A system using advanced Machine Learning models for automatic replenishment for the retail industry

Challenges

One of the largest franchise chains on the Polish market struggled with a time-consuming and misguided merchandise ordering system. Each franchisee was forced to plan its merchandise orders individually – at the expense of the time needed to take care of customers' needs. Performing advanced and accurate sales analysis – taking into account which products sell best and what factors influence such a result – was not possible without the support of advanced solutions. As a result, both the franchisees individually and the client's logistics centers struggled with a shortage of the most popular products and an excess of other products – which had particularly bad financial effects in the case of goods with a short shelf life.

The company was looking for an advanced technological solution to optimise the process at all points. The solution had to be not only effective, but also transparent and convenient to use, in order to shorten the ordering process as much as possible and not distract managers, salespeople, and franchisees from completing daily customer service tasks.

Customer problems we helped solve:

• Long ordering time for franchisees.
• Quantities of goods inadequate for the pace of sales.
• Lack of a medium-term forecast for logistics.
• Possibility of supply chain bottlenecks due to over-ordering.
• Lack of a quick and effective tool to optimise commissions.

Solution

Through business analysis and a full understanding of our client's problems and needs, we created an advanced system based on machine learning to support the ordering process. The implemented tool was based on two solutions:

• Demand forecasting model – the tool indicates what products will sell in a particular store at the index level. Each day, a forecast is carried out in each of several thousand stores indicating which products will sell best the next day
• Order recommendation model – the optimisation system uses data received from the module forecasting stays – including, among other things, information about the amount of inventory the store has at any given time, limits, promotions, seasonality of products, shopping trends, customer behavior, store location, weather conditions, historical data, or weather.

Each day, the system makes detailed and accurate recommendations for several thousand stores, taking into account many internal and external factors, including information on location – a store located near a school or sports facility will sell more products of a given type than a similar store located in a small residential area on the outskirts of a town.

The final results of the analysis are presented in a transparent application – making ordering as easy as possible.

The solution, thanks to advanced and trained machine learning models, is fully scalable and grows with the needs and development of the customer, instantly adapting to their growing requirements.

Key objectives of the project: 

• Reducing the time spent by franchisees on order processing.
• Preparation of sales forecasts for stores.
• Preparation of an optimiser that will generate an order proposal for a given day for a given store in an automated manner.

Effects

The client, thanks to the cooperation with Britenet specialists, received the ability to create demand forecast models based on sales with product rotation, which significantly accelerated, facilitated, and made more efficient the time-consuming ordering process, as well as optimised the execution of sales and logistics requirements.

Key benefits of the solution:

• Automation to ensure availability of goods on shelves
• Significantly reduced ordering time for franchisees 
• Minimisation of losses due to the expiration date of goods
• Quantity of goods tailored to the demand of specific stores
• Ability to better manage logistics centers and their inventories
• Scaled solution based on modern technologies