Promo personalization for customers
Company - food retail, premium segment
1
Objective
Implement a system for selecting targeted offers as an alternative to the current expensive solution.
According to the condition of the problem, the PredTech solution should show better results compared to the results using the client's current system. Margin growth targets were to be achieved
According to the condition of the problem, the PredTech solution should show better results compared to the results using the client's current system. Margin growth targets were to be achieved
2
Work description
- A solution has been developed in which personalized offers are being generated for each client on a weekly basis - 5 personal offers.
- Conducted a cluster analysis of the customer base for various Target Marketing activities
3
Results
- Increased average check, frequency of visits, involvement in the new loyalty program
- Increase in PTO by 5% for the processed customer segment compared to the current approach
- Increase in Gross Margin by 1.1 percentage points for the processed segment compared to the current approach
Customer base clustering and benchmarking
Company - food retail, economy segment
1
Objective
Carry out clustering of the customer base based on the checks historical data
2
Work description
- Cluster analysis carried out
- Benchmarking based on the received clusters has been developed
- Comparability of segments has been assessed
- Missing and incomplete customer segments identified
3
Results
- Conducted cluster analysis and benchmarking
- Formed recommendations for attracting customers to the Target Core cluster (frequent visits, large average check, variety of purchases, various missions of visits)
- A marketing approach has been identified for the further development of the loyalty program
RFM clustering and cashback selection to maximize ROI
Company - a hypermarket chain
1
Objective
Develop an approach to grouping customers to optimize approaches to working with a loyal customer base
2
Work description
- RFM-clustering has been carried out, clusters have been selected according to income, regularity and longevity.
- Based on clustering, 8 groups of consumers have been identified
- A recommendation model has been developed to determine the optimal level of accruals depending on the client group
- For each group, bonuses are determined in the context of cashback accrual
3
Results
Achieving ROI - 53%
Offers’ personalization and the average check increase
Company - food retail, middle class
1
Objective
Finding missing goods according to the typical basket of the cluster and dropping out goods of the client, as well as customers prone to more expensive substitute goods
2
Work description
A system has been created to identify personal offers to increase the average check for subsequent marketing campaigns by identifying drop-out products
3
Results
- Average check growth: 3.7%
- Purchase frequency increase: 0.6%
- Net response: over 1.5 percentage points compared to the customer's current results
Toolkit for short-term forecasting of revenue, margins and retail KPIs
Company - Fast-Fashion retail
1
Цель проекта
Development of a system for short-term forecasting of revenue, margins and retail KPIs
2
Work description
- A system has been developed for weekly rolling forecasting of revenue, margin, average bill, traffic, conversion for the entire offline stores chain for a 1-month horizon with daily and weekly granulation
- To create the system, data on consumer activity and the functioning of the stores chain were used: the history of customer purchases, average price levels, the availability of promotions, production and weather calendars, etc.
3
Results
The error (MAPE) of the model is only 4-5%, in ordinary months
Development of models for Sell-OUT forecasting
Сompany - a major beverage manufacturer
1
Objective
Building a toolkit for Sell-OUT forecasting
2
Work description
- Models have been built to accurately predict sales volumes
- The model is designed to align the logistics supply chain, align the production process and accurately predict warehouse utilization and inventory
- The model works on a planning horizon of 4-10 weeks
- Prediction has been developed for promo and non-promo weeks
- Points of sale number - over 600
3
Results
- Achieved weighted average accuracy rate of 86%
- Developed a price sensitivity model for the type, size, duration and frequency of promo weeks
System for predicting the number of vehicles for an order
Сompany - a major manufacturer of low-alcohol drinks
1
Objective
Develop a system for planning vehicle orders in the context of Warehouse-Tonnage-Day, which would increase the accuracy of vehicle forecasting. At the same time, it was necessary to achieve a reduction in the cost of re-ordering and idle transport
2
Work description
- A predictive model of the number of vehicles has been built to meet the supply plan based on historical data.
- A daily forecast of the number of vehicles has been made with an accuracy of 95%, indicating the required tonnage and the final delivery point.
- When modeling, compound routes and vehicle restrictions have been taken into account
3
Results
- A model has been developed with an overall accuracy of 90%, the accuracy of the surplus is 93%
- Added increased penalties to the model to increase surplus accuracy: 89% overall accuracy, 98% surplus accuracy
Greenhouse yield forecasting
Company - one of the largest vegetable producer
1
Objective
Toolkit development for predicting yields (kg) in greenhouses for various hybrids of tomatoes and cucumbers
2
Work description
- An automated import of data on the facts of collections, as well as data from smart sensors (light, watering, temperature, CO2, etc.)
- Implemented automatic parameter selection for models (self-learning and self-actualizing models)
- The final approach is based on the use of a trend line model (short and long trend), weather conditions (including weather forecast), phenological parameters, and historical collection data.
- A yield forecasting model has been built with planning horizons one and two weeks ahead
3
Results
- An increase in accuracy up to 8 percentage points for predicting the yield of various hybrids of cucumbers in comparison with the existing approach
- An accuracy increase of up to 5 percentage points for predicting the yield of various tomato hybrids in comparison with the existing approach
Development of models for end-to-end forecasting
Joint forecasting for a beverage manufacturer and a large retail chain
1
Objective
Development of predictive models for end-to-end forecasting of the goods volume for a retail chain and a beverage manufacturer
2
Work description
- Backpropagation approach for forecasting has been used
- Sales volumes are predicted for retail outlets, on the basis of which the volumes of orders for DCs are predicted. Further, based on the volume of orders at the DC, orders are predicted for the supplier's warehouse
3
Results
- Forecasting warehouse shipments using the backpropagation method with an increase in accuracy by 20 percentage points
- Weekly forecast accuracy for stores - 85%
- Weekly forecast accuracy for DC orders - 79%
Dashboard reporting development
Company - a major clothing manufacturer
1
Objective
Develop a reporting system for the company's operating activities in the format of dashboards
2
Work description
A reporting system has been created that contains:
- Sales data by sales channels (money, units)
- Information about the achievement of sales KPI by stores
- Data on the distribution of sales by regular price and markdowns
- Information on sales levels by product category
- Data on the speed of sales of goods depending on the total output.
- A special color system for the speed of sales for the prompt correction of the new production volume
3
Results
- Dashboard reporting has been developed using Power BI
- Additional reports have been developed in Excel using VBA programming to fully automate work with multiple data sources
System for assessing satisfaction with the technical service quality
Company - Telecom Service Provider
1
Objective
Develop a system for assessing satisfaction with the quality of IPTV services
2
Work description
- About 90 predictors have been identified for the duration and frequency of television viewing, as well as information from IPTV set-top boxes for signal quality and an aggregate indicator of the number of dead pixels at the time of channel switching
- The model has been trained with the gradient boosting algorithm using cross-validation
3
Results
- The achieved accuracy of the model - 81%
- Reducing incoming service requests - by 20%
System for identifying subscribers inclined to purchase additional services
Company - Telecom Service Provider
1
Objective
Develop a system for identifying Internet access service subscribers inclined to purchase IPTV services, and identifying television service subscribers inclined to purchase additional channel viewing services
2
Work description
- 920 predictors have been generated for billing usage, services, and DPI statistics.
- For each channel package, the model has been trained with the gradient boosting algorithm using cross-validation
3
Results
- Growth of the average Cumulative Lift - 4.5 times
- Concentration of subscribers inclined to purchase additional services - 65%
Identification of churn-prone subscribers for further retention
Company - Telecom Service Provider
1
Objective
Develop a predictive model to proactively identify churn-prone Home Internet service subscribers using PON technology for their subsequent retention
2
Work description
- Binary classification has been used
- More than 400 billing and traffic predictors have been generated
- An ensemble of predictive models has been developed
- Clustering of the subscriber base has been carried out, a model has been trained for each segment with the gradient boosting algorithm using cross-validation
3
Results
- Growth of the average indicator of Cumulative Lift - 15 times
- Achieved accuracy of forecasting for top 10 thousand - 75%
BSSL International Business Startup Services Ltd