The logic of stocks is complex. Starting from the management of (non-) perishable items with techniques like LIFO (Last-In-First-Out) or FIFO(First-In-First-Out), continuing with logistic costs, going ahead with lost, damaged, stolen or expired items, etc.

The warehouse people are the last line of quality assurance for sold items and purchased ones. On the other side, they are responsible for counting them and giving a consequenced value of the warehouse.

ML in choosing where and how to stock

The ERP tracks every items’ movement from being purchased or produced to its shipping. This information calculates the actual maintenance costs like the time to be awaited for having the good available, the time consumed by the warehouse people for moving it, the shipping costs, the risk of damage, expiration, etc.

ML provides the best solution by cost, time or quality optimisation. It gives the option to change the stocking phases if the calculation using the metrics mentioned above would be worth the change. For example, when the quality assurance of the incoming goods coming from a supplier is by historical track reasonable, then the dropshipping would become an option.

ABC categorisation with ML

The ABC methodology divides the items into three categories by importance: A – Top, B – Moderate, and C – Relative. Every good in the warehouse gets the association with one of these categories based on the Pareto principle of 80/20. The reference values involved in the calculation are, for example, its cost, the number of transformations during the production process, the frequency and the level of quality assurance, etc.

ML helps distribute them into the most appropriate category providing the impact of every reference value for calculating the actual result. In some cases, the category’s change can happen only for a specific period caused by events occurring during that period.

Counting items optimisation

A physical inventory is always an investment in time and resources. It is necessary for giving the value of the warehouse (or part of it) and resyncs the expected quantities with the confirmed availability.

ML helps here to provide the sequence of items. It optimises by position and importance. Also, it considers the time used by the operator to count the goods, considering the time slot of the shift, the foreseen logistic operations that should occur, and the historically encountered difficulties.

Conclusions

The ML helped here, presenting options for choosing with enhanced information. The algorithm improves on each selection, understanding the operator’s preferred line for each case.