Recommendation driven UI: Startup Challenges

This week our team at Syfto introduced its new UI (www.syfto.com). This UI is essentially an experiment in finding an effective way to present recommendations to the user. Ideally, we would like to present the user with the most relevant content while requiring them to do minimal work. The challenge of finding the right balance between ‘work’ and ‘discovery’ is central to defining the right user experience.

My primary expectations of experience as a user are:

  • Find something I like
  • Browse content that interests me
  • IF I return, I will find more content I like to browse
  • The site helped me wade through a lot of stuff easily

Secondary:

  • I will buy something I found.
  • I will recommend this site/app to a few friends.
  • I want to see what my friends like.
  • I want to see what everyone liked.

The UI challenge is to balance content presentation as well as feature-discovery by the user. A bigger challenge is to quickly take the user from their first-interaction with content to presentation of relevance-filtered content (recommended content). As we begin to think about a UI that presents the quickest path from initial contact with the user to discovery of relevant content, there are a few things we can learn:

Learning from users

There are a number of possibly ways to determine and present recommendations to users. A few we’re working on include:

  • User profile and their social presence (Facebook)
  • User interaction with content on the site (browsing, views, vote up/down,…)
  • Implicit affinities/relationships within content including:
    • category
    • brand
    • designer
    • price/discount
    • natural aggregation by function (a ‘look’, complimentary items in home decor etc)

How

There are a few conventional ways in which user participation with content creates data:

  • Ask the user
  • Data mining user’s interaction with content on site
  • Leverage other data sets

The UI challenge

In order to derive ‘user-style’ or user-DNA from data, there are a few choices though not many:

1. Ask the user a lot of questions – at signup/onboarding as well on an ongoing basis. There are a few sites that do this well but none seems to find the right balance between annoying and helpful.

2. Make users interact with a lot of content. This is easier said than done since overwhelming the user with content is exactly what we’re trying to avoid in the first place.

So neither path seems to be a great way to do it. Over time, a few sites have done this well are:

The Startup Dilemma

For a startup working off a seed financing, not a lot of data exists a priori or can be created in the time permitted by first financing or within a pivot experiment. While social data sets via Facebook (or Twitter) help to some extent, they are typically sparse in order to compute recommendations solely based on that data/mining. Other sources of good data include ‘trending’ content on the site and social proof (what your friends liked on the site).

The challenge to present this information to the users such that we enable guided-discovery is a complex one. The inherent paradox of producing useful recommendations/suggestions to the users while not asking them to do ‘work’ is the one we have to solve. Our current UI stays within conventional representations of ‘recommended items’. Over the next few weeks we will be introducing new ways for users to discover what we find and recommend to them.

If you think you can help us navigate this challenge better, we’d love to hear from you. @rohit_x_ or @syfto or via email (firstname @ syfto.com)