Today
Every CIP clean leaves long curves: temperature, conductivity, flow and time for pre-rinse, caustic, intermediate rinse, acid and final rinse. At the end the controller often reports only “cleaning finished”, and nobody compares every curve with the validated values. A cycle that was too short or too cold shows up only when a swab is positive or a product spoils.
An example
At Tarvina, CIP station 2 cleans the mixing tank and pipes of the sauce line every night. For Tuesday night's cycle the controller reports “cleaning finished”. The agent looks closer: in the caustic phase the return temperature stayed above 75 °C for only 12 of the validated 20 minutes, while conductivity was fine. It flags the cycle before production starts at 5 am and shows that the return temperature has been falling slowly for three weeks because scale is building up in the heat exchanger. The shift leader has the equipment cleaned again, and maintenance descales the heat exchanger at the weekend.
Your data, wherever it lives
Your cleaning data lives today in the CIP controller and its exports, production orders in the MES, the validated values in the validation report as a PDF or in the QM system, cleaning records and detergent use in spreadsheets. demi reads, understands and links them per cycle, object and line, with your rights. The AI checks every cycle as soon as it ends instead of spot checks at the audit, and quality or the shift leader decides on release or re-cleaning.
Where the data lives today
- Machines & sensors
- MES
- QM system
- PDFs
- 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: trends of temperature, conductivity, flow and valve positions from the CIP controller, cleaning objects and production orders from the MES, validated parameters from the cleaning validation and the QM system, plus cleaning records and detergent use from spreadsheets. demi replaces nothing and links every cycle with its object, its program and the next order on that equipment.
- 2 · PermitPermission Layer
Production and quality see the cycles of their equipment. Only the authorized person in quality or the shift leader may release an object after a deviation, and cleaning programs are changed only by whoever is responsible in change control. Every decision is recorded.
- 3 · ActIntelligence Layer
The agent splits each cycle into its phases and checks per phase the minimum return temperature, concentration via conductivity, flow and hold time against the validated values. It recognizes skipped steps, routes without flow and trends across many cycles, such as a heat exchanger that is slowly losing performance. Limits come from the validation; the language model only explains the finding.
How it runs
- 1
After every cycle, demi reads the curves from the CIP controller and assigns them to object and program.
- 2
The agent checks each phase against the validated values for temperature, concentration, flow and time.
- 3
It flags a suspect cycle before the next production start on that equipment.
- 4
Across many cycles it shows trends, such as falling return temperatures or rising detergent use.
- 5
Quality or the shift leader decides on release or re-cleaning, and the evidence is recorded.
What changes
Every cleaning cycle checked, not just the conspicuous one
Suspect cycles caught before start-up
Audit evidence without searching through curve printouts
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
+240 hfreed per year
Assumptions and calculation
demi compares every cleaning curve with the validated values; you decide on release.
10 min → 6 min per case
How we calculate
300 cleaning cycles × 10 min × 40% × 1.0 (manual) × 12 months = 14,400 min = 240 h
Value per year, before demi costs
Freed per year+240 h
Assumption: labor cost of $48.62 per hour in US manufacturing in June 2026 (BLS), plus workplace, IT and overhead.
= Value per year, before demi costs
+$15,600
gained per year
240 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
Isn't the CIP controller's message enough?
The controller often only checks that the program ran through and measures temperature at the supply instead of the return. The agent compares each phase with the values from your cleaning validation and so also catches a cycle that formally finished but was too short or too cold.
Where does demi get the validated values from?
From your cleaning validation and program specifications, per object and cleaning program, whether in the QM system, as a PDF or as a spreadsheet. Quality confirms the limits once; after that the agent checks every cycle against them.
What happens to a suspect cycle?
The agent flags it and shows the phase, the measured value and the specification side by side before the equipment produces again. Whether it is re-cleaned, checked by swab or ATP test, or released, the authorized person decides, and that is recorded.
Does this work with older plants without a data historian?
Yes, if the controller writes the values to a file or has a chart recorder. demi reads CSV exports, printouts and scanned charts and turns them into phases and values. If a chart is incomplete, the agent says so instead of assessing the cycle.
