Analytics : More Than Just Dashboards

To many, the term ‘analytics’ seems to conjure up thoughts of nice flashy QlikView/Tableau/what-have-you dashboards. I strongly believe that everybody benefits by expanding their perception of its potential use cases.

So, what is Analytics?

We Humans can perceive 3 (maximum 4) dimensions.  Think of Analytics as a N dimensional space where 2 or more disciplines meet in 2 to N - 1 dimensions.

Nirmalya Ghosh, 05 June 2015

We’ve heard of terms such as Big Data, computer vision, data mining, decision science, graph theory, machine learning, natural language processing, pattern recognition, predictive modeling, etc.

Most analytics projects employ a combination of these and other techniques.

In recent years, the term analytics is often used interchangeably with ‘data science’. Unfortunately, given the current hype, often other disciplines such as ‘business intelligence’ (BI) is also being referred to as ‘analytics’.

Does it matter? You might ask.

It depends on who you ask. If one’s livelihood depends on BI, it is only natural to ride the growing trend towards all things ‘analytics’.

My opinion : I think it matters only if it limits one’s perception of its potential use cases.

Over the years, I have used several examples to help expand others’ perceptions. Here are a few, some of which I’ve been involved with.

Predicting Intent : Pregnant Shoppers

In early 2012, the New York Times ran a story about how Target was able to figure out which of their shoppers were pregnant and (interestingly?) when they were due to deliver - just based on their credit card transactions (and/or email address).

Predicting Intent : “Flight Risk”

In 2011, HP piloted a system which attempted to predict which workers were most likely to leave. The underlying prediction model(s) might consider many factors (referred to as ‘features’) such as employment length, number of jobs till date, number of promotions since joining date, average rating over past N evaluation cycles, whether or not he/she has had the same manager since joining date, perceived % difference in pay from best in industry, etc.

Earlier this year, there was a competition held to identify & model key indicators of employee attrition for a flash memory storage manufacturer.

Employee attrition models are not very different from, say, how customer churn prediction models work. The features used are of course very different.

Figuring Out Competitors’ Likely Bid In An Auction

In Singapore, land parcels are auctioned off by the government. The difference between a winning bid and the nearest looser is often several million dollars. It is imperative that wide margins be minimized. Considering multiple features, for a particular parcel, it is possible to figure out if a company’s bid is likely to be successful in an auction and if so by how wide a margin.

Figuring Out One’s Location In The Near Future

It is possible to predict where one is most likely to be at a specific date & time in the near future (say, 7 days), even if the predicted location is a place one seldom visits.

More recently, there was a Kaggle competition aimed at trying to improve the efficiency of a taxi dispatching system by trying to predict the final destination of a taxi while it was plying a fare.

Predictive Policing

PredPol is a system that helps significantly reduce crime by predicting the likelihood of criminal activity occurring in an area on a specific date and time in future.

Automated Meeting Minutes

A startup applies machine learning and natural language processing to determine which parts of a meeting conversation were most important.

And The List Goes On..

For those who haven’t yet heard of Kaggle, now might be a good time to browse through some of their past and current competitions. Examples include predicting if an online bid is made by a machine or a human, predicting salary based on a job ad, predicting whether a mobile ad will be clicked, etc.

Hope this helps.