When recommendation algorithms don't quite hit the mark: 4 quick questions to keep in mind

As I was prepping for this week’s blog post, I noticed that WordPress recommended that I purchase another domain name. I laughed since it seems the automated recommendation algorithm came up with the combination of terms of Batman and Robin and thought I’d be interested in purchasing BatmanYap.com for $18/year. First off, GoDaddy is cheaper and secondly, well, you guessed it, the recommendation was inaccurate. Robin can also be attached to Puck, Williams, or Hood. I didn’t get those choices. Maybe because they already exist?

So here are my 4 quick questions that you would want to address to recover when your automated response system fails.

1. Do you have a process in place to check that the recommendations are correct, in the first place? Maybe ensure that your social network team are customers themselves getting recommendations and are actively interacting with the system. Or have a cross-section of the company as part of a user interface focus group? I’ve engaged members of communications, marketing, human resources, technology, at one point in my career when I have programs that needed another dozen sets of eyes to ensure the program is impacting stakeholders positively.

2.  How to address potential mishaps? Is it by sending a personalized tweet? An online coupon? Some other call-to-action marketing campaign that kicks in automatically?

3.  Do you have a way for customers to contact you so they can get improved recommendations in the future? Is there a link somewhere in the campaign that connects the customer to an “improve your experience” page by putting the onus of the recommendations back to the client?

4. Have you incorporated another recovery process to address the original challenge? An example of this last question is from this tax bulletin boo boo.

Finally, should I get batmanyap.com? Maybe not.

Want to learn more about recomendation algorithms? Check this research paper: Recommender systems: from algorithms to user experience

1 comment to When recommendation algorithms don’t quite hit the mark: 4 quick questions to keep in mind