TABI - Near Real-time Business Intelligence

Timely Analytics for Business Intelligence (TABI) is a system that delivers near real-time business intelligence for analysts relying on massive amounts of data. It was developed by IBM Research and first field-tested in a First-of-a-kind (FOAK) project with Telco industry.

The project integrates two research assets from IBM Research – Haifa: the Massive Collection System (MCS) from the Software and Services department; and the Parallel Machine Learning (PML) toolbox developed jointly by ML group and the Data Analytics department at the IBM T.J. Watson Research lab.

The potential TABI use cases are prevalent in fields such as telecommunications, banking, and transportation. 

TABI connects to the data stream of the customer, extracting only relevant information and making rapid business intelligence decisions using methods such as prediction, clustering, social connectivity graph analysis, and association rules.

TABI has demonstrated that it can learn from hundreds of millions of records per day, with relatively modest hardware requirements.

An IBM Case Study

A joint research project between IBM and a large mobile communications provider offered strategic insight into the wireless customers’ social network calling patterns, giving the provider valuable information it could use to improve customer loyalty and reduce churn.

Mobile phone customers are notoriously fickle, and churn rates are a major headache for most wireless carriers. IBM’s Timely Analytics for Business Intelligence (TABI) uses sophisticated software analytics and algorithms to predict the likelihood of defection among current subscribers, giving the provider the opportunity to target those customers with special offers to encourage retention.

IBM Research conducted a joint first-of-a-kind project with the client, using software analytics and algorithms developed by IBM Research Haifa. Through TABI, the IBM Research Services team and the client discovered that customers’ calling patterns provide important clues about their brand loyalty. Looking at a customer’s network of friends and determining who calls whom is proving to be a reliable predictor of whether and how seriously that customer will consider jumping to a competitor.

The IBM Research Services team worked with the client to analyze more than 250 million call detail records (CDRs) every day using IBM’s TABI technology. TABI is based on a combination of IBM’s Massive Collection System and Parallel Machine Learning system. Unlike traditional methods, which “warehouse” and analyze data that is quickly outdated, TABI perceives patterns by analyzing massive amounts of CDRs as they are generated. This ongoing analytical process, which can handle billions of CDRs a day, helps TABI deliver results that are up to date. The use of social analytics provides predictions that are up to 50 percent more accurate than previous techniques used to predict churn. Customer privacy is maintained by identifying social network patterns based on CDRs, which are decoupled from any personal customer information.

For each set of data, TABI can apply one of several analytical operations to generate a Business Intelligence model that clients can use to glean actionable knowledge. There are currently five operations, including: prediction, in which forecasts are made according to the Key Performance Indicators of each user or entity in the data; clustering, where users are divided into classes according to similarities between them; connectivity graph analysis, which is used to analyze the data according to links between users; association rule mining, which automatically determines correlations between data items; and quality of service analysis, which provides advanced methods that ensure the network is functioning at a high level of service.In addition to loyalty reinforcement and cross-selling, TABI can be used in marketing campaign management to target likely customers and track progress, and for fraud or abnormal activity detection by rapidly adjusting its prediction model to spot recent examples of behavioral anomalies. It can also be used in network quality tracking, alerting a network operator to a potential problem and helping provide the opportunity to react quickly to emerging problems before they have far-reaching effects.

The Net of TABI

- a cluster forming algorithm

- good at finding the lead of cluster

- near real-time decision-making

News:

• IBM analytics tapped to predict, prevent churn (2009.04.15)

Links:

• TABI (IBM Research)