Using Machine Learning to Determine Drivers of Bounce and Conversion

Calling all marketing and sales to the boardroom. At Velocity Conference Santa Clara 2016, two of my favourite and most respected voices in performance - Patrick Meenman and Tammy Everts - presented findings from a collaborative study between Soasta and Google. Determining the drivers of bounce and conversion are crucial for both companies being that they're extremely active and have vested interested in performance on the web. As such, they teamed up for a two part research study: Using Machine Learning to Determine Drivers of Bounce and Conversion
Making sure that all marketing and sales are in the boardroom for this will help connect the dots when they are traditionally the ones who have been referenced when performance has trended south - ads, a/b testing, insistence on marketing materials like large images to name a few. So this talk certainly spoke to an audience outside of the typical team of DevOps/engineers.
The June talk, which made public interesting findings, was then followed by a subsequent study in the fall in NYC. Both were great talks, with even greater findings. Better yet, the conversation that surrounded the study, and that the following statement help true:

Everything matters - Tammy Everts.

Certainly, performance was in focus and of the 93 inputs collected (and Soasta collects a lot of data), they looked at the top 6 bits of info affecting conversion rates:

  1. of elements

  2. of images on pages

  3. of scripts

  4. Front end load times
  5. Full page load times
  6. Back end load times

Then did the same for bounce rates:

  1. DOM load
  2. Full page load times
  3. of page elements

  4. Front-end load times
  5. of scripts

  6. Back-end load times

Watch both Patrick Meenman and Tammy Everts discuss the findings in full and be sure to read the coordinated posts.

Part 1: Velocity Santa Clara 2016

Part 2: Velocity NYC 2016