator hosting in Europe.
· What I learn from your presses stays with your presses: no data trains a model, yours or a third party's.
· Encryption in transit and at rest, and role-based access — rights follow the job: a shift technician sees his machines, the methods engineer sees the fleet, the manager sees everything, a contractor sees only the machine he is working on.
· A full log: who asked what, when, and what was approved.
How we got here: a one-day audit, free of charge, then design, integration and testing — three months, without stopping a press.
And the commercial argument, with dates: 2 of your 3 main customers already require hosting inside the European Union in their specification, and your last quality audit raised an observation on the traceability of your digital tools. You answer both in writing, architecture diagram in hand.
What I suggest: that I keep that architecture sheet current, continuously. It used to cost you three days a year; it will now cost you one read-through. technical-framework_where-your-curves-live.pdfLocal inference, role-based access, processing in the EU targeted
✎ Framework · deployment architecture, access log, supplier specifications of the 3 main customers
A flag here is a technician saying “it is running hot”. What does one look like with you?
One page that reads in thirty seconds and holds everything needed to decide. Here is last night's, on press 17.
· What is moving: hydraulic oil temperature at 62.4 °C against a 54.1 °C average over the previous 90 days, at comparable load and workshop temperature — a steady rise for 19 days, +0.44 °C a day.
· The comparison period, stated plainly — the 90 days of running against which today's measurement is compared, excluding shutdowns and ramp-ups: 62 °C in August and 62 °C in January do not tell the same story, and the page says which.
· Your ranges: manufacturer alarm at 75 °C, controller trip at 82 °C, today's margin 12.6 °C. They are your values, displayed as they are.
· The projection: manufacturer alarm reached in 28 days, 22 at the earliest — place the job on the early date and you keep the whole margin.
· What Saturday's window earns you: 2 h 10 and €180 of parts, against 7 hours and €8,050 of margin for the same heat exchanger caught out.
And I put the cause on the table, quantified — it is the work nobody has time to do: across your 18,400 job reports, a rise of 0.3 to 0.5 °C a day came from a fouled heat exchanger 23 times out of 31, from a hydraulic pump 5 times — it climbs twice as fast — and from workshop cooling 3 times, in summer. So I put the exchanger first, at 23 out of 31, figure in hand.
The check that settles it takes ten minutes: the gap between exchanger inlet and outlet, above 6 °C it is fouling. The March 2024 procedure is attached, the part is in stock, the work order is written.
What it shifts in time: documenting a flag took 25 % of handling, about 40 minutes; it now takes 4 %, that is 6 minutes of reading. Say yes and the check goes out with Saturday's job sheet. flag_press-17_line-by-line.pdfMeasurements, comparison period, ranges, projection, cost of waiting
⛓ Sourced · press 17 controller, 90 days of measurements, manufacturer ranges, line hourly costs
· What is moving: hydraulic oil temperature at 62.4 °C against a 54.1 °C average over the previous 90 days, at comparable load and workshop temperature — a steady rise for 19 days, +0.44 °C a day.
· The comparison period, stated plainly — the 90 days of running against which today's measurement is compared, excluding shutdowns and ramp-ups: 62 °C in August and 62 °C in January do not tell the same story, and the page says which.
· Your ranges: manufacturer alarm at 75 °C, controller trip at 82 °C, today's margin 12.6 °C. They are your values, displayed as they are.
· The projection: manufacturer alarm reached in 28 days, 22 at the earliest — place the job on the early date and you keep the whole margin.
· What Saturday's window earns you: 2 h 10 and €180 of parts, against 7 hours and €8,050 of margin for the same heat exchanger caught out.
And I put the cause on the table, quantified — it is the work nobody has time to do: across your 18,400 job reports, a rise of 0.3 to 0.5 °C a day came from a fouled heat exchanger 23 times out of 31, from a hydraulic pump 5 times — it climbs twice as fast — and from workshop cooling 3 times, in summer. So I put the exchanger first, at 23 out of 31, figure in hand.
The check that settles it takes ten minutes: the gap between exchanger inlet and outlet, above 6 °C it is fouling. The March 2024 procedure is attached, the part is in stock, the work order is written.
What it shifts in time: documenting a flag took 25 % of handling, about 40 minutes; it now takes 4 %, that is 6 minutes of reading. Say yes and the check goes out with Saturday's job sheet. flag_press-17_line-by-line.pdfMeasurements, comparison period, ranges, projection, cost of waiting
⛓ Sourced · press 17 controller, 90 days of measurements, manufacturer ranges, line hourly costs
And on a machine we have just refurbished, or that we start on a new reference? Everything changes there.
That is where the comparison has to be fine, and it is: I split your measurements by regime, and I have identified 178 of them over six years.
A regime is a combination of part produced, rate and material: the same press does not have the same signature depending on what it is making. A press moving from an ABS housing to a glass-filled technical part changes temperature, pressure and cycle time without any wear being involved — and that is exactly what the split neutralises.
The three rules in force, all changeable with a word:
· After a refurbishment or a sub-assembly change, I exclude the first 30 days from the comparison period: a rebuilt machine deserves its own baseline.
· On a regime seen for less than 40 hours, I give you the measurement and its short history rather than an alert — you then know exactly what you are looking at.
· Ramp-ups and Monday-morning restarts are excluded: they pull every average without announcing anything.
Those three rules were born of a figure I publish myself, and which is already corrected: of the 96 flags of the half-year, 22 turned out to have nothing behind them, that is 23 % — 17 of the 22 came precisely from refurbishments and reference changes that my comparison period took in; the 30-day rule was written and applied straight away, and the rate is 7 % over the last eight weeks. Cause measured, correction applied, result verified, and not one euro spent.
What I suggest to bring it down further: one line in the CMMS at the moment of the refurbishment, which your technicians already fill in. It is the only thing I ask of you, and it is worth the 9 hours of checking of the half-year. comparison-periods_178-regimes.pdfWhat is compared with what, and what is deliberately excluded
⛓ Sourced · 6 years of CMMS, 178 regimes identified across the 34 presses, log of the 96 flags of the half-year
A regime is a combination of part produced, rate and material: the same press does not have the same signature depending on what it is making. A press moving from an ABS housing to a glass-filled technical part changes temperature, pressure and cycle time without any wear being involved — and that is exactly what the split neutralises.
The three rules in force, all changeable with a word:
· After a refurbishment or a sub-assembly change, I exclude the first 30 days from the comparison period: a rebuilt machine deserves its own baseline.
· On a regime seen for less than 40 hours, I give you the measurement and its short history rather than an alert — you then know exactly what you are looking at.
· Ramp-ups and Monday-morning restarts are excluded: they pull every average without announcing anything.
Those three rules were born of a figure I publish myself, and which is already corrected: of the 96 flags of the half-year, 22 turned out to have nothing behind them, that is 23 % — 17 of the 22 came precisely from refurbishments and reference changes that my comparison period took in; the 30-day rule was written and applied straight away, and the rate is 7 % over the last eight weeks. Cause measured, correction applied, result verified, and not one euro spent.
What I suggest to bring it down further: one line in the CMMS at the moment of the refurbishment, which your technicians already fill in. It is the only thing I ask of you, and it is worth the 9 hours of checking of the half-year. comparison-periods_178-regimes.pdfWhat is compared with what, and what is deliberately excluded
⛓ Sourced · 6 years of CMMS, 178 regimes identified across the 34 presses, log of the 96 flags of the half-year
96 flags in six months is still a lot. I have eight technicians, not thirty.
Your eight technicians only handled 25 of them: the 96 arrive sorted by criticality, and 71 were settled at Monday's meeting.
Criticality here is not the value of the machine: it is what its unavailability costs in production, given the line's workload and the customer contract being served.
How the 96 break down:
· 8 red — shutdown to be planned within 15 days, line committed to a contract carrying penalties. All 8 were planned, 0 unplanned stoppage.
· 17 amber — dealt with at the next weekly shutdown, 0 hours of production lost since Saturday is already down.
· 49 green — watched, reported on a single page at Monday's meeting.
· 22 checked and closed, with the cause and the correction I have just given you.
What your eight technicians gained, and it is the figure they quote first: they were called out 14 times a year, 9 of those at night or at the weekend. Over the half-year: 3 call-outs, 1 of them at night. The work moved from 3 a.m. on Sunday to 9 a.m. on Saturday, and 17 jobs settled into the weekly shutdown, where they cost not one minute of production.
What I suggest next: review each line's criticality at the monthly workload review, with the production manager. A line that comes under a penalty contract changes criticality on the day it is signed, and that day I move its machines to the front — that mechanism is what put the 8 red ones at the right moment. criticality_96-flags-sorted.pdf8 red, 17 amber, 49 green, 22 with nothing behind them
⛓ Sourced · 96 flags of the half-year, ERP workload plan, customer contracts, call-out log
Criticality here is not the value of the machine: it is what its unavailability costs in production, given the line's workload and the customer contract being served.
How the 96 break down:
· 8 red — shutdown to be planned within 15 days, line committed to a contract carrying penalties. All 8 were planned, 0 unplanned stoppage.
· 17 amber — dealt with at the next weekly shutdown, 0 hours of production lost since Saturday is already down.
· 49 green — watched, reported on a single page at Monday's meeting.
· 22 checked and closed, with the cause and the correction I have just given you.
What your eight technicians gained, and it is the figure they quote first: they were called out 14 times a year, 9 of those at night or at the weekend. Over the half-year: 3 call-outs, 1 of them at night. The work moved from 3 a.m. on Sunday to 9 a.m. on Saturday, and 17 jobs settled into the weekly shutdown, where they cost not one minute of production.
What I suggest next: review each line's criticality at the monthly workload review, with the production manager. A line that comes under a penalty contract changes criticality on the day it is signed, and that day I move its machines to the front — that mechanism is what put the 8 red ones at the right moment. criticality_96-flags-sorted.pdf8 red, 17 amber, 49 green, 22 with nothing behind them
⛓ Sourced · 96 flags of the half-year, ERP workload plan, customer contracts, call-out log
Our old hands know which symptom announces what. Two of them retire next year.
What they know is already written down, in your 18,400 free-text job reports — and it can now be questioned in one sentence by any of your technicians.
What I pull out at every flag: the past jobs whose signals looked alike, with what was found, what was changed, the time taken and the cost.
On last night's press 17, the three closest:
· Press 22, March 2024 — same rise, +0.39 °C a day over 24 days, fouled heat exchanger. 2 h 10, €180 of parts. Report: “scaling and fouling on the water side”.
· Press 9, November 2023 — +0.9 °C a day, hydraulic pump. 6 h, €4,300. The rate of rise separates the two cases, and it is the most useful information on the page.
· Press 17 itself, June 2022 — +0.41 °C a day, heat exchanger. Same machine, same symptom: the most telling of the three.
What I conclude, quantified: at +0.44 °C a day, we are on the “exchanger” case, three hundredths from its own history. The ten-minute check confirms it on the spot, and it costs less than the question.
What it shifts in time: finding the precedents took 40 % of the handling of a flag; it now takes 5 %. Your methods engineer used to spend Friday afternoons on it; he spends six minutes.
And what it keeps for you: your two old hands' knowledge stays available, in a second, to a technician who arrived last month.
What I suggest before they go: that they reread with me the 40 most consulted precedents and add what the report does not say. Forty sheets, two half-days, and the archive is complete. comparable-precedents_press-17.pdf3 precedents, what separates them, what they cost
⛓ Sourced · 18,400 job reports over 6 years, stores part costs, job times recorded in the CMMS
What I pull out at every flag: the past jobs whose signals looked alike, with what was found, what was changed, the time taken and the cost.
On last night's press 17, the three closest:
· Press 22, March 2024 — same rise, +0.39 °C a day over 24 days, fouled heat exchanger. 2 h 10, €180 of parts. Report: “scaling and fouling on the water side”.
· Press 9, November 2023 — +0.9 °C a day, hydraulic pump. 6 h, €4,300. The rate of rise separates the two cases, and it is the most useful information on the page.
· Press 17 itself, June 2022 — +0.41 °C a day, heat exchanger. Same machine, same symptom: the most telling of the three.
What I conclude, quantified: at +0.44 °C a day, we are on the “exchanger” case, three hundredths from its own history. The ten-minute check confirms it on the spot, and it costs less than the question.
What it shifts in time: finding the precedents took 40 % of the handling of a flag; it now takes 5 %. Your methods engineer used to spend Friday afternoons on it; he spends six minutes.
And what it keeps for you: your two old hands' knowledge stays available, in a second, to a technician who arrived last month.
What I suggest before they go: that they reread with me the 40 most consulted precedents and add what the report does not say. Forty sheets, two half-days, and the archive is complete. comparable-precedents_press-17.pdf3 precedents, what separates them, what they cost
⛓ Sourced · 18,400 job reports over 6 years, stores part costs, job times recorded in the CMMS
And our calendar-based maintenance plan? We change parts on a schedule, it reassures us.
It protects you, and I have made it cheaper without touching its cover: of 68 parts removed to schedule last year, 41 still had more than 40 % of their life left.
Calendar-based maintenance changes a part at a fixed interval; condition-based maintenance changes it when its condition calls for it. Both have their place, and the optimum is the mix of the two.
What your own inspection sheets at removal show: 68 valves and cartridges removed over the year, 41 still good, at €380 a part, that is €15,580 of material and 31 hours of labour on components that were running well. Those are your workshop records, not my estimate.
The revision I have written, family by family:
· 3 valve families move from 6 to 9 months, watching the differential pressure your sensors already report: − €9,400 a year, identical cover.
· 2 belt references move from 12 to 8 months: they have failed before their due date twice each. + €1,800 a year, and those failures disappear.
· Safety devices and the 12 machines with no condition measurement: calendar unchanged. There the calendar is the right protection, and it delivers it well.
The net: €7,600 a year, with better cover than today.
The order I recommend: sign the tightening of the two belt families first — it protects from next week. The three lengthened families will bring you the €9,400 over twelve months, and you will judge them on the differential-pressure records I will have built up by then. calendar-plan_what-leaves-too-early.pdf41 parts out of 68, €15,580, and the 2 families to tighten
⛓ Sourced · calendar maintenance plan, 68 inspection sheets at removal, stores prices, failure history
Calendar-based maintenance changes a part at a fixed interval; condition-based maintenance changes it when its condition calls for it. Both have their place, and the optimum is the mix of the two.
What your own inspection sheets at removal show: 68 valves and cartridges removed over the year, 41 still good, at €380 a part, that is €15,580 of material and 31 hours of labour on components that were running well. Those are your workshop records, not my estimate.
The revision I have written, family by family:
· 3 valve families move from 6 to 9 months, watching the differential pressure your sensors already report: − €9,400 a year, identical cover.
· 2 belt references move from 12 to 8 months: they have failed before their due date twice each. + €1,800 a year, and those failures disappear.
· Safety devices and the 12 machines with no condition measurement: calendar unchanged. There the calendar is the right protection, and it delivers it well.
The net: €7,600 a year, with better cover than today.
The order I recommend: sign the tightening of the two belt families first — it protects from next week. The three lengthened families will bring you the €9,400 over twelve months, and you will judge them on the differential-pressure records I will have built up by then. calendar-plan_what-leaves-too-early.pdf41 parts out of 68, €15,580, and the 2 families to tighten
⛓ Sourced · calendar maintenance plan, 68 inspection sheets at removal, stores prices, failure history
And if I ask you who was at the controls when the drift started? We have differences between shifts.
I produce it, and the file that makes it usable is already written: definition of the indicator, information note, consultation material. You sign, it runs the next day.
The three conditions, and the papers that meet them: prior information of the employees concerned, consultation of the staff representatives before it is brought in, an indicator proportionate to the aim pursued. With 240 employees, everything is in place.
What I recommend trying first, quantified: a per-shift indicator becomes a target within three months — your scrap indicator fell 18 % in 2023 while customer returns rose —, and the same question is settled on the machine, where the figure cannot be steered.
Because comparing your 178 regimes, I found something more actionable: on 6 presses, the mould temperature setting varies by 4 to 7 °C between shifts, for the same reference. The sheet gives a range, and each person sits where they were taught; those at the top of the range consume 6 % more and run the oil 3 °C hotter.
What that is worth: €11,600 of electricity a year on those six presses, and 3 °C less thermal stress — enough to push back the exchanger drift we were discussing. And it corrects a sheet, not a person.
One single thing falls to the law, and the lawful path is right beside it: inferring an operator's emotional state or fatigue from a camera or a microphone, which the European regulation on artificial intelligence prohibits at work — and the result you are after I get from the machines: cycle times, micro-stops and setting corrections by time of day say at what hour the process slips, and what is then corrected is a sheet or a handover.
What I propose, in order: the narrowed sheet on the 6 references is written and waiting for the methods technician; one month of measurement confirms the €11,600; and the per-shift indicator file goes out tomorrow on one word from you.
✎ Framework · 178 regimes compared, setting sheets, consumption records per press, history of the 2023 scrap indicator
The three conditions, and the papers that meet them: prior information of the employees concerned, consultation of the staff representatives before it is brought in, an indicator proportionate to the aim pursued. With 240 employees, everything is in place.
What I recommend trying first, quantified: a per-shift indicator becomes a target within three months — your scrap indicator fell 18 % in 2023 while customer returns rose —, and the same question is settled on the machine, where the figure cannot be steered.
Because comparing your 178 regimes, I found something more actionable: on 6 presses, the mould temperature setting varies by 4 to 7 °C between shifts, for the same reference. The sheet gives a range, and each person sits where they were taught; those at the top of the range consume 6 % more and run the oil 3 °C hotter.
What that is worth: €11,600 of electricity a year on those six presses, and 3 °C less thermal stress — enough to push back the exchanger drift we were discussing. And it corrects a sheet, not a person.
One single thing falls to the law, and the lawful path is right beside it: inferring an operator's emotional state or fatigue from a camera or a microphone, which the European regulation on artificial intelligence prohibits at work — and the result you are after I get from the machines: cycle times, micro-stops and setting corrections by time of day say at what hour the process slips, and what is then corrected is a sheet or a handover.
What I propose, in order: the narrowed sheet on the 6 references is written and waiting for the methods technician; one month of measurement confirms the €11,600; and the per-shift indicator file goes out tomorrow on one word from you.
✎ Framework · 178 regimes compared, setting sheets, consumption records per press, history of the 2023 scrap indicator
The hard part is not knowing we must intervene. It is finding when, without stopping a line that is running.
Three windows for press 17, each costed and each workable: the choice is yours, and it takes three minutes.
A production window is a slot in which the line can stop without missing a customer commitment: the weekly shutdown, a changeover or a dip in workload.
The three scenarios, taken from your workload plan:
· Saturday the 14th, weekly shutdown. €0 of production, 2 h 10 from the press 22 precedent, 2 parts in stock, 22 days of margin on the projection. This is the one I recommend.
· Thursday the 26th, a 6-hour changeover. €0 of extra production, and I place the job at the head of the changeover sheet: 2 h 10 out of 6 h leave 3 h 50 of margin to the same technician.
· The 11th of next month, shutdown already planned. €0 immediately, and I secure it with a daily reading of the exchanger inlet-outlet gap: at the first point above 6 °C, I move it to the Saturday before, which is free in the workload plan.
The comparison that decides: the Saturday job costs 2 h 10, €180 of parts and €0 of production; the same exchanger caught out costs 7 hours, €8,050 of margin and a material rework — that is the press 9 precedent, invoice attached.
What is already prepared: the work order drafted in the CMMS, the two parts reserved in stores, the March 2024 procedure attached, and the note to the production manager in his usual format.
What carries your signature, and it is one click: blocking the window, because a slot taken out of the workload plan commits a line and customers. You tick Saturday the 14th and it all goes together: order approved, parts issued, production notified, gap check added to the job sheet. Thirty seconds from you, and €8,050 that stays with you. shutdown-windows_press-17_3-scenarios.pdfWhat each window costs, and what it risks
⛓ Sourced · ERP workload plan, customer contracts and commitments, stores inventory, CMMS precedents
A production window is a slot in which the line can stop without missing a customer commitment: the weekly shutdown, a changeover or a dip in workload.
The three scenarios, taken from your workload plan:
· Saturday the 14th, weekly shutdown. €0 of production, 2 h 10 from the press 22 precedent, 2 parts in stock, 22 days of margin on the projection. This is the one I recommend.
· Thursday the 26th, a 6-hour changeover. €0 of extra production, and I place the job at the head of the changeover sheet: 2 h 10 out of 6 h leave 3 h 50 of margin to the same technician.
· The 11th of next month, shutdown already planned. €0 immediately, and I secure it with a daily reading of the exchanger inlet-outlet gap: at the first point above 6 °C, I move it to the Saturday before, which is free in the workload plan.
The comparison that decides: the Saturday job costs 2 h 10, €180 of parts and €0 of production; the same exchanger caught out costs 7 hours, €8,050 of margin and a material rework — that is the press 9 precedent, invoice attached.
What is already prepared: the work order drafted in the CMMS, the two parts reserved in stores, the March 2024 procedure attached, and the note to the production manager in his usual format.
What carries your signature, and it is one click: blocking the window, because a slot taken out of the workload plan commits a line and customers. You tick Saturday the 14th and it all goes together: order approved, parts issued, production notified, gap check added to the job sheet. Thirty seconds from you, and €8,050 that stays with you. shutdown-windows_press-17_3-scenarios.pdfWhat each window costs, and what it risks
⛓ Sourced · ERP workload plan, customer contracts and commitments, stores inventory, CMMS precedents
The real trap is the part. We once turned a 2-hour job into a 9-hour one because it was not there.
Six times last half-year, zero this one — and what tipped it is a mandate you gave me and that you take back with a word.
What your stores said: 1,940 references, of which 310 have not moved in three years and 47 regularly absent at the moment of intervening. The stock was not too small: it was badly distributed, and that is a free correction.
What I do on my own, and it commits nothing: at each flag I check the likely part and reserve it in stores. A reservation buys nothing and releases itself within 15 days.
What commits your money carries a mandate, and here it is:
· Scope: the references in the maintenance plan, from your 4 approved suppliers.
· Caps: €800 per reference, €6,000 a month, automatic stop at the cap.
· Condition: stock below the threshold you set, or a reference attached to a red or amber flag.
· Term: 3 months, renewable. Withdrawal with one word, immediate effect.
· Keep your signature: a part outside the plan, an unapproved supplier, an express order at a premium price, anything above the cap.
What the mandate produced: 0 shortages at the moment of intervening against 6, that is 42 hours of stoppage and €48,300 of margin that stayed with you. And the stock fell 4 %, the 310 dormant references no longer being replenished.
The next step I suggest: an option — free, cancellable 48 hours ahead — on a slot with your machining subcontractor as soon as a red flag lands on equipment you do not repair in house. Over the past half-year it would have saved you 9 days on two jobs. sourcing-mandate_capped-and-dated.pdf€800 per reference, €6,000 a month, withdrawn with a word
⛓ Sourced · stores inventory (1,940 references), shortage history, prices from the 4 approved suppliers, shutdown log
What your stores said: 1,940 references, of which 310 have not moved in three years and 47 regularly absent at the moment of intervening. The stock was not too small: it was badly distributed, and that is a free correction.
What I do on my own, and it commits nothing: at each flag I check the likely part and reserve it in stores. A reservation buys nothing and releases itself within 15 days.
What commits your money carries a mandate, and here it is:
· Scope: the references in the maintenance plan, from your 4 approved suppliers.
· Caps: €800 per reference, €6,000 a month, automatic stop at the cap.
· Condition: stock below the threshold you set, or a reference attached to a red or amber flag.
· Term: 3 months, renewable. Withdrawal with one word, immediate effect.
· Keep your signature: a part outside the plan, an unapproved supplier, an express order at a premium price, anything above the cap.
What the mandate produced: 0 shortages at the moment of intervening against 6, that is 42 hours of stoppage and €48,300 of margin that stayed with you. And the stock fell 4 %, the 310 dormant references no longer being replenished.
The next step I suggest: an option — free, cancellable 48 hours ahead — on a slot with your machining subcontractor as soon as a red flag lands on equipment you do not repair in house. Over the past half-year it would have saved you 9 days on two jobs. sourcing-mandate_capped-and-dated.pdf€800 per reference, €6,000 a month, withdrawn with a word
⛓ Sourced · stores inventory (1,940 references), shortage history, prices from the 4 approved suppliers, shutdown log
And the day a flag lands at 3 a.m. on a Sunday, on a line carrying penalties?
It goes to the on-call technician with enough to settle it in five minutes from a phone — and twice out of three this half-year, the answer was “it holds until morning”.
What he receives, at 3:12 a.m.: the machine, the quantity that is moving, the measurement and its comparison, the manufacturer range, the closest precedent with its job duration, the stock position of the likely part, and two costed options: go in now, or hold until 6 a.m.
What makes the second option safe: I give the margin, not just the alert. “Controller trip at 82 °C, reading at 71.4 °C, rise of 0.8 °C an hour: there are 13 hours left” is settled in one sentence. Of the 3 call-outs this half-year, 2 waited for the morning shift on that calculation, and each turned 7 hours of Sunday into 2 h 10 of Saturday.
What I prepare meanwhile: the work order is drafted and the part reserved before the technician arrives at 6 a.m. — twenty minutes less in stores for him.
The guarantee that makes on-call comfortable: my connection to your controllers is read-only, and the separation is physical. Stopping a machine, changing a controller parameter, altering a setpoint: your technicians and your safety functions keep exclusive control of those. That guarantee is what let you wire me into 34 presses without reopening your risk assessment, and it is every auditor's first question.
The figure for the half-year: 3 call-outs against 8, 1 at night against 5, 0 unplanned stoppages on the 8 red flags.
What I suggest: one hour reviewing the on-call arrangement with your 8 technicians. They know which line to add to the 3 a.m. message, and every night handed back is worth 4 h 50 of job time saved.
⛓ Sourced · on-call alert log for the half-year, work orders generated, comparison with the previous half-year
What he receives, at 3:12 a.m.: the machine, the quantity that is moving, the measurement and its comparison, the manufacturer range, the closest precedent with its job duration, the stock position of the likely part, and two costed options: go in now, or hold until 6 a.m.
What makes the second option safe: I give the margin, not just the alert. “Controller trip at 82 °C, reading at 71.4 °C, rise of 0.8 °C an hour: there are 13 hours left” is settled in one sentence. Of the 3 call-outs this half-year, 2 waited for the morning shift on that calculation, and each turned 7 hours of Sunday into 2 h 10 of Saturday.
What I prepare meanwhile: the work order is drafted and the part reserved before the technician arrives at 6 a.m. — twenty minutes less in stores for him.
The guarantee that makes on-call comfortable: my connection to your controllers is read-only, and the separation is physical. Stopping a machine, changing a controller parameter, altering a setpoint: your technicians and your safety functions keep exclusive control of those. That guarantee is what let you wire me into 34 presses without reopening your risk assessment, and it is every auditor's first question.
The figure for the half-year: 3 call-outs against 8, 1 at night against 5, 0 unplanned stoppages on the 8 red flags.
What I suggest: one hour reviewing the on-call arrangement with your 8 technicians. They know which line to add to the 3 a.m. message, and every night handed back is worth 4 h 50 of job time saved.
⛓ Sourced · on-call alert log for the half-year, work orders generated, comparison with the previous half-year
Six months. What has it earned us, in euros?
Four lines, and the fourth arrives with its correction already measured.
1 — Unplanned shutdowns. 3 this half-year against 8 in the comparable half-year, 17 hours lost against 54. 37 hours avoided, at €1,150 of hourly margin: €42,550, readable in your shutdown log rather than in mine.
2 — Parts shortages at the moment of intervening. 0 against 6, that is 42 hours and €48,300 of margin kept with you — and a stock down 4 %.
3 — The calendar plan. €7,600 a year net, once the two tightened families are paid for.
4 — Time. Tracking the signals 65 % → 7 %, finding the precedents 40 % → 5 %, documenting a flag 25 % → 4 %, that is 40 minutes down to 6. And 212 machines tracked instead of 38, with no hiring and no extra sensor — the figure I like best.
The figure I publish myself, with what I did about it: 22 flags out of 96 turned out to have nothing behind them, 23 %, at 25 minutes of checking each — 9 hours. 17 of the 22 were on machines refurbished or moved to a new reference that my comparison period took in; I now exclude the first 30 days and no longer flag under 40 hours of regime, and the rate is 7 % over the last eight weeks. Correction written, applied, measured, at zero cost — and those 7 % are the price of detection at three weeks rather than three days.
What I suggest next: one CMMS line at every refurbishment. It is the only thing I ask of you across the whole half-year, and it is worth the 9 hours. half-year-report_42550-euros-and-23-percent.pdfWhat was avoided, what was got wrong, what was corrected
⛓ Sourced · shutdown log across 2 half-years, CMMS, stores inventory, maintenance department time records
1 — Unplanned shutdowns. 3 this half-year against 8 in the comparable half-year, 17 hours lost against 54. 37 hours avoided, at €1,150 of hourly margin: €42,550, readable in your shutdown log rather than in mine.
2 — Parts shortages at the moment of intervening. 0 against 6, that is 42 hours and €48,300 of margin kept with you — and a stock down 4 %.
3 — The calendar plan. €7,600 a year net, once the two tightened families are paid for.
4 — Time. Tracking the signals 65 % → 7 %, finding the precedents 40 % → 5 %, documenting a flag 25 % → 4 %, that is 40 minutes down to 6. And 212 machines tracked instead of 38, with no hiring and no extra sensor — the figure I like best.
The figure I publish myself, with what I did about it: 22 flags out of 96 turned out to have nothing behind them, 23 %, at 25 minutes of checking each — 9 hours. 17 of the 22 were on machines refurbished or moved to a new reference that my comparison period took in; I now exclude the first 30 days and no longer flag under 40 hours of regime, and the rate is 7 % over the last eight weeks. Correction written, applied, measured, at zero cost — and those 7 % are the price of detection at three weeks rather than three days.
What I suggest next: one CMMS line at every refurbishment. It is the only thing I ask of you across the whole half-year, and it is worth the 9 hours. half-year-report_42550-euros-and-23-percent.pdfWhat was avoided, what was got wrong, what was corrected
⛓ Sourced · shutdown log across 2 half-years, CMMS, stores inventory, maintenance department time records
What do you do without asking us? I want the list, and I want to know how we cancel it.
Three actions, not one more, and each undoes as fast as it does.
· I reserve in stores the likely part for a red or amber flag. A reservation buys nothing, releases itself within 15 days, and if you release the part by hand, it stays free.
· I draft the work order in the CMMS, as a draft, marked “proposed by the agent, not approved”, visible only to the human who opens it.
· I alert the on-call technician when a quantity crosses a range you have set, with the margin remaining. It is governed by the threshold, and the threshold is yours.
And six acts carry a signature, with the file built in advance: block a production window · trigger a job · stop a machine · write into a controller · order beyond the mandate · change an operating range. Six acts, six people's names — and the signature takes a minute because everything else is already done.
The result: 0 decision taken without human approval across the 96 flags of the half-year. That is how I am wired in: I read the controllers, writing to them is not wired, and that separation is physical.
And for your customer audits, which is the real reason for your question: every flag keeps its measurements, its comparison period, the precedents supplied, what was proposed and what was decided, with the name and the time. Two of your three main customers audit your maintenance every year: the file comes out in a minute instead of a day.
What I suggest: a quarterly review of this list — an automatic action that has served no purpose in three months comes off it, and its shortness is what makes it unanswerable to an auditor. who-decides-what_three-actions-six-limits.pdfThree reversible actions, six reserved acts, 0 writes to a controller
✎ Framework · list of automatic actions approved by the maintenance manager, approval log for the half-year
· I reserve in stores the likely part for a red or amber flag. A reservation buys nothing, releases itself within 15 days, and if you release the part by hand, it stays free.
· I draft the work order in the CMMS, as a draft, marked “proposed by the agent, not approved”, visible only to the human who opens it.
· I alert the on-call technician when a quantity crosses a range you have set, with the margin remaining. It is governed by the threshold, and the threshold is yours.
And six acts carry a signature, with the file built in advance: block a production window · trigger a job · stop a machine · write into a controller · order beyond the mandate · change an operating range. Six acts, six people's names — and the signature takes a minute because everything else is already done.
The result: 0 decision taken without human approval across the 96 flags of the half-year. That is how I am wired in: I read the controllers, writing to them is not wired, and that separation is physical.
And for your customer audits, which is the real reason for your question: every flag keeps its measurements, its comparison period, the precedents supplied, what was proposed and what was decided, with the name and the time. Two of your three main customers audit your maintenance every year: the file comes out in a minute instead of a day.
What I suggest: a quarterly review of this list — an automatic action that has served no purpose in three months comes off it, and its shortness is what makes it unanswerable to an auditor. who-decides-what_three-actions-six-limits.pdfThree reversible actions, six reserved acts, 0 writes to a controller
✎ Framework · list of automatic actions approved by the maintenance manager, approval log for the half-year
We still have a blind spot: drift that shows on the part produced, not on the machine.
Half of it is already in my logs: over the half-year, 3 of your 5 scrap peaks fell within the 72 hours following a flag raised on the press concerned. The machine drifts before the part does, and it is your own data that shows it.
What I do with it from Monday: I match my 96 flags against your dimensional records over the same six months. That matching runs on the history, with nothing to install: you have it Monday, peak by peak, with the delay between the machine signal and the first out-of-tolerance point.
What I recommend next, quantified:
· A production quality monitoring agent, on dimensions and process parameters: it would see the drift on the part when I see it on the machine, and the two together give the cause. Your 5 peaks this half-year weigh 118,000 parts and €21,400 of material.
· A non-conformity tracking agent for what follows: your 3 main customers require an action plan within 10 working days, your average is 17 days, and that agent brings the 17 under 10, with the evidence attached.
What I go on holding, myself: the condition of your 212 machines, the documented flags, the precedents and the windows. Each to its own ground, and the two talk through your data.
And the argument to keep for your customer reviews: all of this lives in France, on a machine at your site, with the deployment objective of processing and access operated within the European Union. Your curves, your cycle times and your settings — your know-how — stay in the plant. At the next renewal you will be the only one on the list able to show it in writing.
Say the word and the matching is on your desk on Monday morning.
⛓ Sourced · scrap log for the half-year, flags raised, non-conformity action plan lead times, customer specifications
What I do with it from Monday: I match my 96 flags against your dimensional records over the same six months. That matching runs on the history, with nothing to install: you have it Monday, peak by peak, with the delay between the machine signal and the first out-of-tolerance point.
What I recommend next, quantified:
· A production quality monitoring agent, on dimensions and process parameters: it would see the drift on the part when I see it on the machine, and the two together give the cause. Your 5 peaks this half-year weigh 118,000 parts and €21,400 of material.
· A non-conformity tracking agent for what follows: your 3 main customers require an action plan within 10 working days, your average is 17 days, and that agent brings the 17 under 10, with the evidence attached.
What I go on holding, myself: the condition of your 212 machines, the documented flags, the precedents and the windows. Each to its own ground, and the two talk through your data.
And the argument to keep for your customer reviews: all of this lives in France, on a machine at your site, with the deployment objective of processing and access operated within the European Union. Your curves, your cycle times and your settings — your know-how — stay in the plant. At the next renewal you will be the only one on the list able to show it in writing.
Say the word and the matching is on your desk on Monday morning.
⛓ Sourced · scrap log for the half-year, flags raised, non-conformity action plan lead times, customer specifications
Local inference · no data outside the EU
Your case is not here? That is exactly what a 15-minute conversation is for. Book the free audit →
Use cases
What does the agent actually do?
One agent, several angles on maintenance. All these uses work in support, subject to your approval.
Included in your agent
The 3 capabilities essential to this promise are included, at no extra cost.
From 693 € excl. VAT / month
Tracking the signals
Gathers the measurements from your sensors across the whole fleet.
Documented flags
Flags movements away from the equipment's usual behaviour.
Comparable precedents
Finds the past interventions on similar signals.
Controls and safeguards
These 3 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
Explain alerts, scenarios and operational trade-offs
What the agent must be connected to
These 2 connections are required for the agent to work. They concern your information system and are scoped during the audit.
Connect ERP/WMS/MES/IoT without creating double entry
Honour constraints, statuses, approvals and error recovery
Need to go further?
These agents handle a different business process, with their own owner and their own price. They are added to this one.
Does your need fall outside this?
Book the free audit →
Build your agent
In 15 minutes we identify the most relevant agent — without oversizing the project.
The gain
How much equipment can a team follow closely?
By taking on the tracking of signals, the effort shifts towards scheduling and intervening. How large the gain is depends on your volume and remains to be confirmed by a pilot.
Tracking signals across the fleet
Finding the precedents
Documenting an alert
Indicative figures, not contractual, to be confirmed by a pilot on your number of items tracked and the density of your sensors. Scheduling a shutdown or triggering an intervention commits production and safety: those decisions belong to the maintenance manager.
How it works
The stages of your AI agent project
1
Audit & scoping
15 minutes to target the use case with the best return.
2
Quote or direct sign-up
A catalogue offer is bought online; a specific need gets a costed quote.
3
Design
We design the agent and its guardrails.
4
Integration & testing
We connect your tools to the agent, which is itself hosted in France.
5
Rollout
Going live and training your team.
6
Operation
Continuous supervision and improvement.
Pricing
One package, one agent
A predictive maintenance agent (signals, precedents, documentation), installed and operated for you.
Agility
Setup + controlled subscription
10,775 € excl. VAT setup
then 693 € excl. VAT/month — you invest at installation and pay a reduced subscription. Ideal for keeping the cost under control over time.
- Installation, configuration and training for your teams
- Operation, human oversight, updates and support
- Sovereign hosting in France, a dedicated and isolated resource
The simplest
Serenity
All inclusive, no setup fee
1,293 € excl. VAT /month
all inclusive, immediate start. No upfront investment: a single subscription. Ideal for starting quickly and simply.
- Setup included (installation, configuration, training)
- Operation, human oversight, updates and support
- Sovereign hosting in France, managed end to end
100% Sovereign
On site, you own it
15,000 € excl. VAT setup
then 927 € excl. VAT/month · + hardware from 5,500 € (one-off purchase, in addition) — a sovereign computer installed on your premises, maintained remotely. Models run locally, your data returned at the end of the contract. 36-month commitment.
- Hardware installed on your premises (you own it)
- French / European AI models run locally
- Secure remote maintenance (Pro support included)
Not included in the packages: AI consumption (model tokens), re-invoiced at real cost with no margin, and tracked in real time in your client area. Maintenance and supervision subscription for an initial term of 12 months for the Agility package, 24 months for the Serenity package and 36 months for the 100% Sovereign package, renewable; support levels (SLA 72 h / 24 h / 4 h) optional. Bespoke development, additional integrations or exceptional volumes are quoted separately. Support Monday to Friday, 9am to 6pm. Prices exclude VAT.
AI model:
none of the AI models offered currently carries a fixed surcharge. When the selected model carries a cost, that cost is shown when you choose it, before you order, and re-invoiced at the cost incurred, with no mark-up; usage is billed at the publisher's price. Publishers' prices are published in US dollars: the amount re-invoiced is the amount in euros actually borne by Blue Lemon Agent on the publisher's invoice, at that invoice's exchange rate, with no commission or mark-up.
Included components and additional components Components included in the base offer: the Blue Lemon Agent software foundation, the AI models listed in the order journey, the standard channels (Microsoft Teams, Slack, WhatsApp Business, email, website chat, calendars, Microsoft 365 / Google Workspace, file storage, market VoIP telephony, professional social-media pages and accounts, Google Business Profile), hosting in France for the package chosen, backups, supervision, updates and support. If adapting the AI agent to your constraints, your needs or your requests requires other paid components — a third-party publisher's software licence, paid API access to one of your applications, hosting of health data, for which French law requires an HDS-certified host (art. L. 1111-8 of the French Public Health Code), SecNumCloud-qualified hosting, a speech synthesis service, particular hardware —, they are offered to you as an option or on quotation and re-invoiced at the cost incurred; nothing is committed without your written agreement. Where the artificial intelligence model you choose entails an additional cost, that cost is shown to you before you order and re-invoiced to you at the cost incurred, with no margin.
What to expect
Go-live
2 to 3 weeks
Agent designed, channels connected, team trained.
Steady state
4 to 7 weeks
After a few weeks of real use, once the agent's behaviour matches what you expect.
Indicative estimate, adjusted to the options you keep. It is not a delivery commitment.
Our commitment
Four guarantees that matter to your industrial equipment
Your industrial data stays with youLocal inference or an isolated resource hosted in France; no sensor data and no history entrusted to a third party, no data used to train a model.
Data in France, under French lawYour industrial data and your failure history: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The maintenance manager keeps the decisionThe agent produces documented intervention recommendations, which can be checked and altered; no approval is automated.
Human oversight & traceabilityOn your number of items tracked and the density of your sensors: systematic logging and tracking, in line with the AI Act.
Going further
Related resources
Frequently asked questions
Your questions, our answers
Does the agent trigger interventions?
No. It flags, documents and provides the comparable precedents. Scheduling a shutdown commits production and safety: the decision falls to the maintenance manager.
What are the flags based on?
On how the measurements are moving against the equipment's usual behaviour, with the comparison period shown and the past interventions on similar signals.
Do we need additional sensors?
The agent draws on the sensors already in place. Whether additional sensors would help is assessed during the audit, not presumed.
How does this differ from quality monitoring?
Quality monitoring follows process drift on the parts produced; predictive maintenance follows the condition of the equipment itself.
Is our industrial data protected?
Yes. The agent is hosted in France, on local inference or an isolated resource, with the deployment objective of processing and access operated within the European Union and an architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. Your data is not used to train a third-party model.
How long does it take to deploy this agent?
Several months as a rule, depending on the number of items, the density of the sensors and the depth of the history, after a free audit then phases of design, integration and testing.
Related use cases
Other agents for industry
Let's talk
Let's size up the potential across your equipment fleet
15 minutes to frame your equipment and your sensors — hosted in France, supervised, with no commitment.