In thedelivery service intheweb store, there was alarge percentage of non- redemption goods inthe"order online with POS pickup".
We have built ascoring model ofredemption probability estimation and optimised processes ofcommunication with therisk group.
On top ofdemand forecasting model, theclient required totake into account goods amortization and supply chain processes. Key features: estimation ofsafety stock according toeconomic factors and theeffective distribution ofgoods in thedeficit conditions.
The implementation of thesystem increased cost savings by25% from EBITDA.
Airport imitation model
An airport required asimulation model that would allow it todetermine theeffects of changes to theconfiguration of ground transportation and airport network soit would not depend solely ontheexpert opinion.
We built ahigh performance simulation system.
Implemented visual "What-If analysis" tool.
Forecasting ATM cash demands
Major ATM network requested to improve anexisting expert model for replenishing ofcash levels at ATMs.
We built amodel for optimal cash inventory levels estimation andreplenishment procedures forecasting.
Company (top5 inRuNet) implemented its own targeted banner advertising system, which had low margins from advertising campaigns.
We conducted aresearch of advertising banner traffic system features.
We built mathematical optimization models ofadimpressions and pricing campaigns and bids for RTB.
We implemented new high performance algorithms, which increased thecompany's margins.
Optimization ofproduct recommendations for remarketing
The company had aB2B-service for remarketing ofproducts with sufficiently low proposal conversion (close tobreakeven limit).
We developed machine learning models for product classification according to their compatibility etc., based onproduct data, consolidated from various sources.
We developed models and algorithms topersonalize product offers based on user behavior history available on thecompany's customers and its partner sites. We included solvency, LTV and other integrated customer data.
Proposals conversion was increased by 34%.
Predicting server failures
The company has alarge pool of data centers with server hardware (hundreds of units). It was crucial to provide anoptimal SLA, which would take into account failures, aging, and seasonal characteristics ofload balancing.
We developed anadaptive model of server hardware failures based on trend analysis ofits functional characteristics. After the model implementation, server failures decreased byorder of magnitude.
Data Center Service Management received aservice that optimized its work and reduced the peak load onthestaff due to themassive failures etc., maintaining theawareness level of deviations within 3σ.
Company management received reporting system for failures and maintenance of server hardware, depending on many factors, including design features, components layout and temporary factors, which allowed tomake better decisions about purchase and design ofequipment in thelong term.
If you have questions or suggestions, feel free tocontact us through thefeedback form