The Hadoop cluster has since grown to 64 nodes, holding up to 70TB of data, which is "growing at a pretty rapid pace".
"The beauty of this system is that going from a 32 node cluster to a 64 node clusters in Hadoop, it is not a massive investment at all. To process this kind of data in your traditional databases, you are talking a 10x multiple," Saraf said.
The cluster is used to analyse a variety of structured and unstructured data by a team of data scientists. This includes transactional data, click-stream data from Omniture to provide insight into digital services, log-monitoring from Splunk and risk management information.
Through the Hadoop platform, Western Union is able to support the pattern recognition and and predictive modelling that allows it to personalise user experience.
The company has also invested in other analytics technologies. This includes a range of Tibco tools to speed processing and enable real-time decision making, SAS business intelligence software, Tableau for data visualisation and reporting, and Informatica for data integration.
By harnessing its data Western Union is able to enhance its relationship with customers in a number of ways, Thompson says. For example, for a business with a global reach, there is value in tailoring its services for returning customers to provide them with the right currency and targeted recommendations.
A major part of Western Union's business is supporting consumer transactions - which account for around 80 percent of its overall revenues - particularly migrant workers sending earnings back to their home country. According to the latest World Bank figures, remittance payments are expected to reach $550 billion this year, and set to grow to $707 billion by 2016.
By analysing the patterns of worker movements across the globe, Western Union is able to anticipate the needs of customers as they move across the world with greater precision - monitoring the flow of data across the thousands of different 'corridors' through which Western Union operates, and identifying customer behaviour.
"We are now able to clearly see immigrants living in, lets say, the Netherlands, and very quickly pick up the habits of that country. They might prefer payments that are native to the Netherlands, meaning they won't use credit cards as much as they use online banking and transfers," he said.
"Very quickly we saw that, when we offered the payment method customers wanted in certain regions, we would see a dramatic uplift in our conversion rate - it has a direct correlation."
Gaining this feedback was possible before, says Saraf, but it was slow to capture, and hard to scale. This made it difficult to recognise returning customers and preempt the services they would require or decisions they would make - such as offering recommendations based on past transactions.
Sign up for Computerworld eNewsletters.