Customer Success: Improving operational efficiency for US based Telecom Manufacturer with $60B revenues
The client was handling millions of alarms on a daily basis. The NOC engineers spent lot of time on monitoring false alarms and so were blindly suppressing 95% of the alarms. Existing measures were all reactive.
A Pareto analysis was performed to obtain the fraction of the significant alarms. This analysis showed that 5% of the alarms ids were causing 85% of the alarm. Next, 54 sequences of simultaneous alarms were extracted from the “simultaneousness” study. This reduced the false alarms by 8%. Defined and constructed a matrix of Normal and Abnormal behavior of alarms. Based on mathematical methods and heuristics, we obtained subsets of abnormal behaviors of alarms. It constituted 58% of the total alarm volume. So, the NOC engineers had to attend to only 42% of the alarms.
Customer Success : Predicting Anomalous Patterns generated from US Pacemaker equipment manufacturer
The client was pacemaker manufacturing company which is regulated to ensure that the requirements of the Medical Device Directives are met. Allowable percentage of defects was 1.5%. The client manufactured and sold around 7000 devices with around 3% defect.
We collected the electrical impulses generated by electrodes within the pacemaker. Used pattern recognition methods to identify the inconsistent behaviors and raised triggers well in advance and bring down the percentage of defects to acceptable number.
Customer Success: Improving Warehouse Planning for a growing retail chain
A growing retail chain with 15 outlets, 3 warehouses, 3500 items required a back-end system for warehouse decision support. In addition, they wanted analytics models for predicting the procurement and cash flow gaps.
Designed a centralized data warehouse and generated much more informative reports than were done. Built predictive models for demand forecasting and also supervised learning methods were used that considered floor manager inputs for better decision.
Customer Success: Identifying customer segment for a leading multichannel home shopping retail client
The client sends about 300M catalogs with a dismal success rate of 3%. They wanted to identify the optimum customer base for maximum ROI. They also wanted to measure and display loyalty shifts across brands to help decide optimal campaign strategy
Segmentation based approach was adopted to classify customers into most probable responders and non-responders. This was based on several characteristics or behavior retrieved from transactional data. Preditive models were built which increased success rate to 5%
Customer Success: Maximizing Returns while minimizing Risk for Fund Manager
A leading Mutual Fund Manager wanted to build a portfolio that had an optimal diversification of stocks such that the allocation of funds among the stocks resulted in maximizing returns.
Historical data was investigated to understand the risk and returns of individual stocks. Designed a quantitative allocation strategy which allocated capital to produce optimal risk adjusted returns. Various experiments were conducted to identify the investment durations and frequency of reallocation to maximize returns.
Customer Success : Optimal infrastructure Upgrade process for an Oil and Gas Major with revenues $25B+
Clients infrastructure has over 70 thousand servers and every year they spend several million dollars on the infrastructure upgrades that is often done without in depth understanding of the actual requirements
We developed a mathematical framework to analyze and measure the systems performance of a cluster, and by using Artificial Intelligence techniques like Genetic algorithm we developed predictive model for a cluster based on historical data. A decision support system was built which allowed client to schedule the projects and identify the time periods when upgrades would be required.