Today
In maintenance, every work order is free text: “gripper sticks”, “zone 7 cold”, “sensor replaced”. Nobody can say which machines and components eat the most repair hours without reading thousands of orders. So gut feeling decides, and the same weak point is repaired for years instead of being removed once.
An example
At Polivo Kunststoffverpackungen, the agent reads about 6,300 maintenance orders from three years for 41 injection moulding, stretch blow moulding and thermoforming machines and sorts them into 14 damage types. At the top is the 32-cavity mould for the Ø 38 mm closure: a heating zone in the hot runner fails every nine days on average, on the two identical moulds every 70 days, and each repair takes 4.5 hours on average. That adds up to about 180 hours of downtime and 26,000 euros of repairs a year, almost always a thermocouple whose cable breaks at the connector. According to the mould maker's quote, a new cable set costs 8,400 euros; the maintenance manager approves the conversion for the next mould change.
Your data, wherever it lives
Your maintenance history sits today as free text in the work orders of the maintenance system, spare parts and costs in the ERP, run times and stops in the MES, and much of it only in shift logs and spreadsheets. demi reads, understands and links it per machine and component, with your rights. The AI works out in minutes which machines cost you the most hours, a job that otherwise means weeks of counting orders, and your maintenance manager decides what gets improved.
Where the data lives today
- Tickets
- ERP
- MES
- Office files
How demi solves it
Three layers in this order: connect the data first, then the rights, then the AI.
- 1 · ConnectData Layer
Data sources in this case: the maintenance system, the ERP, the MES and shift logs kept in spreadsheets. demi replaces nothing and joins, for every machine, breakdowns, work orders with their free text and times, spare parts fitted with their cost, and run times from the MES. Every repair then has a machine, a component, a duration and a price.
- 2 · PermitPermission Layer
Maintenance and production management see the analysis for their areas; hourly rates and part prices are visible only to those who see them today. The analysis counts machines, components and hours, not names. The maintenance manager owns the catalogue of damage types and every correction to the assignment, and that is recorded.
- 3 · ActIntelligence Layer
The agent reads every order text, abbreviations and typos included, and assigns it to a component and a fixed damage type; uncertain cases go to a person for confirmation. demi computes MTBF, MTTR, repair hours and cost per machine and component from recorded times and bookings, not by estimate. For the biggest items it sets a year of repairs against the cost of an improvement.
How it runs
- 1
demi reads the maintenance orders of recent years with their free text, times and parts fitted.
- 2
The agent assigns each order to a machine, a component and a damage type from your catalogue.
- 3
Maintenance confirms uncertain assignments once; after that they apply to similar texts.
- 4
A ranking per machine and component shows MTBF, MTTR, repair hours and cost.
- 5
The maintenance manager picks the weak points to improve rather than repair and approves the measure.
What changes
Clarity on which machines and components cost the most hours
Improvements where they pay back quickly
Figures from real work orders instead of gut feeling
What does it bring you?
Check it with your own numbers
Every starting value is an assumption you can change. The calculation sits right below, step by step.
Your numbers
Result
+288 hfreed per year
Assumptions and calculation
demi shows which components cause the most repairs; you decide what gets improved.
2 h → 1.8 h per case
How we calculate
120 repair jobs × 2 h × 10% × 1.0 (manual) × 12 months = 288 h
Value per year, before demi costs
Freed per year+288 h
Assumption: labour cost of €49.50 per hour in German manufacturing in 2025 (Destatis), plus workplace, IT and overheads.
= Value per year, before demi costs
+€18,720
gained per year
288 h ÷ 1,760 h per role = 0.2 full-time roles
An estimate, not a promise. The starting values are assumptions for an industrial company with 150 to 500 people.
Combine with more cases and get the packageQuestions about this use case
Our work orders are full of abbreviations and typos. Does this still work?
Yes, that is exactly what it is for. The agent reads abbreviations like “HR Z7”, typos and half sentences and assigns them to a damage type. Uncertain cases go to you instead of being guessed; a confirmed assignment then applies to similar texts.
Where do the run times for MTBF come from?
From the MES where there is one, otherwise from shift plans and downtime bookings. The agent shows which run time it used for each machine. If none exists, it uses calendar time and says so explicitly.
Is this used to rate our technicians?
No. The analysis counts machines, components and hours, not people. Who worked on an order stays in the order but feeds no figure. How the analysis is used in your plant is something you agree with your works council.
How does demi work out whether an improvement pays off?
From a year of recorded repair hours, part costs and downtime, at your hourly rates. You enter the cost of the improvement or attach a quote, and demi computes the payback. Whether to convert is the maintenance manager's decision.
