Gilad Barash of Dstillery speaks at Metis

Feb 23, 2016    

Gilad Barash, data scientist at Dstillery came to Metis to share about his work in Ad Tech on Tuesday 2/23. It was a comprehensive and technical overview of data science in the online advertising space. Here are my notes:

Best resource?

Data Science for Business by Foster Provost and Tom Fawcett. Gilad advises Metis to be able to look at data science from a revenue and business perspective in mind.

Advice for Metis Students?

  • 80% of work is in preprocessing, so be ready and be great at that step.
  • Also document process for every project so it’s repeatable in the future. This step is unfortunately missing in the field today.

First Gilad gave an overview of how Ad Tech works. Specifically, he gave some great illustrations of ‘Programatic Advertising’, a real-time bidding of advertisements on websites.

Advertisement started off on billboards, print, and media where audience sizes are large. But today with “Super Targeting”, marketers can get to “SegOne” where each individual consumer can be profiled and targeted for ads; audience sizes are now < 1.

Data today can from all sorts of places:

  • Data from browser behavior with cookies
  • Device behavior based on apps used, time of day, and
  • Location. Where do you work, live, and travel to.
  • Place of interests

In addition, they can perform crosswalks which tracks people between devices. So maybe you start the day off checking the news on your tablet, then respond to a few emails on your commute using your cell phone. Then you login to work PC and perform some research online. They can identify that it’s the same person using difference devices! Very interesting and maybe a little spooky.

There are still some challenges though:

  • Mobile devices provide lots of information, but location is not always known or accurate.
  • Bot traffic is prevalent on the internet. How can companies like Dstillery tell them apart from humans?

After the session, I ask Gilad about growth in Ad Tech, and he shared 3 sources:

  • Better customer targeting and segmentation with machine learning
  • More marketers and websites getting into the market
  • Faster analysis and more accurate pricing estimation to be competitive in bidding