Departments · Production AI in Production: Orders, Inspection Records and Shift Reports

Production tasks

The following six examples describe typical processes, not client projects. AI reads, organises and drafts material using approved data. It does not decide whether to approve a product or intervene with a machine. Which steps can be automated depends on your documents, interfaces and inspection requirements.

1. Prepare production orders from customer orders

Items, quantities, dates and variants need to be transferred from customer orders to production orders. Questions arise when details are missing or do not match the master data.

• Input: approved customer orders from sales, item master data, bills of materials and production planning requirements.
• AI prepares: reads line items and notes, checks descriptions against master data and flags missing or conflicting information. It creates a draft production order.
• Production planning checks and approves: items, quantities, dates, material availability and discrepancies. Only then is the order created or changed in the system of record.
• Output: a checked production order with supporting documents and a record of outstanding questions.

The information feeding into this process may come from supplier confirmations in purchasing or from customer orders. What matters for production is that the source and approval status of the information remain clear. Missing data must not be filled in without drawing attention to it.

2. Create shift reports and handovers

A shift generates records of output, downtime, scrap and unfinished work. A handover is useful only if the next shift can understand and verify the information.

• Input: shift log entries, production data, fault reports and notes from the people responsible.
• AI prepares: summarises events by machine and order, assigns times and flags gaps or conflicting information in the sources.
• Shift supervisor checks: quantities, descriptions of causes and outstanding items. They add observations that are not in the systems.
• Output: an approved shift report with references to the original entries and a list for the next shift.

A report should not consist solely of readings. The person responsible must confirm from what actually happened whether downtime was caused by a material issue, a changeover or a fault. This prevents an unsupported cause from becoming part of the summary.

3. Prepare inspection records and 8D reports

Inspection results are often stored separately from order data, images and written findings. When a complaint arises, these details need to be brought together for review.

• Input: inspection plans, measurements, affected orders and batches, and documented complaints.
• AI prepares: transfers existing values into a draft inspection record, cites the sources and separates known facts, open questions and proposed actions in a draft 8D report.
• Quality assurance checks and approves: measurements, batch assignments, descriptions of defects, causes and actions. Unsupported statements are removed or marked for further investigation.
• Output: an approved inspection record or an editable 8D draft with traceable supporting evidence.

An 8D report documents a structured approach to resolving a problem. AI can organise complaint data and prepare a report. It cannot establish a cause merely by describing one, or confirm whether an action has been effective. Those checks belong to the qualified people responsible.

4. Analyse complaints and quality data

Individual deviations are usually recorded. It is harder to identify recurring patterns across products, machines or batches without conflating different types of defect.

• Input: complaint texts, defect codes, inspection records, batch details and documented actions.
• AI prepares: groups similar descriptions, shows related cases and flags missing links between records. It drafts questions for specialist investigation.
• Quality assurance checks: whether the cases are genuinely comparable and which findings warrant further investigation.
• Output: a traceable overview of cases, data gaps and outstanding checks.

The overview is not an automatic fault diagnosis. The same term in two complaints may describe different issues. The original records must remain accessible when decisions are made. If information first needs to be extracted from files, document processing can form an earlier step.

5. Schedule maintenance requests

Faults are recorded in shift logs, machine alerts and maintenance systems. Requests sometimes lack details about the machine or its history that are needed for follow-up.

• Input: fault reports, machine identifiers, maintenance plans and available production data.
• AI prepares: brings related reports together, adds references to known incidents and proposes how to classify them.
• Maintenance team checks: urgency, possible connections and planned work. The team decides on any intervention with the machine.
• Output: a maintenance request with the missing details added, or an approved planning task in the maintenance system.

This is not yet predictive maintenance. Reliable indications of a possible failure require suitable measurements, known operating conditions and specialist assessment. If these foundations are missing, recording and classifying fault reports properly is a more useful first step.

6. Check work instructions against bills of materials

Changes to components or processes often affect several documents. Older versions must not be used as current instructions without anyone noticing.

• Input: approved drawings, bills of materials, change notices and existing work instructions.
• AI prepares: compares identified versions, identifies affected passages and drafts an updated instruction.
• Production engineering checks and approves: technical statements, sequence, validity and affected workstations under the established change procedure.
• Output: a checked version with a clear version reference and documented changes.

Valid documents must be clearly identified. If several versions are available without a status, AI should flag the conflict. It must not decide which drawing or bill of materials is authoritative.

Systems and machine data

Most production facilities already use several systems. A new AI‑supported process should connect to them. Clear rules are needed about which system is the source of record for orders, measurements and approvals. Not every process needs a connection to every machine.

MES stands for Manufacturing Execution System. It supports processes between planning and production. BDE is the German abbreviation for operational data collection. It refers to recording production data, including quantities, times and faults. The terms describe different tasks. You need to check which functions your existing systems cover. We do not sell an MES or require you to have one.

For integration, we first establish which interfaces are available and what data they actually provide. We can then determine whether information should only be read or also written back after approval. An Excel spreadsheet can be a useful input. It should not inadvertently become a second system of record. If a write operation fails, the failure must be visible and available for review.

Approvals and quality

Prepared information is not the same as specialist approval. Inspection records, product holds, deviations and changes to work instructions follow your business procedures. Quality assurance makes quality approval decisions. AI can flag missing information and gather supporting evidence, but it cannot replace an inspection of the product.

Traceability depends on properly linking the order, item, batch or serial number, and inspection documents. Automatically generated text without those links is of little use when an issue needs investigating later. Corrections should also remain visible: who entered a value, who checked it and which version was approved? Document control follows the quality management standards and internal requirements applicable to your business.

If data on individual employees’ performance is collected or analysed, any employee representation rights should be assessed early. In Germany, section 87 of the Works Constitution Act (BetrVG) may apply; in Austria, section 96a of the Labour Constitution Act (ArbVG) may apply. Whether and how these provisions apply to a particular business requires specialist and legal assessment. We do not plan automated evaluations of employees. This is not legal advice.

Data: orders, machines and quality

A workable process starts with an assessment of the available data. For orders, this includes item numbers, variants, bills of materials and approval status. For machines, unique identifiers, timestamps and the meaning of alerts matter. Quality data also needs the inspection characteristic, unit, batch and version of the inspection plan. If these details do not match, the discrepancy must be visible rather than disappearing into a report.

Machine data does not have to be transferred wholesale to an external service. We work with your IT team to establish what processing suits the intended workflow and which data should remain on site. Selected events may be enough for a shift report. Raw machine data is useful only if its meaning is clear. Access to confidential order and quality data is also limited to the relevant purpose.

Custom apps for the shop floor

Not every entry is easy to make at an existing workstation. A supplementary app can capture details where they arise: at a machine, during an inspection or at shift handover. A checklist app can require specific fields and a link to an order or batch. A digital shift log can pass on outstanding items. A fault reporting app can bring together the machine, time and observation. Tablet-based inspection checklists can assign entries directly to the correct inspection plan.

Such apps supplement ERP, MES or quality management software. They should not create separate item master data or approvals detached from the systems of record. Clear input fields, appropriate access rights and a defined approach to a lost connection matter. An entry must show whether it has only been captured locally or has reached the system of record.

For app development, we therefore start with the actual workflow. Who enters the information? Who checks it? What happens if an entry is wrong? Only then can we determine whether a custom app is needed or an existing form will do. How the app will be used on the shop floor is part of planning from the outset.

Where to start

Choose a paper-based process or a recurring transfer between two systems. Which information is copied more than once? Where do questions arise? Who must make the final decision? A suitable first process has identifiable inputs, a person responsible for checking them and a clear output in an existing system.

Then check a specific case to see whether the supporting documents and master data are sufficient. For a production order, that might be a customer order specifying the variant. For an inspection record, it means measurements linked to a valid inspection plan and batch. If those links are missing, the work starts with organising the data, not drafting with AI.

We discuss the process and the systems involved with you. This helps establish which steps automation can prepare, where approval is needed and whether a supplementary app would help. You do not need to replace your production software or change an entire department at once.