Saturday, 1 November 2014

The "TIME" variable in data analytics and personalization

Ritu and I are now serious online shoppers. Last month we started buying vegetables and fruits online and with this, almost 90% of our monthly needs are shopped online. Very thrilling.

And with our own “desi” portals giving world class online shopping experience, we should be thrilled. Right?

Well, almost.

6 months ago, we re-started to play tennis. Ritu and I. And we used to buy a set of 3 balls every month. The recommendation engine on FLipkart and Amazon were identical. Suggesting other brands of balls. And telling me that people who had bought tennis balls, also bought racquets etc etc. Ok. Thank you.

But its been 6 months now and Ritu and I now buy a set of 3 balls every 10 days.

What should Amazon have figured from this? That we play frequently, and we are getting better at the game. And possibly play longer. So the recommendation engine should have started to give me energy drinks/ hand towels/ shoes/ maybe some books on tennis……Nope. Nothing like that. It still gives me the same other brands of balls etc.

This is not “personlisation” that big data analytics was supposed to do. This is the same small data analytics extrapolated to big data. And this happens on every category. Kids diapers….my kids gradualed from diapers to kids underwear. BAM! That should have told FLipkart something. Start suggesting books- other interesting stuff for 3 year olds. The change in product signals a big shift – it’s a counter that allows Amazon or FLipkart to estimate when my kids will be 5/6/7/15 or 30. And plan to seed me and my kids. Dont recommend basis what 1 million other dad bought when their kids moved from diapers to underwear. Mine did it in 3 years, some one else's may do it in 4. The kids are different.

Netflix i believe does a lot of reasearch on consumers and is very kicked about its recommendation engines. I was curious and looked at the analytics they do. Lots of cross tabs...etc.

But lets look at this case: Ritu Venkatesh (my wife’s name) has lived in 3 cities in the last 5 years. She was watching movies like “Marley and me” till 4 years ago when she switched to Snowhite and the dwarfs. Then last year, she also added “Frozen” and “Toy Story” for my eldest who turned 4…

So Net flix should know that this is a female customer, North Indian (Ritu), married to a South Indian (Venkatesh) on a transferable job; watching kids movies of different genres (have multiple kids)… while they have a lot of recommendations..i realised they were losing marketing or cross promotional opportunities basis what more they can put together on the customer.

They could be cross selling Ritu massage treatments/ nanny recommendations / governesses/ early development kids games etc…

Really excited as I write this. Recommendation engines use millions of data points to predict what I should need. But forgetting the rate at which I migrate from one category to another. This we believe would be different from user to user and adds a layer to the modeling exercise that today we dont see at some large retailers. Will keep busy think of the applications of this approach in consumer marketing.

The TIME variable in data analytics and personalization.

Ritu and Venkat

No comments: