AI in manufacturing — plant knowledge under control

Retaining expert knowledge before experts leave

Adoption and skills

Problem: Key knowledge about machines and processes is in the heads of two people about to retire

How it works: The knowledge audit (CDF-F0.K); the Knowledge Gap Register and a digital twin with a Learning Log that captures the expert’s way of working

What you gain: The departing expert’s way of working stays in the company in a form the successor can use — before it is too late

Assistant for standards and process sheets

Knowledge and documents

Problem: Finding the right standard or process sheet takes too long

How it works: A central RAG knowledge base with OCR and source preview

What you gain: The process engineer finds the right sheet or standard with a question and a citation — without searching binders and drives

Automatic production reporting

Analytics and reporting

Problem: Production reports are prepared by hand from several systems

How it works: Scheduled script runs (RRULE) and in-chat code execution for calculations on production data

What you gain: The shift and daily report is produced automatically from the same calculations — the manager reads the result instead of compiling it

AI readiness assessment of legacy systems

Integration and reliability

Problem: Old MES and ERP systems — nobody knows which are ready for AI

How it works: CDF-F0.L: assessment of older systems for AI — a list of modernisation priorities

What you gain: Modernisation goes where AI adds value; systems not worth touching stay on the “postpone” list with a rationale

A quick decision: scale the pilot or stop

Adoption and skills

Problem: AI pilots drag on for months without a deployment decision

How it works: A Cognitive Sprint (CDF-F4) with a Cognitive SLA and a “scale or stop” gate

What you gain: The pilot ends in 2–4 weeks with a decision based on measurable criteria — there is no “let’s extend it a bit more” option

Operational automation without a developer

Process automation

Problem: Small automations require a developer

How it works: If/then rules in natural language; n8n flows; a Decision Learning Layer that proposes automating repetitive decisions once approved

What you gain: The shift manager defines a rule in plain language, and the system proposes further automations based on repetitive decisions

Local AI for production data and recipes

Security and sovereignty

Problem: Production data and recipes must not reach the cloud

How it works: Data classification and routing to local models; blocking the pasting of confidential data

What you gain: Recipes and process parameters stay in the plant — technically, not just on paper

Where to start: what problem are we solving and how much is the loss costing us today?

With one process with a measurable loss, e.g. time spent searching for instructions or manual reporting. The Initial Assessment in the CDF methodology selects the use case and sets the criteria for the decision after the pilot

What data is needed?

Documents are enough to start: instructions, standards, service reports. TWIN:DESK also processes scans

What happens after an AI alert?

A human makes the decision. AI points to the source and proposes steps, and every action is recorded