Does FBX produce different time-of-day conversion patterns?

 

Today’s post is by Beth Logan, PhD, Director of Optimization at DataXu.

Despite recent reports that users sometimes “take a break” from Facebook, the site is phenomenally popular. It is a highly engaging place where many users spend a disproportionately large amount of time, both viewing and creating content. Therefore, when we started running campaigns on Facebook Exchange (FBX), we were keen to investigate whether conversions followed new time-of-day patterns there.

A Triggit post last year mentions that users convert “all day” on Facebook vs. post-work hours for other platforms, based on an analysis of click-through conversions. Our analysis, using the more popular view-through conversion model, shows that the patterns are not so different after all. We did find, though, that proportionally more early-morning (post-midnight) impressions led to conversions on FBX.

The study

We studied four campaigns running on both FBX and other exchanges. Two campaigns were in the financial vertical and two were retail campaigns. We pulled a week’s worth of data and looked at the local time that users saw impressions on FBX and other exchanges, and whether those impressions led to view-through conversions. We normalized the action through rate (ATR) and fitted a smoothed curve to the plots, which appear below.

conversion pattern FBX

Normalized ATR vs. Hour of day the impression was shown for 4 campaigns running on both FBX and other exchanges

From the graphs, we can see that three of the four campaigns, both financial campaigns and the retail 1 campaign, have very similar conversion patterns for FBX and the other exchanges. The only major apparent difference is a higher tendency for early-morning impressions on FBX to lead to a conversion.

The Retail2 campaign also over-indexes for early-morning (and late-night) impressions on FBX leading to a conversion, but much more markedly so. Additionally, for this campaign, proportionally fewer impressions seen in daytime hours on FBX led to conversions vs. other exchanges. This may be because the product advertised appeals to students, who are heavy users of Facebook and who may keep irregular hours.

Conclusion

From these few examples, it seems that the conversion behavior of users may depend more on the product than the exchange. The only clear trend here is that late-night/early-morning impressions have a higher chance of success on FBX compared with other exchanges.

Though we are still in the early days of FBX and it would be unwise to draw too many conclusions at this point, these findings do seem to support a non-channel-centric approach, indicating that advertisers are better off optimizing to each specific channel. As more advertisers take advantage of FBX, a clearer pattern will emerge of how users engage with advertising on this fascinating site.

beth logan dataBeth Logan holds a PhD in Speech Recognition from the University of Cambridge, UK. She has worked in many fields including speech, music indexing, computational biology, medical informatics, and activity monitoring, and has over 30 publications and 11 issued patents. Since joining DataXu in 2009 she has turned her talents to online advertising, relishing the vast amounts of data available and the many interesting problems to be solved.

This entry was posted in FBX, Most Read. Bookmark the permalink.

One Response to Does FBX produce different time-of-day conversion patterns?

  1. Prachi M says:

    [Disclaimer: I work at Triggit]

    Beth, great to see more and more people blog about the analysis they found on FBX and other display advertising channels. However, a few things still don’t stand in your argument about why FBX and traditional exchange retargeting patterns don’t vary very much. When comparing notes a few questions popped up, like:

    1) Did you aggregate the hours for the entire week? We definitely found different conversion patterns by day of week; for example, weekend conversion rates always perform lower than on other days. So Monday and Tuesday ATRs by hour could vary significantly for Fri-Sun.

    2) How do you define viewthrough rate and how were viewthrough conversions attributed to specific exchanges? Currently, unless utilizing viewthrough attribution for A/B test purposes, viewthrough attibution can be extremely murky. This is why most of our data focused on post-click conversion rates, since most advertisers find this metric to be a universal standard. As you can see from the following chart, bit.ly/YDvGrV, when we looked at post-click conversion rates on any given weekday, there’s a clear pattern that a) the recency rates are much higher on FBX, and b) we see 2-5x higher post click conversion rates even 2 weeks out.

    3) In addition to the blog post you referenced in this article, we completed an extensive study earlier this year focusing solely on viewthrough conversion rates through an A/B test, and still found some differences in behavior by exchange. We saw that if you put aside all click value, we saw 36% more conversions on FBX than anywhere else – including Google Display. We used campaign data from hundreds of the largest direct response advertisers on FBX, which were retail and travel advertisers that see over 1MM unique visitors on their site per month. In our analysis, finance actually performs unlike any other traditional DR client because the interaction and conversion definitions are so different, so we try to carefully match up our analysis with advertisers that have similar conversion events. For more info on our viewthrough conversion analysis, you can check it out here: triggit.com/facebook-exchange-playbook

    I agree that optimization by different channels is key. Hopefully as a greater number of FBX partners get more and more refined data, they’ll see that there are important distinctions to make when viewing the conversion behaviors of potential customers on Facebook vs. other exchanges.

Leave a Reply

Your email address will not be published.

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>