AI helped Verny supermarket chain to increase the forecasting accuracy of effect of actions by 1.5 times

On January 21, 2020 the GoodsForecast company announced that the Verny supermarket chain increased the forecasting accuracy of effect of promotion actions by 1.5 times, having implemented the GoodsForecast.Promo system. A system using machine learning technologies independently analyzes keyword parameters of the previous promotion actions and predicts results of future. Initial data for forecasting are loaded into it automatically.

To GoodsForecast.Promo system implementation specialists of Verny network predicted effects of promotion actions at the level of branches of network. The tool allows to build forecasts already at the level of shops that allows to avoid redundancy or, on the contrary, insufficiency of promotional goods in specific outlets.

Keyword parameters which are considered for effective forecasting of results promo are: goods and its category, mechanics and promotional period, discount depth. In addition to them, also other properties of promoaktivnost, for example seasonal coefficients for separate goods and recovery of demand in scarce actions analogs are considered. For formation of the most exact forecast a system considers data on similar and similar actions which took place in network within the last two years, — Kirill Chernikov, the manager of projects of GoodsForecast company speaks

Implementation of GoodsForecast.Promo in Verny network began in October, 2018 and took about 9 months. As a result of network it was succeeded to reduce considerably a commodity remaining balance after carrying out promo in each of the shops, without creating at the same time deficit of goods. The indicator of an absolute forecast error of effect of promotion actions decreased by 1.5 times. At the same time operating time of the personnel of the company involved in forecasting of a commodity remaining balance was reduced that allowed to pass to the detailed analysis of specific forecasts to which it is required to pay attention. A basis of a system are the database under control of the Microsoft SQL Server and also the self-training forecasting models integrated into a DB implemented in language R. Business users of a system work in the convenient web interface stylized in corporate style of retail chain stores. Now the GoodsForecast company jointly with specialists of Verny network develops functionality of a system and the interface.

We improve a system first of all in terms of ergonomics and transparency for users. But also we try to expand its functionality using accounting of additional parameters when forecasting, such as price segments of similar goods, existence of occasional seats of the calculation in shops, overflowing of the goods demand in a specific action and adjacent promoaktivnost, — Kirill Chernikov, the manager of projects of GoodsForecast company speaks