As a central bank, industry developer and regulator, the Monetary Authority of Singapore (MAS) relies on a broad set of transactional, behavioural and social media data to craft policies and drive the development of Singapore’s financial industry.
By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers.
MAS’s challenge, however, is not about the availability of data, but how to collect data in a way that’s scalable and makes sense so it can be effective in fulfilling its mission.
David Hardoon, its chief data officer, says the answer lies in smart data.
But what is smart data? To Hardoon, who spoke at the Chief Data and Analytics Officer Singapore conference earlier this week, smart data has several attributes. First, it is data that has intrinsic value, and contributes to an organisation’s key performance indicators.
Second, smart data needs to have a certain quality so that it can be trusted by people, Hardoon says, emphasising the importance of data governance. “If you do not have trust in what’s happening with the data, you may think that things are not true or accurate.”
Hardoon says the MAS has put in place data governance processes through automation, which is all “about taking processes and routines to ensure that data is at a level that we can trust.”
Once that has been established, MAS then looks at suitable methodologies that can used to make sense of that data, whether it’s deep learning or machine learning in both supervised and unsupervised forms.
Hardoon also touched on the accuracy of data, which he says is “entirely irrelevant” and needs to be contextualised.
“If you don’t a baseline for data accuracy and you move to 65% accuracy which has helped to improve customer satisfaction and efficiency, do you really care about 65% accuracy?,” he quipped.
“Just because we have shifted to a data-driven, statistical world doesn’t mean we need to have 90-95% data accuracy.”
To Hardoon, smart data also means building data models that can be generalised and applicable not only to specific instances. “What we build can’t just be a one-hit wonder. It must have generalisability and robustness,” he says.
Finally, smart data should be interpretable, which means an organisation needs to know why and how a data model works. Hardoon says to many people, a data model may seem like a black box, but when something untoward happens, somebody needs to know why.