I had a discussion recently about the future of analytics, specifically about what new analytics trends that I'm geeking out about, and I wanted to share it with you.
We are living in the most exciting time of all, because the world of analytics is going through a tremendous transformation. The old form of analytics is being replaced with a number of new ways of thinking about data. Ways that are far more in tune with how we humans behave.
But let me briefly summarize the four different types of analytics that I'm fascinated by at the moment.
The first one is 'scored analytics'.
Scored analytics is a form of analytics where you try to measure the importance of an article (or similar things) by looking at the value of the interactions.
For instance, having a person actually read an article is much more valuable than just having someone view an article. But it's not just what people do on the page that it's important, it's also how people get to the page, and what people do afterwards.
For instance, a visitor coming to an article from your newsletter is generally more valuable than just a random view via Facebook. Having people share an article might indicate another form of value (although not always the type you think). And having people return is another signal as well.
The problem is that, with normal analytics, all of these metrics are presented as single data points with none of them making that big of a difference.
This is where scored analytics comes in. With scored analytics, you attribute a value to each type of interaction, and then you add them all up. So, one article might have a total score of 290, while another one might have a score of 470. Meaning the second article was more valuable not because of any single metric, but because of the combined value of all the interactions combined.
This is a very interesting way of thinking about analytics, especially for publishers.
The second type I'm fascinated about is 'behavioral analytics'. Behavioral analytics focuses on measuring how people behave, in order to give us a much better idea of the value of each interaction.
Let me give you a simple example.
If you are using Chartbeat, it will help you measure how much attention a page gets, but it doesn't tell you whether that was a useful form of attention or not. This is where behavioral analytics comes in.
For instance, imagine you want to measure if people read a page. The way this is usually done is like this
Here we check if people start scrolling down the page, and then we also check when they reach the end. And the way we then determine if people actually read the article is by looking at how long this takes.
If an article takes 7 minutes to read, but people scroll from the top to bottom in only 15 seconds, then they didn't really read it.
This works fine for simple stuff, but it's not very accurate.
A better model would be to look at how people scroll down a page. Like this:
Here we observe that people started scrolling and then stopped when the next part of the article was in view. Then they scroll again, and stopped to read even more.
This is a much better way to measure this, because now we are observing actual reading.
The problem is that when you start to measure it like this, you very quickly realize that people don't necessarily behave this way.
For instance, with my Plus articles (which are usually about 30 pages long), people often start to read them, leave, but then return later to finish them. So, I might see a pattern such as this:
So, behavioral analytics is much more complex (and harder to implement), but also incredibly fascinating because it can help you identify patterns that normal analytics completely miss.
And, of course, this doesn't just apply to measuring read-rates. Think about more advanced stuff, like measuring the impact of engagement over time in relation to getting people to subscribe.
There are so many things that we just don't see if we only measure the overall interactions (like views). We need to get into the details and observe how people interact as well.
Thirdly, we have the very exciting world of what I call 'learning analytics'. Learning analytics is what Google and Facebook is using, and something I have written about before. Learning analytics approaches analytics in a very different way, in that, instead measuring pageviews, etc, it focused on building up profiles around people (mostly).
Facebook is a good example of this. The way Facebook measures things is to look at what you are doing as a person. It measures what you look at, what you engage with, what you click on, what connections you create and many other things. But instead of measuring this as 'person [x] liked 287 posts last month', Facebook uses this data to build up a profile about you.
In other words, they are using analytics to 'learn' who you are ... hence why I call this 'learning analytics'.
In many cases, learning analytics is a much more powerful way to measure things, because what articles people see (or even the number of articles) doesn't really mean much.
Again Facebook is a good example.
Facebook doesn't care about the performance of any individual post, because people see 100s of post each day. But they care a lot about who you are, what you are interested in and what you are likely to engage with. So, their analytic focus is to learn who you are instead.
So learning analytics is a really interesting to think about data.
The fourth a final new analytics trend that I'm geeking out over is the future of machine learning and how that will impact the way we do analytics. I wrote about this in my latest plus article.
Machine learning is fascinating because it's defined around measuring patterns and probabilities rather than individual data points. And from an analytics perspective, this is important because often the way people behave is so complex that you cannot look at any single thing, nor even do things like scored analytics.
Machine learning has the potential to redefine the way we think about analytics because of how it is turning this whole concept on its head.
It starts by looking at the outcome, like the total sales or overall subscription performance, and then we teach the machines to look at all the data as a whole in order to figure out why.
This is a fundamentally different way of defining analytics, because it enables us to identify patterns across hundreds of different data signals, each of which might not seem very important to us. In other words, machine learning has the potential to mimic the way we think as humans.
When we make a decision to do something, it's rarely because of a single thing. Instead, it's the result of multiple interactions, over time, that causes us to connect more with one brand than another.
For publishers, this is incredibly important, because people don't subscribe (or connect) to magazines or newspapers because of single metrics. They do it because of the loyalty we build up over time.
This is the potential insight we gain from the future of machine learning, but you can read more about that in my previous article.
These four new types of analytics have the potential to transform our world. Each has different strengths and weaknesses, but all of them provide us with a level of analytics that wasn't available before.
It's still early days for some of this. For instance, behavioral analytics is still in the early implementation phase, and we are just starting to realize the potential with machine learning. But other things, like scored analytics, is already offered by several analytics startups, and learning analytics is in active use by most large tech companies, and even big publishers are starting using it very interesting ways.
The real question, though, is where we will be in five to ten years from now.
Will you still be using traditional analytics, where your dashboard is merely telling you how many visits you had and what devices or countries they came from?
No, of course not.
This is the change that is happening right now. And, at least to me, it's incredibly exciting.
Creating a propensity model is one of the most important tools publishers can have.
Many people say you can't measure trust. But you can, although before you do that, you first have to create trust.
When you are an independent publisher, analytics can sometimes be tricky because we don't enough data to work with.
Several publishers have found that reducing volume leads to an increase in revenue
The potential with machine learning is amazing, but it's not enough to identify a result. We also need to be able to do something about it.
Time is such a critical metric for publishers, but it's also a very complicated one.
When you are monetized by advertising, you tend to favor the least valuable metrics, but when you are focusing on subscriptions that changes to the most valuable metric.
Everyone talks about conversion rates, but that often doesn't tell you anything about how well you are converting people. Let's talk about conversion value.
Many large publishers are now turning to advanced analytics to understand their audiences, but what if you are not a big publisher? Can you still do it?
Publishers who start their own data studios need to take extra steps to identify real people.
Founder, media analyst, author, and publisher. Follow on Twitter
"Thomas Baekdal is one of Scandinavia's most sought-after experts in the digitization of media companies. He has made himself known for his analysis of how digitization has changed the way we consume media."
Swedish business magazine, Resumé