A few weeks ago, one of my editors pointed out that Gawker’s traffic graphs on Quantcast looked less ‘spike-y’ and asked me if I had noticed. I pulled up a graph of our traffic over the last 2+ years and immediately saw what she was referring to.
It looks like our week-to-week traffic volatility started coming down over the last year: we aren’t seeing the same peaks (or, to a lesser extent, valleys) that we saw in 2014.
This might seem like a minor observation about our traffic, but it isn’t: over the last few years, online media has been a hits-based business dependent on heavy traffic from major aggregators, primarily Facebook. The spikes in our (and other sites’) traffic were primarily driven by stories that ‘hit’ on the social platform.
This traffic pattern incentivized sites that make money from display advertising to go after the same stories that were likely to go viral. John Herrman hilariously tracked The John Oliver Sweepstakes, but that’s just the most glaring example. Gawker itself had a public experiment with traffic-whoring.
There has been a lot of hand-wringing about the loss of Facebook traffic, much of which is based in truth, but overblown. But if Facebook traffic, and more importantly the hit-driven dynamic that it facilitates, is now less important, that potentially has implications for our business.
Is it just us?
My immediate concern was whether we are the only ones seeing this decrease in volatility. If so, it might be because we’re just losing these traffic spikes to other publishers.
To figure out if it’s just us, I took a look at the traffic volatility* of 21 large digital publications: 6 of ours (I excluded Gizmodo because of the io9 merger late last year) and 15 others (mostly the media companies on Quantcast’s top-100 list with public data).
The results were pretty clear: 13 of the 15 other sites had seen a decrease in traffic volatility over the last two years, just like we have.
Week-to-week traffic changes peaked at about 20% in early 2015. We’ll remember this as the top of the hype about virality. This was the fertile ground that the dress post appeared in the midst of.
Those days appear to be ending: traffic volatility is down by about 1/3rd since that time**. For almost digital publishers, this week’s traffic is going to look a lot like last week’s traffic.
What does it mean for us?
So if traffic is becoming more stable, what does that mean for us? I think it’s a mixed bag. On one hand, massive traffic growth, like we saw a few years back, probably isn’t in the cards.
On the other hand, the smoothing out of traffic numbers might be a sign that publishers who have reached scale are less likely to lose that scale. Maybe the land rush for traffic and social attention is over? Anecdotally, it’s much harder to grow Facebook fans and Twitter followers today than it was a few years back. Our existing scale would serve as a difficult-to-reproduce moat.
More stable traffic also allows us to monetize the traffic we do get more efficiently. Our business team can forecast inventory and size ad sales better, which should mean better fill rates and happier clients.
It’s also possible that the fight for attention has just moved from websites onto individual platforms like Facebook Live Videos and Snapchat Stories, which aren’t included in these numbers. We don’t have a sense of what is ‘normal’ for these new platform distribution points yet. We know other media sites are focused on them, as we are. I’m hoping to write more about our experience on these platforms in the coming days.
* I looked at unique visitors, week-over-week, to remove the weekly seasonality of traffic, then took the mean change over the trailing 13 weeks (~3 months). A true finance equivalent would have looked at the second-order volatility, but I think this is more tractable.
I only used US traffic since A) there are fewer chances that there are reporting anomalies due to international partnerships, and B) the vast majority of the economics come from domestic advertising.
** The chart uses mean, but taking the median gives the same result. If I use raw traffic swings instead of percentages, the change is closer to 50%.