GOALS AND OBJECTIVES
To increase customer loyalty and reduce marketing costs through targeted work with customers.
To develop and implement a purchase forecasting system.
- A system based on Machine Learning algorithms
- Oracle Siebel CRM and Oracle BI
Jet Infosystems specialists have built a mathematical model trained on extensive data from the company’s CRM system, including: the register of commodity items, transactions for the previous year, product turnover and delivery, receipt information and demographic data from discount card holders. The solution uses a whole range of machine learning methods, including gradient boosting, random forest, collaborative filtering, etc.
The newly implemented ML-model follows a two-stage scenario. First it analyzes purchase patterns and identifies a target group of customers (from among 2.6. million loyalty card holders) who are most likely to purchase specific goods during the next two weeks.
The model’s second stage then forecasts the two products which a given customer is most likely to wish to purchase. The forecast works up to exact SKU accuracy, based on all the company’s available items (several tens of thousands). A mailing list with personal discount offers is made to a specifically generated list of customers; a discount amount within allowed values is generated, and the system calculates personal offers for each customer, individually.
Machine learning methods thus help the company determine a ‘golden segment’ of loyalty card holders; both the list itself and all parameters are continuously adjusted in real time.
With this ML-model data and the targeted work with customers it makes possible, the retailer has significantly reduced the cost of its marketing campaigns, while also increasing sales.
About 47% of loyalty card holders now return to purchase a second time (the previous the figure was only 22%). The average customer purchase amount among those to whom offers with personal discounts were made was 42% higher, on average, than the average purchase amount of other customers. During systems testing, personal offer customers accounted for approximately 7% of company income, though they make up only 1% of the company’s total customer base.
Correlations identified during the project entail enhanced facility for making personal client offers and also make it possible to further increase the accuracy of the mathematical model, making it possible, for example, to expand the sample by adding previously unused indicators to the data array (information about warehouses, product ratings, etc.). In order to continue optimizing marketing campaign costs, an analysis of the effectiveness of all customer interaction channels is also required.
To date, Jet Infosystems specialists have implemented more than 50 projects using Machine Learning technologies for a diverse client base including banks, retailers, industry, insurance and other sectors. The range of tasks that require solving is always based on relevant market trends, and ranges from improving the efficiency of marketing campaigns to preventing manufacturing defects and combating fraud.
Project implementation period
Outlets across Russia covered by the project
Accuracy of personal product recommendations
Number of loyalty card holders
Increase in average customer purchase amount after system implementation