What Makes for “Bad Data” for Personalization?

We’ve all heard the expression: “Garbage in, garbage out.” It means that the quality of the input to a process will determine the quality of the output. If your inputs are bad, then your output will be too. The adage certainly applies to personalization — any occasion when you decide what experience to deliver to an individual based on something you have learned about him — because even the most advanced machine-learning algorithm designed to provide a highly individualized experience won’t be successful without a deep understanding of each individual. That understanding comes from data.

Think about it this way: if you’re a sales associate in a retail store, you probably want to provide a personalized experience to every customer. But if you’re trying to help a shopper find what she’s looking for or to recommend additional products she may be interested in, your help will be completely ineffective if you don’t know anything about her. You need to have a conversation with her before you can offer any valuable assistance. In the digital world, the data you collect acts as that conversation to help you form an accurate representation of each person.

In the context of personalization, “bad data” can have several different meanings. Data is bad or ineffective when it’s incorrect, when it becomes outdated, and/or when it’s inadequate. Let’s dive into each of these areas in this blog post.

Incorrect Data

It’s easy to understand why using incorrect data would lead to inappropriate or inaccurate personalized experiences: if an input is wrong, the output will be wrong too. For example, you could use data from your CRM to deliver a message within your web app only to customers who are using the product for a particular use case. But if that data was entered incorrectly into your CRM for certain customers, the resulting experience will be targeted incorrectly and won’t make sense.

This seems obvious — no one wants to have incorrect data in their organization! But there are often situations when you don’t trust a data source that you have available in your organization. A 2017 study of senior executives from North American, Britain, and Ireland by Dun & Bradstreet and Forbes Insights revealed that only 42% are confident in the quality of their organization’s data. And if you’re not confident in the data you have, you won’t be confident in the personalized experiences delivered based on that data.

Can you rely on your data sources? Consider weighing the cost of using an inaccurate source against the benefit of the information it gives you.

Outdated Data

Just because something was true a day or even a minute ago doesn’t mean that it’s still true, so even data that was once accurate can become outdated sooner or later. Technology allows us to move very quickly, and personalization based on data that doesn’t keep up won’t be accurate.

For example, if a visitor to a travel site researches and books a ski vacation all in one session, it won’t make sense to recommend additional ski vacation packages to him when he’s on the site again later that day (although it may still make sense to recommend ski-related content to him). While it is true that those packages were relevant to him recently, they are not still relevant to him.

Can your data keep up? The only way it can is if your personalization platform is able to take in information and then act on that information in real time. While the term “real-time” gets used a lot, true real-time actually means “in the moment” with a response time of less than 20 milliseconds (faster than the blink of an eye). Anything longer could result in ineffective, next-time personalization.

Inadequate Data

Although it might not be obvious to the topic of “bad data,” not having enough data can produce less-than-accurate personalized experiences too. We found in our annual study that the most common criteria marketers use to target personalization campaigns are campaign source, location and demographics.

Those areas are a good start, but they don’t tell an individual’s full story. What pages has she viewed on your site? Was she interested in what she found on those pages, or not interested at all? If it’s a retail site, is she a loyalty program member? What are her favorite categories and brands? If it’s a B2B site, at which stage of the journey is she? What topics is she interested in? If it’s a financial services site, is she an existing customer? What products does she use? If you only have a few data points to paint a portrait of your visitor, you’ll only create a hasty sketch.

Collecting in-depth behavioral data (not just the pages that someone clicked on but how she engaged with those pages and what those behaviors say about her preferences and intent) and combining that information with data from other sources in a central location is absolutely critical for effective personalization. Because while you can deliver a message targeted to someone knowing only her location or campaign source, her whole experience won’t be as relevant as it could be if you knew more about her.

Final Thoughts

Data is a critical component of any successful personalization program. Clearly, if you have incorrect data, outdated data, or inadequate data, your personalized experiences will be affected. And realistically, a personalized experience based on bad data isn’t personalized at all. It’s an endeavor that misses the mark.