Erica Smith
Aug 23, 2017 · 2 min read

Hi Michael. Of course, I’ll give you an example from the company where I work.

We run a website where customers can choose a piece of artwork from a large catalog, and get that artwork printed on a product, like a t-shirt or a phone case. We would like to provide personalized recommendations of artwork that we think people might like enough to order.

The base experience allows users to search for keywords, and then we show them artwork which matches those keywords. They can view the artwork, and if they decide they like it enough, they can buy it.

From this experience, we might get the following data:

  • User A searched for “cats”
  • User A viewed artwork #123
  • User A viewed artwork #234
  • User A viewed artwork #345
  • User A purchased artwork #345
  • User B searched for “cats”
  • User B viewed artwork #345
  • User B searched for “kitten”
  • User B viewed artwork #456
  • User B purchased artwork #456

This is just data though. Here are some insights we might be able to draw about user A:

  • She probably likes cats (high probability)
  • She might also like kittens, because another user who searched for cats also searched for kittens (medium probability)
  • She might like artwork #456, because another user who looked at artwork #345 ended up buying artwork #456 (medium probability)
  • She may prefer cartoons of cats to photos of cats, because the three artworks she viewed all had the “cartoon” tag as well, and she didn’t view any of the cats in the first page of search results that had the “photo” tag (low to medium probability)

This could then be build into experiences:

  • When she visits the homepage, show some cartoon cat related products above the fold.
  • When this user searches for cats next time, blend in one or two search results from kittens as well, to see if she responds positively. Also, give a search ranking boost to artworks which are also tagged ‘cartoon’.
  • When sending an email with a promotion, customize the products included in the email to include artwork #456.

Then, of course, loop around. Watch to see if the data from the tweaks you made are producing favorable results, adjust your insights accordingly, etc, etc.

)

    Erica Smith

    Written by

    CTO and obsessive maker of things.