Ranking and Suspensions From A Data Science Perspective With Yev Marusenko

Yev is sharing on data science and a perspective on how Amazon’s algorithm is spotting artificial spikes and measuring them against ‘real spikes’ and against other similar accounts for a relative measure to assess signals and patterns that may be driving this pattern of suspensions.

  • Doing lots of maths and statistics
  • Approaching everything with this background
  • Work on ZonTracker


  • Is to help with more data on external traffic
  • Shows Amazon sales coming from Facebook ads and direct link to Facebook ad manager
  • Automated and able to pull Amazon data in Facebook to do re-targeting or build lookalikes
  • Integrating Facebook with Amazon

The Hypothesis

  • A combination with recent chatter and breaking news
  • Maybe people are making more out of it than there is
  • Seeing the data from 2 different ways
    • Talking with hundreds of Amazon sellers and Facebook advertisers and seeing the data
    • Having a data-driven perspective and how what the data tells us
  • Amazon can switch the algorithm but data is always the same

Account suspensions

  • Relationship between discounted giveaways
  • Link with using specific URLs
  • Never known an account be suspended exclusively based on one a promotion or any type of URL
  • No absolute proof of this
  • How Amazon sees the data
  • Different factors that Amazon looks at on data science
  • This has to be an automated match
  • Amazon knows using a prediction model
  • They can tell what a new account will ‘look like’ and a newer account has less products than older accounts
  • This is one of the factors that is correlated with other factors
  • Newer accounts compared to older accounts data breakdown

Old School Marketers

  • They don’t try and game the algorithm coming in
  • These are grandfathered in accounts
  • They just have popularity maybe off-Amazon and they are not tech savvy but have lots of searches and velocity
  • They are getting the right factors because of their popularity not because they are gaming the algorithm
  • Very different signals being given to Amazon in the data spikes coming in

Artificial Spikes

  • When a brand generates an off-Amazon traffic spike through publicity in the real world
  • This spike will generate with it secondary spikes as newspapers and other media pickup on the publicity, these filter through to Amazon
  • Typically Amazon Sellers produce a different kind of ‘Artificial Spike’ with our tactics to draw in traffic and Amazon’s algorithms can spot this
  • From a data perspective it makes sense that Amazon sees variability in the spikes

Relative to other Accounts

  • Assume you are running your account and all of your sales come from a discount code and that’s a signal to Amazon
  • If you’re bringing in new products every couple of months and the same artificial spikes then that might be a flag to Amazon
  • How is that data relative to similar accounts – this is a comparison that Amazon might be using

Data Points Triggers

  • Amount of sales on a day – unusual activity
  • Organic Sales and sales bought through a coupon code
  • Age of the account co-correlated with a bunch of other factors
  • Diversity of the traffic – variability is different is the data science view
  • 2-3 year old account with a 300-500 spike of sales from a one type of traffic
  • Might have got lucky on a tv show and shows a same spike of activity
  • But the effects are different as it’s not just one type of discount code bringing it in
  • The amount of variability will send signals to the Amazon algorithm from a diversified account from a non-diversified

Diversification of Traffic

  • Diversification of the traffic is one key factor that impacts across a bunch of the other factors across your account
  • Sales Reviews is a very direct factor
  • Taking one of these factors and splitting the atom on it with a data science perspective
  • This would explain how some activities on one account will not trigger a suspension but will on another account
  • A 3 year old account might be different to a one year account
  • How can you diversify within a promotion – just flex

Funnels Diversification

  • There are so many different funnels going on with some brands across pinterest, facebook ,etc
  • They are diversifying it within a single campaign – a survey, a chatbot, ads – many ways to bring in consumers across different URLs
  • Amazon will think it’s a real publicity spike as it is diversified and coming from all over within their ad campaigns

How to Diversify your Traffic

  • When running your ad you have to keep optimising it
  • Maybe 20-30 ads and narrow down to the top 3 and optimise those
  • WOuld have said focus in on a certain area but that is the opposite of diversification
  • If you are testing a Messenger Bot or a Landing Page – have another form in the other type as well
  • Get it up and running just to drive the diversity
  • Use 3rd Party people talking about you, a blog or influencer on another website can help
  • Small bloggers on their own site will review your product and post
  • Simple set of diversifying the funnel you’re running
Share This
%d bloggers like this: