← All projects

AI inventory monitor

I built a monitor that assesses product data locally and sends only new or significantly changed cases to AI.

Implemented and operational Brand name withheld on request
Illustration of the AI inventory monitor
Illustration of the scope of the solution.

Selecting products that require attention

I developed a monitor that analyses stock and selects cases for closer review. Its rules account for stockout risk, urgent reorders, changing sales trends and excess stock. Each case is prioritised according to the product’s situation.

First the calculations, then the interpretation

Local rules calculate a suggested order quantity, deadline and risk assessment. Selected data is then passed to an AI model to help interpret and explain the recommendation. Calculations are kept separate from the written explanation.

AI triggered on significant change

I added a comparison between each new case and the previous recommendation. Changes in priority, event type, trend direction, delivery quantity or reorder deadline can trigger another analysis. Thresholds limit repeated model calls when the situation has not changed significantly.

Urgent products at the front of the queue

Selected cases are sorted by priority and recommended order date. A limit on the number of cases analysed directs model resources towards products needing more urgent decisions.

Managing analysis costs and repeated runs

I added a model usage log and cost cap. Each recommendation has an identifier based on key event details, helping detect duplicates. Users receive structured information to review, while AI use remains controlled.

Reports to support purchasing decisions

The module helps assess stock and flags products for review. Its AI component requires access to the model. The user decides what to order, comparing recommendations with delivery plans and current business conditions.

Scope of implementation

  • AI agents
  • Custom systems
  • Selection of exceptions requiring analysis.
  • Cost limit and model usage records.
  • Reports to support inventory decisions.

The AI portion requires active access to the model. The module does not place orders for the user.

The brand name is withheld on request. This description explains the work without disclosing customer data or internal materials.

Do you need a similar solution?

Describe your process and what you want to improve.

Let's talk