I was reading a news item recently about another startup in the marketing analytics space. This startup analyzes eye movements as customers browse in a store and, based on that, tries to determine the customer's interest in certain items, which triggers email offers to their mobile phones in real time.

Wow. I hope that company lists customers on their website so I know where not to shop.

Email promotions that personalize offers based on the interests of each individual customer will increase response rates. OK, but how do you know what offers will be most interesting to each customer? That's not so easy.

If you have thousands of customers and hundreds of products or more, advanced analytics is the only effective way. But the sources of information used by analytics platforms vary widely.

Some build profiles of each customer using all available information. Some analyze website interactions and clicks. Some decipher posts on social media. Some look at customer activity, location or even eye movements.

They are all designed to give marketers "insights" about who customers are, what they want, and how or when to email them.

All of this information can be useful if given enough time and money to source and integrate the data continuously, analyze and interpret the data, figure out how to apply it to email marketing programs, and test it all until it proves effective. The conventional wisdom of most database experts, data science and marketers that usually have to collaborate on this sort of thing is that the more data you can analyze about each customer, the better.

This is fueled, of course, by the ever-increasing amount of data that can be collected. It won't be long before there is temptation to use data that some customers might consider "sensitive" in ways that are "creepy."

Watching a customer's every move across websites or social media might lead to insights, but if you then start to anticipate their needs and interrupt them with messages strangely coincidental to personal problems or current activities, that sort of creepy surveillance could backfire. And if you misinterpret all the "insights" by sending irrelevant email messages, you may find customers annoyed, if not hostile to your company.

I don't know about you, but I don't want an email for each product I happened to look at in a store.

While better insights are possible from analyzing more information, in most cases it is not actually necessary to make good predictions about customers. Depending on what you want to know, most of the predictive power can be derived from just a few bits of information as long as they are the right bits.

For example, in the case of predicting future purchase behavior, analyzing past purchase behavior (i.e., transaction data) of every customer across a large body of customers usually yields better predictions of any given customer's future purchase behavior than just about any other kind of information — eye movements included.

And there's nothing creepy about analyzing transaction data for marketing purposes. That's been going on for something like 100 years. It's not controversial, your company already owns it, and it's probably among the cleanest, most reliable data available about your customers.

Turns out, you don't need all that other data to get highly reliable predictions (probabilities and propensities) about customer loyalty or what products they are interested to buy. And specific buying predictions are the kind of insights marketers can actually use to personalize their email marketing campaigns.

Much can be gained by combining all sorts of information to build rich profiles of your customers. But considering how hard it is to pull off, sometimes a much simpler approach is far more practical and can be just as effective.