When I gave my talk about mobile performance and business KPIs at Velocity Berlin a couple weeks back, one of the areas I got the most questions about later was the experiments we were able to do in which we delayed HTML on a customer’s site and tracked the results over a 12-week period. I thought it might be useful to break some of this out into its own post.
As I mentioned at Velocity, I was insanely jealous of Google and Bing a couple years ago, when they revealed their own in-house experiments with HTML delay. Most of us in the performance community would kill for that kind of experimentation platform. So I was extremely happy and grateful when one of our customers at Strangeloop expressed an interest in figuring out the value of time for their business.
We conducted a split test over the course of 12 weeks, in which we segmented mobile traffic into four groups: fully optimized, 200 ms delay, 500 ms delay, and 1000 ms delay. We monitored four metrics: bounce rate, conversion rate, cart size, and page views. We also monitored and analyzed user behavior for 6 weeks after the test ended, to gauge the long-term impact, if any, of slow performance even after users begin to receive an accelerated site.
The results of the 200 ms delay weren’t significant, but the customer and I were both taken by the dramatic impact of the 500 ms and 1000 ms slowdown. Our customer was blown away that they were losing 3.5% percent of their conversions when their site was delayed by just one second on a mobile device. This was a major epiphany for them, and it’s already helped them change their business and how they view mobile.
I’m including this next graph to reinforce the connection between load time and user behavior.
Over the 12 weeks, you can see that, while the HTML delays were constant — 500 ms and 1000 ms — users’ reaction to these delays fluctuated. The drop in bounce rate ranged from around -2% and -12% for users who experienced a 1000 ms delay, and 0% and -6% for those who experienced a 500 ms delay. While the bounce rate may have varied, the one thing that was constant was the fact that the behavior trends are strongly linear for both groups, and the bounce rate for the 1000 ms group was consistently worse.
Finally, and most interestingly to me, we wanted to look beyond just the effect of delay in the timeframe of the experiment. We know that slower pages have an immediate impact on user behavior and customer satisfaction. We wanted to find out if there was any long-term impact on customer satisfaction.
So we looked at our traffic data for the 6 weeks immediately after the experiment — specifically at the behavior of returning visitors. As any site owner will tell you, repeat customers are the bread and butter of an e-commerce vendor. These are the people you need to keep happy. If you look at the graph below, you can see that, even after the experiment was over and the shoppers in the 500 ms and 1000 ms group started to be served the same accelerated site as the baseline group, they were significantly less likely to return to the site. By the end of the 6-week period, you can see that return traffic is slowly improving as visitors seem to finally be recovering from their poor experience.
As I’ve already mentioned, these findings have been a huge revelation to the company that owns this site. It’s had a major impact on how they’re tackling performance on their mobile platform. The most important over-arching takeaways from this experiment were:
- Mobile shoppers are now fully engaging with e-commerce sites in significant enough numbers that we can analyze their behavior as a group.
- Even a 500 ms delay has a major impact on metrics. For mobile sites, which can suffer excruciating load times (the latest Keynote index is about 12 seconds), this is a wake-up call that they need to take a hard-line approach to optimizing their pages.
- The damage of poor performance is lasting.
See the full presentation: Case Studies from the Mobile Frontier: The Relationship Between Faster Mobile Sites and Business KPIs