What Is Replenishment?

What Is Replenishment?

It is a system that automatically complements the shortcomings in the goods in the shop. It is easy to define this mechanism.

It's a vending machine whose job is to ensure the proper supply of stores.

This application analyzes the stocks in the store and sends orders to suppliers or manufacturers in the absence of goods.

Slots of ordering appeared in Poland with the development of mass sales in supermarket chains. It was impossible to manually handle such a large volume of turnover.

Replenishment replaced numerous industry and supply companies, became an essential element of the logistics system of large retail networks.

It is worth noting that the use of the term "ordering machine" is more appropriate than the English word "replenishment," which simply means complement.

This name is firmly rooted in the supermarket industry. Replenishment is a system that constantly analyzes warehouse states in stores.

The system collects information from hundreds of stores about thousands of products. Individual products in specific stores have so-called upper and lower store limits.

These are the ranges of the quantities of individual goods on shelves. The lower limit determines the smallest quantity of goods on the shelf, the upper limit is the quantity of goods that will fit maximum on the shelf.

If the store ordered more goods than it had, it would have to be stored in the back or somewhere in the free space of the store.

The simplest version of the ordering machine is based on static stock limits for individual stores.

If the condition of a specific item in a particular store approaches the lower limit of the warehouse status, the system automatically creates an order.

The volume of the order must not exceed the maximum stock limit.

W-MOSZCZYNSKI-2023-5-27

Problems

The simple version of replenishment has some drawbacks that, on the vast scale of supermarkets and retail chains, cause huge losses,

resulting mainly from lost revenue due to the lack of goods on shelves. The machine sends an order when the warehouse is approaching the lower limit.

The delivery time is a few days. During this time the goods are bought out and the empty shelf remains until the ordered delivery arrives.

This means no sales and costs resulting from it. The solution. is the setting

lower control limits at a higher level. In this case, the machine sends an order at a time when the stock is still high.

The problem is, you don't know how much to order.

If the goods are ordered at the level of the current absence and the duration of the order is, for example, a week, the goods will be bought out until delivery.

At the time of delivery, the machine will again have to make an order because the current stock will still be low.

The supply of the store by small suboptimal volumes of supply is another example of unnecessary costs caused by poor quality of the process.

What if the goods in question are not sold and the vending machine has ordered more goods because the limits are set?

There will be a exceedance of the upper limit of the warehouse and the store will stay with excess goods.

The problems described may concern at the same time a thousand or more goods in one store. The uprising of chaos, which will be difficult to control.

In addition, there are also problems related to the expiry date for the intake of FMCG products. There is an aspect of different delivery times from different suppliers.

How should the automaton behave when the stores will organize promotional campaigns, then for goods covered by the promotion statically set storage limits will be too small.

On the other hand, goods not covered by the promotion, which will be replaced by goods promoted, will have too high limits.

It is also worth mentioning that goods in retail networks can also be shipped between shops.

It is easy to guess that a simple ordering machine is unable to meet all these challenges.

A simple replenishment based on control limits is unable to predict seasonal changes. For example, ice creams are sold mainly during the summer period.

So the limit for each type of ice cream should be set individually and then according to the season or weather forecast should be changed.

At what level should Ice cream limits be set? Sales of ice cream in large part depends on the weather. A worker setting limits should suggest long-term weather forecasts.

It turns out that a shopkeeper needs to update the limits every few days to order the machine properly. It's easier to order ice cream in person. It's obvious absurd.

The labour intensity and low efficiency of continuous adjustment of control limits for hundreds and thousands of FMCG goods was the main reason for abandoning the control limit approach.

The control limits have therefore been replaced by simple models of time series autogressive moving average model of ARMA or ARIMA type. These are simple models based on trends.

You can say in a very simple way that these models are sort of walking averages, and that this solution is mostly based on simplicity.

These models provide for sales based on sales behaviour in recent years. The algorithm bases each of the thousands of goods in stores to the auto-aggressive model.

The model calculates the future stock.

For example, on Monday the machine orders because it predicts that on Thursday the warehouse status of the specified item will be closer to the lower warehouse border.

The solution based on medium-sized autoregression models is primarily autonomous, does not require editing control limits for storage states.

The problem with models is that they predict on the basis of the last period. If last time it was hot and ice cream sold in large quantities,

this ARIMA model will assume that in the future, according to the trend, ice cream will also sell in large quantities.

So the model will order again a lot of ice cream that will not be sold because it's coming autumn.

Sales of ice cream will fall dramatically, and the reply based on auto-aggressive models of walking averages will conclude that in the next period no more ice cream will be ordered.

As a result, stores will be left with ice creams that have been ordered unnecessarily. This phenomenon will appear simultaneously on a scale of several thousand goods in several hundred shops.

This means significant costs for retail networks, which is why the models of self-aggression of medium moving more advanced machine learning have been replaced over time.

And we're talking about models like these that contain elements of artificial intelligence, and these models are based on a much longer history of sales, even a few years.

This makes it possible to take account of seasonality.

These models are based not only on sales themselves, but also on variables describing the phenomenon of sales, such as weather or days of the week.

The contracting machine based on an advanced classification model will predict most future changes in demand and thus will make an appropriate order that leads to a reduction in

losses caused by too low or excessive store handling.

The operation of such a system is complicated because it is based on thousands of models which provide for the behaviour of customers buying specific goods based on variables describing.

To build a forecast model, you have to choose the right algorithm and then conduct training of this model. The model learns on time series of sales for many years.

It is well trained to predict the level of sales with accuracy between 75 and 95%. Modeling takes a long time and requires specialist knowledge.

Modern replenishment systems are placed in the cloud. Models operating there are retrained automatically so that they do not lose their forecasting ability.

Moreover, such a solution allows to create further models for new, emerging goods.

Replenishment is a very important element of the logistics infrastructure of the retail industry.

Knowledge of this mechanism can be useful for suppliers of goods, bakers, bakers.

Some bakery chains also sell a wide range of food products. Many bakeries and bakeries supply large-area shops.

I hope that this publication will contribute to an increase in knowledge about the operation of sales network logistics, especially sales of FMCG products.

Wojciech Moszczyński — graduate of the Department of Econometrics and Statistics of Nicolaus Copernicus University in Toruń; specialist in econometrics, finance, data science, and management accounting. He specializes in the optimization of production and logistics processes. He conducts research in the area of the development and application of artificial intelligence. For years he has been engaged in the popularization of machine learning and data science in business environments.