How Machine Learning Helps Find Sros
Confrontation of purchases and demand
Retailers form the necessary stock to meet the demand of customers by purchasing wholesale lots of goods from suppliers directly or through distribution centers. At the same time, the store constantly faces two risks: shortage or excess of goods.
With a shortage of goods, the store loses profit from sales. Customers do not find the product of interest on the shelf, their loyalty is reduced: they go after the "favorite toothpaste" to the competitor.
Excess of goods leads to problems in logistics associated with storage and distribution. If the store is working with perishable goods, it will face the costs of writing off and disposing of damaged products. The retailer has to discount the stale goods in order to free shelves for new lots. Working capital of the company, on which it was possible to purchase a new product, is frozen.
All this in total - a huge loss of profit.
It would be nice if you could accurately predict the turnover, but there is no ideal forecast in retail. A competitor launched aggressive advertising, massive sneakers went out of fashion, snow didn’t fall in winter - and the market situation changed dramatically.
But uncertainty exists not only on the demand side, but also on the supply side. The supplier may delay delivery, bring not all the ordered volume, or deliver goods of inadequate quality.
All this leads to the fact that finding a balance between the risks of shortage and excess of goods is an extremely difficult task.
Superiority of artificial intelligence
A software product is able to estimate the uncertainty of demand and thereby optimize purchases more efficiently than simply calculating on average if you use machine learning technology for this. In this case, the system will evaluate not average sales, but all possible potential scenarios.
We operate with the same statistics of the retailer: sales volume, sales history of this and substitute goods, planned prices, delivery schedules - and we calculate for each potential scenario the probability of its occurrence and marginality. In this case, the insurance and average reserves are formed based on the desired scenario. We go from the economy and evaluate different options for demand and different business strategies. Achieve fewer charges or more sales? Now, purchasing managers can easily track how beneficial a particular scenario will be for a particular supplier or group of products.
The superiority of the application of algorithms in comparison with the "old-fashioned" system of formulas is obvious. For example, in one of the pilot projects with a large food delivery platform, we focused on optimizing the purchase of perishable products (vegetables and fruits, dairy and meat gastronomy, meat and poultry) - they accounted for more than 70% of all write-offs and markdowns.
The use of machine learning methods reduced write-offs of goods by 20% and reduced cases of lack of goods in stock by 8% - and this is only the beginning. In foreign practice, there are already examples of the effectiveness of the method: for example, thanks to a similar solution, the digital equipment retailer reduced the turnover period of goods in their warehouses by 20% and reduced the cost of inventory storage by the same amount, and the food chain reduced out-of-stock by 75%. Not surprisingly, the optimization of procurement using machine learning methods is in demand.
The uncertainty of supply and demand is faced by electronics stores, fashion boutiques and other categories of retail chains. The system, whose work is based on machine learning technologies, is able to take into account the specifics of each business, offering the most effective and marginal scenarios.
Where to begin
We recommend starting a trading network that is ready to use new technologies in the procurement process by setting specific business goals and developing methods for measuring commercial indicators that need to be improved. Everyone understands that with a shortage and an oversupply of goods, costs increase - but few about how to measure, digitize, correlate them with the global goals of the network. If the retailer begins to collect and analyze this data, then over time he will learn to competently manage it and more carefully assess risks.
Therefore, before you start implementing solutions for the optimization of procurement, it is worthwhile to ask the objectives of this optimization, study business processes, determine the methodology for calculating risks. After that, as a rule, a site is selected for testing the solution — a limited list of outlets or a certain category of goods — and the solution is piloted on it. Based on the results of the pilot project, we look at how effective the deployment of the solution to the entire trading network will be.