Today
Every system shows only its own slice: the ERP the batches, the MES the process data, the QM system the defects. The cause often sits right in between, for example when a material causes trouble on one line only in damp weather. Seeing it would take days of exporting and stitching data together, and day-to-day work leaves no time for that.
An example
At Arnstein, more AlSi10Mg housings AH-2140 fail the leak test in autumn. demi overlays test results, casting parameters, melt batches, the shift plan and weather data: 41 of 52 failures come from days with high humidity on which the melt was degassed for a shorter time. Quality management has the degassing time adjusted, and demi now flags humid days to the foundry in advance.
Your data, wherever it lives
Today the pieces sit in separate systems: batches and suppliers in the ERP, process data in the MES and on the machines, test results and complaints in the QM system, shift plans in spreadsheets, weather data on the web. demi reads, understands and links them by order, batch and time, with your rights. The AI works through in minutes what would otherwise take days of exports, and your team decides which lead to follow.
Where the data lives today
- ERP
- MES
- Machines & sensors
- QM system
- Office files
- Web & portals
How demi solves it
Three layers in this order: connect the data first, then the rights, then the AI.
- 1 · ConnectData Layer
The case needs data from the ERP, MES, machines, QM system and spreadsheets, plus public data such as the weather. demi replaces nothing and joins the records by order, batch, serial number, machine and timestamp into one picture per part or lot.
- 2 · PermitPermission Layer
Quality and production management see the analysis. It runs by line, machine, shift, batch and supplier, never by name. Anything that goes to a supplier is released by quality management, and that is recorded.
- 3 · ActIntelligence Layer
The agent works with statistics, not guesses: it tests which factors are linked to scrap, complaints or downtime, how strongly and how stably. Every finding shows the records behind it. The language model reads free text such as complaints and shift logs and explains the results in plain sentences.
How it runs
- 1
You pick the question, such as scrap on a line, complaints about a part or downtime of a machine.
- 2
demi joins the data of every connected system by order, batch, serial number and time.
- 3
The agent tests statistically which factors play a part, across weeks and seasons too.
- 4
Every finding comes with the strength of the link, the period and the records behind it.
- 5
Your team assesses it and plans a trial or an action; demi tracks the effect and warns when the pattern returns.
What changes
Causes between systems become visible
Decisions on evidence, not gut feeling
Early warning when a known pattern returns
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
+230 hfreed per year
Assumptions and calculation
demi links the data of every system and computes the correlations; you assess and decide.
8 h → 4.8 h per case
How we calculate
6 cross-system analyses × 8 h × 40% × 1.0 (manual) × 12 months = 230 h
Value per year, before demi costs
Freed per year+230 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
+€14,950
gained per year
230 h ÷ 1,760 h per role = 0.1 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
How does demi know which data belongs together?
Through shared keys such as order, batch, serial number, machine and timestamp. Where a key is missing, demi matches by time window and marks the link as uncertain. Every link can be traced back to the single record.
Is a statistical link already the cause?
No. demi shows how strong and how stable a link is and how many cases support it. Your team establishes whether it is the cause, often with a targeted trial. demi suggests which data will best show the result of that trial.
Which outside data goes in?
Weather, public holidays or raw material prices, for example, when they matter for the question. External data is labelled as such. You decide what goes in.
Does the search run only once?
No. A confirmed pattern becomes a rule that demi checks continuously. If it shows up again, say humid days together with short degassing, demi alerts the shift in charge so it can act before scrap is made.
