| | December - 20189of what the data is going to be used for. With data flowing into organizations at lightning speeds, it's becoming important to focus on extracting meaning from data early in the data creation/acquisition process. More organizations have to deal with garbage or irrelevant raw data occupying mind-boggling volumes in the lake. The problem is starting to become significant enough to give rise to real-time analytics. Rather than just storing away data, real-time analytics focuses on storing `insights.' But, this is not easy and it requires sophisticated analytics, and sophisticated resources. Organizations are investing in Data Science teams to condense the large volumes of data through algorithms and store away only meaningful information. Technologies like Apache Storm, Spark, and Azure Streaming Service, are starting to lay the foundation of real-time use of data science in decision making. Unlike the traditional financial service fraud models that used older platforms and proprietary solutions, which took over six months to refresh data models, the newer real-time platforms have ubiquitous application potential, capable of transacting at millisecond speeds and providing almost instantaneous model refresh capability. Data Science can now react to changing market needs at break-neck speeds.Machine LearningMachine Learning (ML) is very closely related to analytics. In fact, a lot of the techniques used in traditional statistics and analytics form the foundation for modern Machine Learning and Artificial Intelligence solutions. The key difference though is the fact that Machine Learning techniques rely on human labeled data to get into a continuous learning loop with their algorithms. Machine Learning algorithms can use complex correlations between various data streams to learn how to predict important decision points for the organization. While ML represents a great opportunity for almost all industries, enabling production ready ML systems require platforms that are capable of state-of-the-art data ingestion, data transformation and model hosting. Most organizations stop short when it comes to implementing such production-ready ML IT solutions. The key is to embrace change management (within IT and Data Science teams) to develop a new DevOps model that works seamlessly with engineers and scientists alike, while driving the entire company towards an analytics-driven organizational model. Fail-Fast ExperimentationThe goal of moving towards an analytics-driven organization is to allow faster innovation and to experiment with new ideas in quick time. This is especially true when it comes to customer facing departments, solutions and products. Fail-fast experimentation, especially social-media driven A/B testing is rapidly changing the way companies are reaching out and experimenting with their customers. Starting with web technology companies to traditional retail companies, organizations are moving to innovative fail-fast experimentation and pushing the frontiers of traditional marketing. Setting up and running these A/B experiments is hard and it demands great teamwork between various parts of the organization before a viable outcome can be materialized. In order to ensure there is an organized system for mining Big Data and extracting insightful outcomes, it is important that organizations invest in fail-fast experimentation platforms and ensure there is a good process to triage the new ideas and innovations in a structured manner.Testing-in-ProductionTesting-in-Production (TiP) as the name implies, reduces the typical DevOps cycle down such that developers can push out solutions to the customers with little or no delay. Usually DevOps takes long because of the traditional Dev, Test and Release cycles. TiP approaches this problem in an innovative way. For a start, it's common to define a metric of customer dissatisfaction that can be measured very quickly and effectively through one of various customer touchpoint channels. With this metric defined, developers (and experimenters) are given near production release authority. The platform is designed to allow new code release, to serve a new experience, to a randomly selected group of customers, while holding back a control customer population for comparison. The customer dissatisfaction is constantly measured on both groups and even the slightest increase in dissatisfaction will auto trigger the platform to revert to the old experience, thereby rejecting the new code release. This enables the organization to rapidly innovate with little or no friction in the ideation process. TiP works seamlessly with Agile DevOps and some of the pioneering Internet companies are adopting this methodology.
< Page 8 | Page 10 >