What do you do when your data tell you nothing?

Posted on Monday, 3. July 2017 in category Web Analytics. 4 min read • Written by

Črt Podlogar

We love to say that we live in “a data-driven world”. Digital advertisers have hyped growth hacking, conversion optimization, data analysis, clustering, machine learning, and other buzz words to such an extent that most people believe you can learn just about anything if you stare at data long enough.

These days, a belief has emerged that we can look at a set of data and, by applying some statistics, we can understand who our users are, what persona or segment they belong to, what kind of shopping behavior they engage in, and how we can effectively convert them.

This may be possible – but only if you have a massive amount of data, an extremely capable data science department, and AI in place that spells doom for mankind. Otherwise, you can probably expect the same experience I have had in the past few months.

You probably don’t have the data

The first thing we need to clarify is that, on average, the data available on your website is not enough. If you think you’re doing data science and the only thing you have to work with is non-ecommerce such as the Google Analytics setup, you are light years away from being able to perform something that could remotely be called data science.

Even when you have a more complex setup and collect a lot of data, you will run into the persistent problem of identifying users across different devices, touch points (website, app, ads) and trying to wrestle with the fact that users often do multiple things at the same time, distorting your bird’s eye perspective.

Nonetheless, I made an attempt to understand user behavior equipped with a set of 30+ custom Google Analytics events, goals, custom content and enhanced ecommerce data for a webstore. My goal: to identify users who demonstrate purchase behavior, and to separate them from users who are merely browsing the website or looking for how-to content. Having done that, I wanted to build an advertising model that could identify users based on their key behavioral patterns and adjust remarketing.

Three months and zero insights later…

After countless analyses, a lot of charts, graphs and writing on whiteboards to make me feel like I was doing something useful, I ended up exactly where I had started with the project. I analyzed many different user behavior patterns, divided users into content readers, buyers, category browsers, discount hunters and more. And the one piece of information that kept jumping out at me was: everybody was a little bit of everything.

Buyers who checked out discounted products ended up buying pricy products, and vice versa. Content readers sometimes bought things, and buyers sometimes read things. Everybody was browsing categories like crazy, even if they ended up buying something entirely different. Those who read multiple how-to-buy guides without ultimately purchasing a single thing educated themselves on how to buy. They had different genders, age groups and locations, came from different sources, and used various devices at different hours of the day and on different days of the week.

My biggest problem – homepage abandoners! People who turned up on the website en masse from all possible sources and did absolutely nothing. Sometimes they hung around for a while, apparently looking at some banners and random product listings, and then vanished just as quickly as they had appeared.

Don’t get me wrong. I had a lot of data. I had considerable information about users and what they do. But I had zero insights. Despite the scope and depth of my analysis, at the end of the day it was useless.

Eureka! – What if I were to ask users who they are?

I decided to drastically change strategy. If I couldn’t segment users by observing their behavior, I’d ask them to segment themselves, and try to see if there are any patterns that are unique to each segment and don’t overlap in every step.

We made a big change to the homepage and did away with all the classic ecommerce “nonsense” such as banners, top offers, categories, featured blogs and so on.

We replaced this with three large buttons and the simple title: “What is the nature of your visit to our webstore?”

Button A: Just browsing

Button B: Looking for information

Button C: Looking to buy something specific

As soon as the users started choosing these options, the dividing lines between the three segments began to appear. I realized after collecting a sufficiently large set of data that it was stupid to assume users would demonstrate one type of behavior or another. The answer did not lie in what they did; instead, it lay in how much of each thing they did.

With users segmenting themselves, we managed to define a set of criteria that allocated users into one of the three buckets, even if they did not use the buttons.

Leverage user insights to kick-start your understanding

Another significant issue we faced was identifying which product category or product users are interested in. Despite possessing a large set of ecommerce data, including product listing impressions, product page impressions, cart behavior divided by products, brands, and categories, we couldn’t find any obvious pattern.

We applied the same strategy – we asked users which category they were interested in. Their selection was stored inside a cookie on their device and each time they came back we asked if their interest had changed, giving them an option to select a different category.

Again the data came pouring in – we could identify the behavior users demonstrate when interested in buying a product from a specific category and, moreover, we could track when users had shifted their interest from one category to another. This allowed us to track shifts in behavior. We could understand and analyze how users behave when they’re no longer interested in one category and are on the hunt for another.

This could probably also be done without the user, but it was infinitely simpler this way. And predicting patterns of user behavior was much more reliable when it was modeled on actual users belonging to the exact segment we were trying to understand.

My advice to you – don’t be afraid to ask questions! Questions and answers are a fundamental exchange and create the basis for understanding. And this is something that has not changed in our age of data.

Good luck!

 

 

Still here?
Question? Write it in the comment bellow, let's open a debate.

Leave a Reply