The Problem with Recommendation Engines

I was just reading a blog posts on why brands should build recommendation engines to create better content platforms and improve the overall experience for users. While I tend to agree “pushing” content to users is a better strategy than “pulling” them in to consume content, I don’t think the technology is ready to truly support these type of experiences.

Below is the comment I posted.

I agree with you that brands should invest in building “user platforms.” Though, contrary to what you mention above, these platforms lack context to serve up the right content to the right users.

I like to use Amazon as an example… Amazon thinks I like a Hello Kitty backpack because I purchased it or added it to a wish list. In actually, I purchased it for my niece, so recommending I purchase a Hello Kitty doll is not relevant to me. Tis the issue with recommendation engines – they lack context, which leads to irrelevant recommendations. Amazon will not know who I purchased the product for unless I explicitly tell it my purchase of that backpack is a gift for someone (which is data it does not collect).

Same holds true for Facebook. Facebook thinks I am friends with Jamie (name changed to protect this person’s identity). In actuality, I am only friends with Jamie because she keeps sending me friends requests when I unfriend her. I really don’t like her, nor want to know what is going on in her life. Yet, Facebook tells me when her birthday is and every time she posts a new cat photo.

Sadly, the technology to fully supports these types of engines does not yet exist. These platforms are too reliant on user input to educate the recommendation engine on what the user may like, and those learnings are subjective because they lack the context of why that input was made. These engines could fake it by designing an experience that asks users to supply this information in a frictionless way, but most platforms don’t/won’t. The only platforms I see do this well are some dating websites.

What are your thoughts on the current state of recommendation engines?