Anticipate breakdowns and stock-outs, starting from shared definitions.

Three concrete uses

1. 01Predicting equipment breakdowns from your machine data

   AI Act level · Minimal risk

   1. 01Collect measurementsSensors, controllers, manufacturer data
   ↓→
   2. 02Detect driftsDeviation from usual behaviour
   ↓→
   3. 03Propose an interventionWith confidence level and history
   ↓→
   4. CheckSign-off by the maintenance manager
   ↓→
   5. 04Schedule maintenanceSlot chosen by the workshop

   The problem

   Unplanned stoppages disrupt production, yet machines already produce measurements: vibration, temperature, counters. That data sits in controllers or with the manufacturer.

   What the AI does

   The model spots the drifts that usually precede a failure and proposes an intervention with its confidence level. It does not stop the line or change the maintenance schedule on its own.

   Data needed

   Sensor history, past interventions coded by failure type and the maintenance plan.

   Who signs off

   The maintenance manager decides on the intervention and its slot, after consulting the shop-floor supervisor.

   Minimal risk (EU · AI Act); the use would become high-risk if it served as a safety component in managing critical infrastructure such as water, gas or electricity supply (Annex III).

   • Evaluation
   • Human oversight
   • Logging

2. 02Forecasting stock-outs and proposing replenishment orders

   AI Act level · Minimal risk

   1. 01Forecast demandBy item, site and week
   ↓→
   2. 02Calculate the riskStock-out or overstock, actual lead times
   ↓→
   3. 03Propose the orderQuantity, supplier, date
   ↓→
   4. CheckSign-off by the planner
   ↓→
   5. 04Place the orderIn the ERP, by the planner

   The problem

   Forecasts rely on spreadsheets and the experience of a few people. Stock-outs and overstock are discovered too late, often at the same time on different items.

   What the AI does

   The model forecasts demand by item and site, calculates stock-out risk and proposes order quantities. It places no orders: the planner approves, adjusts or rejects.

   Data needed

   Sales and issue history, stock by site, actual supplier lead times, promotions and calendar.

   Who signs off

   The planner approves each proposal; above a set amount, the supply chain manager does.

   Minimal risk (EU · AI Act): the tool deals with items and flows, not people.

   • Semantic layer
   • Evaluation
   • Human oversight

3. 03Tracing a quality drift to batches and suppliers

   AI Act level · Minimal risk

   1. 01Record the defectNature, severity, where detected
   ↓→
   2. 02Trace the batchesMaterials, suppliers, customers delivered
   ↓→
   3. 03Propose causesRanked, with supporting evidence
   ↓→
   4. CheckSign-off by the quality manager
   ↓→
   5. 04Hold or recallQuality manager’s decision

   The problem

   When a defect appears, you need to find the affected batches, materials, suppliers and customers quickly. The information sits in the ERP, the quality system and paper forms.

   What the AI does

   AI links non-conformities to batches, materials and suppliers, proposes likely causes and lists the batches to hold. It triggers no hold or recall: the quality manager decides.

   Data needed

   Batch-level traceability, coded non-conformities, certificates and supplier inspection results.

   Who signs off

   The quality manager decides on holds, any recall and corrective action, with senior management in the event of a recall.

   Minimal risk (EU · AI Act); a recall decision remains governed by your product safety obligations, which cannot be delegated to a tool.

   • Semantic layer
   • RAG
   • Logging