Personalization Pitfalls and How to Fix Them

Personalization—applied properly—can turn a regular website visit into a serendipitous experience. Think of how you felt the first time you heard your favorite song. If you have good content, then you probably have pages on your site that will give your visitors that same feeling.


The best way to help them discover that content is by personalizing your site to show the right page to each visitor. However, personalization is still a new practice and sites are making a lot of mistakes while we’re all in the trial and error phase of figuring it out.

We’ve learned some lessons about what can go wrong and how to fix it. Here are a few.

1) Use the right type of data

The first common personalization pitfall is using the wrong type of data to personalize website content. Facebook has perhaps the web’s largest store of intent data. Facebook knows what you browse, who you talk to, and most obviously, what you “Like”. In fact, many sites will pull in your Likes when you sign in to them using Facebook. Liking, however, isn’t always the best source of intent. On Facebook, I may “like” Ferrari but in real life I drive a Jeep, and I’m not going to buy a private jet charter if you advertise one to me.

Pandora has another interesting mix of declared interests and observed behavioral data. There are lots of online music sources these days, and Pandora’s core value proposition is really one of an easy-to-manage discovery engine. They use my thumbs ups and thumbs downs to fine tune their recommendations for me. But I’ve found that once I get a good station, I’m afraid to mess it up so I stop clicking, it goes stale, and then I stop listening altogether. I wish they took my skip button clicks (a.k.a. behavioral data) as a stronger signal to play fresh music.


When in doubt, remember that actions speak louder than words and that behavioral data will best help you understand what your users really want.

2) Avoid boxing

I’d love it if news sites gave me more tech and business content up front every time I visited, but this isn’t always what’s best for me. Sometimes I really should hear about important world or political news, or about weather that could affect my plans. If all sites automatically narrowed their content down to their understanding of my interests, it could be severely limiting or worse, discriminatory. This is called boxing.

“Boxing is where a consumer’s vision and choices are limited by . . . analytics that make judgments based on their digital history.” –– Martin Abrams, Boxing and Concepts of Harm

In practice, most of us already box ourselves. Most news sources have some sort of bias, and I limit my vision and choice when I only visit the ones that I agree with. The important distinction is that I can quit whenever I want because machines aren’t making those choices for me.

Here’s another example. I visited a major retailer’s website to look at a frying pan and here’s what they recommended to me:


Hopefully this is just a mis-labeled recommendation unit because I find it hard to believe that the top five products people also bought were basically identical frying pans. This site is trying to key off my intent, and reinforce it with additional purchase options. However by thinking so narrowly they’re missing out on an opportunity to drive higher sales.

It’s ok to use a couple of cells to show me other frying pans, but why not use cells to show related products like utensils that work well on cast iron? You know I searched so you can probably expect that I’ll keep looking until I find what I want. Why not take a gamble and show me an unrelated but possibly appealing product to cast iron buyers like a french press or a water filter? Amazon does this well with their “frequently bought together feature.” If you do personalization right, you can get me to buy three things at once, rather than making triple sure that I buy a product I was going to buy anyway.

The best antidote to boxing is to think “discovery.” Help me find new things I didn’t know existed, and I’ll keep coming back for more.

3) Don’t surprise your users

Users are used to getting personalized recommendations these days, and many would say that it improves their experience on your site. This can be great in context if it recommends the right products or content. Take this, for example:


Taken out of context, however, it can be pretty surprising and many users are creeped out the first time they realize they’re being retargeted by online advertising. The ad creative below is essentially showing the same things from the same retailer:


Now, if you were on a different website and didn’t realize that was possible, you might be wondering how that site knows that you were shopping for shoes somewhere else. It gets worse when you share a computer with another person. Users may eventually learn that this is powered by a lot of ad tech behind the scenes, and that the new website they’re on usually doesn’t know their shopping history. That doesn’t mitigate the negative first experience they’ll associate with your brand.

Users know how their offline identity works. They expect waiters to adjust their meal recommendations when they visit a restaurant with small children because they’re used to how much personal data is visible when they leave the house. So the key to avoid creeping out your visitors is to leverage patterns they’re familiar with and avoid surprising them, at least until they get used to the personalized web.

Base your applications of data on your users’ sensitivities and be subtle when they’re not expecting it. Use clear visual cues to let them know the context, but also let them know when they’re anonymous and when they’re not. If you’re using personally identifiable data, make it clear that you’re doing so with indicators of their identity, like their avatar, and give users an easy way to turn it off. Try to leverage real world identity paradigms, especially when using data in a 3rd party context.

Keep these concepts in mind when designing your website personalization, and make sure to do lots of user testing. As an insider, you’re the furthest from the context of a new user, so make sure to talk to lots of them to be sure you’re improving their experience on your site.

Editor’s note: this is Charlie’s fourth post in a series on personalization. You can read the other posts here, here, and here.