Implementation of the system of analytics and prevention of lost sales GoodsForecast.OSA in the Perekrestok chain of stores

• Improve the current system for monitoring the availability of goods on the shelf

• Increase the level of availability of goods for buyers

• Increase turnover

• Optimize staff work

• Integration of large data arrays (checks) during the day in a mode close to online.

• Balancing the complexity of calculations and their speed: on the one hand, there should not be a large lag between the sales anomaly and the employee's reaction, on the other hand, the signals should be as accurate as possible.

• Maximization of the commercial effect in the conditions of limited resources of retail personnel.

At the first stage, the specialists set up data integration processes, including sales and balances on an hourly basis. Further, on the basis of the collected and labeled data, accessibility assessment models were built, the signal algorithm for alerting stores and ranking the issued signals was improved in order to maximize potential additional revenue. This system was integrated with the Perekrestok microservice, which provides signal exchange and feedback with retail outlets.

To evaluate the economic effect of the system, a specialized division of X5 Group Ad Hoc developed a design for A / B testing lasting 2 months.

A pilot group was formed, consisting of 132 stores with a wide geography. A control group was selected for the selected outlets to evaluate the results. The target metric of the experiment was the positive dynamics of the retail turnover of stores. X5 Group experts calculated that for statistical significance it should be at least 0.8%.

Within two months, the model generated up to two hundred signals in each store 3 times a day. Based on such signals, store employees received tasks on mobile devices to check the shelves for the presence of goods, the presence and correctness of the price tag, and the loss of consumer properties. If an employee identified a problem, it was corrected and confirmed using a mobile device that the issue was resolved.

According to the test results, according to the X5 Group Ad Hoc assessment, the model showed a statistically significant result. The point estimate of the RTO increase was 1.4%, and the upper limit of the confidence interval was 2.4%. In addition, a specialized division of X5 Group concluded that with a full-scale deployment, the same results of growth in turnover will be achieved, provided that the quality of processing incoming signals by store employees is maintained. Since October 1, 2022, the signal generation module has been launched throughout the Perekrestok supermarket chain, and the project has already received its development. At the moment, the X5 Group and GoodsForecast team are working on signal-based demand recovery with subsequent backward integration in order to increase the accuracy of the forecast in the Perekryostok network's auto-order system.