Improve loyalty through decreasing irrelevant personal offers
GOAL Develop prediction algorithms
of total spending of all customers for the next 30 days for each service/product category
of individual customer spending in the next month for each service/product category
Data on 7M card transactions by 15K customers, 184 categories of services/products
For Goal 1, an ensemble of models was a solution: time series prediction, linear regression, decision trees
For Goal 2, we built a model based on statistics for prior periods and extended statistics on various combination of attributes
BENEFITS Sberbank Data Science Journey Contest
Goal 1: 8th place
Goal 2: 2nd place
Out of ~ 800 participants
Grupo Bimbo is the worldwide leader in baking with the largest dealer network located in Mexico and USA. The product range consists of over 100 brands, represented in 2.5M stores, which are connected by 52K logistic routes.
Maximize sales by fully meeting the demand
Minimize cost of returns (food spoils)
The goal was to develop a model predicting product demand at POS's
History of sales for 7 weeks (81М transactions) was provided
SOLUTION We applied a machine learning algorithm based on statistics for prior periods and expanded statistics on various combinations (product, group, location, manufacturer, route, channel…) Absence of about 70% of data on weekly sales/returns (an average history is 2 – 3 weeks) complicated solution of the task. Returns could exceed purchases for the prior period, whereas for some products there was no history of demand. BENEFITS In the built model, the RMSLE metric was 0.4436, which allowed decreasing returns of unsold products by 40%, or by about 1% of turnover, or by $140M.
Business Travel Bureau
BusinessTravelBureau is a Russian business travel company providing a full range of services for business and personal travel.
Build a cost-effective system for attracting VIP travelers via digital channels
The goal was to develop a system for searching for people interested in VIP travel services
Use open data
Data were collected using automated webpage decomposition of the TripAdvisor website and extraction of user, location, and hotel features.
Source data contain mainly categorical text features.
Text mining technology and a neuronet were use
BENEFITS The built model showed accuracy of ~85%, which was sufficient to start a new business under the brand of Epico Travel.
Home Depot is one of the world's largest online and offline retailers of home products Turnover > $83 bln 6 countries 2274 huge stores (average floorspace is 9,755 m2) 40k products (offline) 300к+ products (online)
Improve customer loyalty
The goal was to develop an algorithm which would automatically evaluate the relevancy of search results to a search query entered by a user
~ 74 000 search queries on ~55 000 products were provided
A large amount of source data was provided as text
We used all varieties of Text Mining techniques to identify semantic similarity of words and texts and semantic clusterization
The resulting model was built as an ensemble of decision trees.
BENEFITS The suggested model was 15% more accurate than the standard method for estimating the relevancy of search results to a search query, which was based on keyword comparison.
StilGiyin is a Turkish aggregator of products from major online fashion stores (clothes, shoes, jewelry, cosmetics, accessories, etc.) Over 100 000 products (online)
Increase sales through a higher product page conversion rate
The goal was to develop an algorithm which would evaluate the probability of customer clicks on certain products ~ 20 mln products views and clicks were provide No data on visitors SOLUTION
We built a probability model based on the frequency of products displayed togethe
The model showed accuracy of ROCAUC=0.76
BENEFITS The suggested model allows optimizing product search results for individual customers and increase the conversion rate by 14%.
Maximize sales by fully meeting the demand.
Free up assets from stock with a low turnover rate Improve dealers' performance and as a consequence, expand the dealer network.
The goal was to develop a model predicting product demand by dealers
A history of purchase plans and actual purchases by dealers for 2 years was provided
SOLUTION Such data characteristics as a large amount of missing values and no history for some dealers wouldn't allow us to use typical methods for time series predictions. Instead, we built a model based on statistics for similar companies.
BENEFITS By expert evaluation of Company representatives, the suggested model would allow decreasing the number of slow-moving items by 150 — 200% compared to initial dealer orders.
Custom mobile apps development for the Russian Railways
Customer:JSCO "RZD" is a Russian vertically integrated company, both managing infrastructure and operating freight and passenger train services. In 2012 it became one of the three largest transport companies in the world. The company was established on September 18, 2003.
Project Fayrix developed 2 cross-platform mobile apps for RZD: "RZD Online" and "RZD to Passengers".
RZD Online is mobile service, developed for providing RZD customer feedback and to tackle it in a timely manner. The app includes 3 modules: CONSULTING, SOLVING PROBLEMS, NEWS. RZD clients can ask questions or submit complaints via the app and as a result receieve an official response of ZRD dedicated experts. There are 39 feedback topics on the railway infrastructure. The NEWS module lists the latest RZD corporate news and news on Russian public transportation in general.
RZD to PASSENGERS is an official mobile service of JSCO "RZD" to buy long-distance train ticktets without additional fees and commissions. The ticket prices are 100% the same as on RZD official web-site. Users can buy tickets with a bank card, Apple Pay or the "RZD Bonus" card. The app also provides accurate train timetables and ists the latest RZD corporate news and news on Russian public transportation in general.
Online authorization module for VTB Bank
Customer VTB Bank (PJSC) and its subsidiaries form a leading Russian financial group – VTB Group, offering a wide range of banking services and products in Russia, CIS, Europe, Asia, Africa, and the U.S. Project VTB Bank cooperated with Fayix in a Time & Material model. Fayrix developed a brand-new user account system for the bank mortgage partners. The project required to implement one of the most secure authorization and authentication mechanisms, so that bank and partner credentials and operations stay 100% safe. Fayrix provided the bank with 7 full-time developers, 2 QA Engineers with a dedicated project manager. The whole project took 6 months and was successfully launched.
Docflow system customisation for Softline Ltd client
Customer: Softline Ltd client
Project: TEZIS is a modern Enterprise Сontent Мanagement (ECM) system, primarily targeted at the business process management (BPM) area - it serves as a perfect tool to automate a company's processes of any complexity. TEZIS features standard out-of the-box solutions for task management, document approval and standard office processes such as processing inbound and outbound correspondence or managing document archives.
Fayrix has partnered with Softline to provide its client with OOTB software customization service. Fayrix customized EMC TEZIS to particular client's document approval process and integrated the TEZIS software infrastructure with the client's corporate 1C, MS SharePoint and Active Directory portals.
Docflow system customisation for Inteza Bank
Customer Banca Intesa is a Russian subsidiary bank of the Intesa Sanpaolo Group, one of the top banking groups in the euro zone and leader in Italy in all business areas. Banca Intesa is a full-service bank: it provides banking services to enterprises and to individuals. Project TEZIS is a modern Enterprise Сontent Мanagement (ECM) system, primarily targeted at businesses in Russia and CIS states. The focus of the system is in the business process management (BPM) area - it serves as a perfect tool to automate a company processes of any complexity. In addition, TEZIS features standard out-of-the-box solutions for task management, document assignment and standard office processes such as processing inbound and outbound correspondence or managing document archives.
Fayrix has partnered with Banca Inteza to provide it with OOTB software customization service. Fayrix customised two EMC TEZIS modules: the "Document Management" and the "Legal Service" to particular bank document approval process and integrated the TEZIS software with the client's corporate portals.
Sony online accessories store for Android
Customer: CJSC Sony Electronics distributes electrical household appliances, radio and television goods, and other electronic equipment. The company is based in Moscow, Russia. CJSC Sony Electronics operates as a subsidiary of Sony Corporation.
Project Fayrix developed the official Android app for Sony. The app is a mobile online-store of Xperia accessories. The shopping app automatically detecting user's mobile device and suggesting products based on model. Easy navigation, purchase, pre-order and order tracking are featured.
"Movie for a Journey" mobile apps for iOS and Android
Customer: Kino V Dorogu LLC
Project Fayrix has partnered with a startup company Kino V Dorogu LLC to develop iOS and Android apps for their "Movie for a Journey" project. The "Movie for a Journey" service incorporates 2 components:
Mobile apps to select and purchase movies for iOS and Android;
Branded movie points in public transportation terminals: airports, railway stations and bus stations, where customers can download their purchase to mobile devices using a dedicated high-speed wireless connection for watching the movies offline on a trip.
Users can search for movies they like by genre, year, name and critics rating. The app regularly offers discounts, editorial collections, serials and movies for free download.