Telefonica has created a specialist team of data experts to help standardise business intelligence (BI) practices across its South American operations.
The telco giant operates on a federated model, with each international division responsible for its own organisational structure and technology choices. Business intelligence (BI) has always been the domain of marketing teams in each geography, making it hard to create a consistent analytics strategy across the business.
To solve this, the company created what it calls a global “unicorn team”, combining data scientists with complimentary roles such as data journalists (who visualise and present data), data wranglers, data engineers and line-of-business staff to deliver their own data solutions. The term comes from the idea that data scientists are as rare as unicorns.
This central data team allows the company to create a standard approach to BI across 19 countries in South America countries, John Belchamber, global leader of advanced analytics at Telefonica told Computerworld UK.
This blended team assist in the deployment of BI software tools that were previously prescribed by the individual IT departments, an approach which slowed new projects in some cases.
This new approach allows each country data analytics efforts "to be more quickly monetised," by surfacing a more granular view of customer habits in order to deliver more targeted marketing, and to be more reactive in order to reduce customer churn.
Belchamber's team then provides on-line training and a community for the global teams to share best practices once they are up and running.
Telefonica allows each business unit to make their own technology choices, as long as the tools are being employed to a specific end. "We are open source, we take the business goal first and let these teams choose the tools they need to get there," he explains.
The overarching aim of all of this is the same as most organisations: to understand the customer better. The key to this is better customer profiling. The BI team is taking a few different approaches to this, from analysing contact centre call records, to parsing web browsing data into a customer profile
In terms of web-profiling Belchamber's team looks to give each business unit a 'dictionary', which is essentially a piece of bespoke-BI software that parses network data -- so web pages visited, when and on what device -- and builds a media profile of that customer to be used by the local team to deliver targeted marketing.
As a telco Telefonica looks at mobile browsing history or fixed line browsing. Naturally all of this is done within the legal framework of the geography that the technique is deployed, looking at customers that have opted-in, or using anonymised data.
Telefonica has also invested in speech and natural language processing (NLP) to discern sentiment from call centre chat logs. Unfortunately because of data restrictions the unicorn teams couldn't analyse the actual speech of the logs, so they anonymised the MP3 data and looked at the background acoustics. Belchamber says this still allowed them to analyse the "heartbeat of the call".
"We found within Latin American culture, who might be more expressive, there are peaks and troughs from which we can predict a good or bad call". Belchamber says this is the "first stage to understand customer journeys and introduce history into those transactions".
The problem is that most existing NLP APIs, such as Microsoft's, are built for English speakers, so the organisation has had to start from scratch. "We are now building true speech analytics and NLP pieces," Belchamber says.?