Get More Digital Media Updates from Parse.ly

Views, Visitors, Engaged Time & Shares in Context

parse.ly changelog 10-2016

In Parse.ly’s new analytics platform, we provide support for a bevy of new metrics beyond the world’s most popular analytics metric, the venerable page view. Here’s a screenshot from our dashboard that shows all the available sort metrics.

metrics

For customers in our highest tiers of service, we aim to provide all of these metrics both in real-time through minute-by-minute, live-updating screens and over time, as a historical record, rolled up and summarized by day.

New Metrics Galore

Now that these new metrics have launched in the new dashboard, we understandably get many questions about them.

Metrics definitions can vary from platform to platform though, and it’s critical to us that our clients know exactly what each one means. This includes understanding how we are measuring them, defining any latency that affects real-time measurement, and explaining any estimation techniques. The following table dives into the details:

measures_table

I’d like to highlight a couple of areas from this table. First, many of our metrics are sourced from a JavaScript event fired on every page of a publisher’s site. From these events, we can determine page views and traffic sources (aka “referrers”) coming from search engines and social networks. This was our only real data source in the old version of our dashboard. But now, we have unifed several other data sources around your content.

The first big one is Total Visitors, which can also be segmented by New and Returning Visitors. This metric is measured using a “first-party cookie” that uniquely identifies each visitor’s web browser. Cookies are an inherently inaccurate proxy for “people”, but they are the best proxy available to audience managers when measuring anonymous audiences. Parse.ly does not apply any cookie deduplication logic in its visitor counts, but it does not sample your data, either. Instead, to achieve real-time performance, we use probabalistic data structures to estimate your visitor counts, typically within 2-5% accuracy. This is a necessary tradeoff to return visitor counts quickly in the dashboard.

One thing to note about Visitor Counts: It’s very tempting to produce exports of data in Parse.ly using visitor counts, and then attempt to sum up visitor counts on your own, after the fact. For example, if I had my top 3 sections looking like this:

visitor_counts

I might decide to add the visitor counts from the 3 sections together — 154k + 74k + 32k = ~260k visitors. But I’d be making a mistake in doing so. The reason is that the unit of a visitor is, well, a Unique Visitor. You cannot sum Unique Visitors together on your own due to what is known as “the hotel problem”. If a web visitor visits two stories on your website — one in the News section and one in the Sports section — then you only had 1 “website visitor”, but 2 “section visitors”.

So, unlike page views and engaged minutes, visitors cannot be rolled up and summed on your own accurately — you must rely upon Parse.ly’s fast engine to do this work for you! For the geeks out there, this hard work is known as the “count-distinct problem” in computer science circles. Returning real-time visitor counts has been a challenge for many analytics platforms, due to these and other tricky technical issues that plague unique visitor counts.

It’s About Time

Alright, so, Parse.ly knows about your visitors and your views. How about the time they spend with your content?

For measuring time, we introduced the Engaged Minutes metric. Unlike Page Views, which come from a single JavaScript event, this metric is measured via a “JavaScript Heartbeat”. How does that work?

The Parse.ly tracker determines which of your users are currently engaged with your content, and for every individual visitor who is, we produce a “heartbeat” telling Parse.ly’s systems that the user is spending time with your content. This makes it possible to track precise time spent even for single page view sessions (very common on content websites), and also to properly attribute time spent for richer forms of content, such as video, long form, and interactive.

Understanding Traffic Sources

Views, Visitors, Time — now you can really understand your audience beyond the click! But we didn’t stop there. We also decided to break out views into “device views” on mobile, desktop, and tablet. We decided to offer sorting for social referrers (that is, clicks that came from social networks, like Facebook) and by search referrers (that is, clicks that came from search engines, like Google). And for the visitors, we broke them out by new visitors (never been to your site before) vs returning visitors (your core audience).

referrers

Each of these seemed to offer brand new ways of understanding your audience and its interests. For example, many publishers have found relationships between their evergreen content and search referrers, and have unearthed that high mobile traffic is a leading indicator to social virality. Looking at traffic popular with new visitors let them learn about real-time audience growth, and looking at traffic popular with returning visitors let them understand the interests of their loyalists.

One of the newest and least understood parts of our new dashboard is the “shares” metric. This is a metric we generate by connecting directly to the APIs for the 4 largest content-oriented social networks: Facebook, Twitter, LinkedIn, and Pinterest. We aggregate up the share counts from each of these metrics and let you sort your content to find that which is most shared. Note that due to the limitations of these APIs (not a limitation of Parse.ly), this data is not truly real-time — instead it is typically about 5-10 minutes delayed, and can be as delayed as 1 hour at times.

All of this new data is a dramatic change for those used to our old dashboard, where there was only a single metric (page views) throughout! How do we interpret all of this new data?

Here’s that table of metrics again for you to review:

measures_table

Providing Context for Interpretation

One of the design principles that went into our new dashboard is that, since we provide so many more metrics to our users, we should also provide context around those metrics. That is, we should always show a specific sort metric in context of other, relevant metrics. And, when we display a metric on its own, we should aim to also display how that metric compares to historical site-wide averages.

We do this for two reasons. First, we want to help publishers and content producers understand that no single metric should be their ultimate goal (what we call a “golden metric.”) Second, it’s more interesting! Seeing things in context allows you to more easily understand what the metrics mean about the audience, so you can focus on that instead of growing numbers for the sake of big numbers.

As an example, when we display a list in the dashboard sorted by “mobile views”, the metrics we show in context are the percentage of total views that mobile represents, and the comparative number of views for desktop.

For this site, that helps us determine that though News is the section with the highest number of mobile views, this can be explained by the fact that nearly 56% of the section’s views are mobile:

mobile_views

In other words, editors can learn that their breaking news content has a higher propensity to be visited via mobile, so if they have a goal of growing their mobile audience, it is wise to focus on this genre of content. Further, though a smaller absolute number of views come from their Travel section, this is also a genre preferred by mobile visitors.

Another idea — let’s say I’m an author and I want to know the relationship among my visitors, their time spent per post, and the kinds of posts I’m writing? I have two tags available, “type: brief” and “type: report”. The former is applied to 400-word short-form stories, and the latter are 1,500 word+ in-depth articles. (You are using tags, aren’t you?) Are people spending more time on my long-form pieces, as I’d expect?

post_count

Answer: yes! But not by the factor an author might intuitively expect — they’re spending about 50% more time per post, even though the author is writing at least 3x as many words. But there are other insights here, too. What we learn is that this author’s long-form reports drew a bigger audience, and that this audience spent an average of 2 minutes reading the long-form content. The briefs produced a smaller audience, and there was also less time spent per article, as well. There were 15 posts in each of these tags for a 30-day period, so the author is writing an even number of each. Based on this data, the author could be encouraged to continue the writing style of their “reports” to continue winning on both audience and engagement.

How does the author compare to the site overall? By dropping the author filter, we learn that “reports” typically garner ~3 minutes per visitor, across 217 other posts:

sitewide_reports

That’s more than a minute greater than John’s posts! What gives? Answers are only a click away in the Parse.ly dashboard.

So, that’s the power of context. What good are metrics if you don’t have the context to interpret them and draw conclusions that can impact your content strategy?

Going Viral

One of our team’s favorite contextual metrics is “social referrers per share”. This one warrants a little explanation. If you sort all of your content by “Shares” in the Parse.ly dashboard, we’ll show you “social refs” and “refs/share” as contextual metrics. Let’s take a look at the following 3 posts:

shares

How am I to interpret this data? Well, let’s say I’m the Facebook page manager or social media editor for this publisher. I want to choose one of these 3 posts to promote on our page. Which one do I pick?

Answer: either #1 or #3. Why not #2?

The answer is informed by refs/share. Although this post has the second-highest number of shares on the site for a 7 day period (with an eye-popping 8,000+ shares in social networks), it is only earning 0.9 clicks per share. Meanwhile, the story at the top of the leaderboard — on Elon Musk’s Hyperloop — is earning 5.3 clicks per share. Not only is it more widely shared, but it will also draw much more traffic when shared widely. If 1,000 new people share this content, you’ll likely earn about 5,300 new page views. But if 1,000 new people share the flamethrower piece, you’ll only earn 900 new page views. Quite a difference that a small promotion choice can make!

Go Forth and Explore!

Back in February, my colleague Toms described some of the UX decisions that went into the dashboard, and wrote the following, which has become a bit of a mantra on the Parse.ly team: “the challenge now lies in finding how to tell the million stories each dimension is able to provide, without overwhelming users with data.”

dashboard_monitor

We hope that you’ll find this edition of The Changelog serves as a solid introduction to all of these new data points, and helps you understand how to interpret your own data in our new dashboard. Happy exploring!

About The Changelog

The Changelog posts document changes to the Parse.ly dashboard through the eyes of co-founder and CTO, Andrew Montalenti. Interested in trying out the changes for yourself? Login to your Parse.ly account, or sign up to get started with Parse.ly today.

Find out how your team can access insights about readers on your site with Parse.ly’s dashboard.
See how other businesses have used the Parse.ly platform to access insights about their readers.
Webinar:
Audience Engagement:
What We Know, and
What We Don't
Provide your team with easy access to video analytics using Parse.ly’s dashboard.