Improving the accuracy of web analytics for social media tracking
There is a big elephant in the web analytics room… We are less likely to accurately track visits referred from social networks than we are visits from more traditional sources. ‘Less’ likely because the rise of those sources of traffic has coincided with (or helped cause) the rise in use of mobile apps or desktop clients at the expense of traditional web browsers.
So the visits you see as being reported as referred from one of the social networks are only the tip of the iceberg.
You’re almost certainly getting a better return on your social marketing that you think. It’s just not showing up where you’re looking for it. As we near the end of 2011 many are asking:
“If social is so ‘big’ why can’t I see those visits in my site’s analytics?”
If your calculations of the return on your efforts in social marketing are based on those referred visits, plus the visits from campaigns which you started and tagged yourself, you’re only seeing a small part of the story.
How large are the inaccuracies in traditional analytics?
You can get an idea of the inaccuracies from these tests. Thomas Baekdal posted some interesting estimates of the effect of incorrect social referrers on his site (he estimated that for Twitter 90% of visits were missing from analytics). The situation will probably have become more extreme since then as mobile continues to surge ahead.
Note though that during 2011 Twitter has changed the way it tracks referrers – now showing through the T.co domain. As I show in an example at the end of the post this has improved accuracy.
In another analysis by Awe.sm the headline figure was that Twitter drives 4 times as much traffic as you think it does. The majority of visits are shown as direct (null referrers). By the way, check out the Awe.sm tool which tackles this problem if you don’t know it.
I recently had a chance to see this effect for myself. One of the advantages is having some very low-profile personal web sites is that you can sometimes observe things which might be lost in the general noise of a busy commercial site.
Earlier in the year, Avinash Kaushik was kind enough to Tweet about one of my blog posts. The effect of having an industry leader like that post a link is instant and dramatic. I also know the exact source of the spike in visits.
Here’s the spike:
And this is where Google Analytics thinks they came from (date range changed to show just the day of the Tweet):
To break that down (who knows whether you can see the images wherever you’re reading this).
- The Tweet caused a clear spike of c170 visits (I really do believe we can agree that ’caused’ is ok here)
- 30% were counted as referrals by GA
- 60% were counted as direct by GA
Within the referrers (in Google Analytics v5 the Overview reports are now good for this kind of thing) we can see Twitter in top spot with 39 visits, followed by Hootsuite with 10 visits. Mobile.twitter.com sent 1 visit, by the way.
Avinash’s followers are tech savvy and the spike began as soon as he tweeted. The chances are the majority of those people were using some kind of app on their desktop or, more likely, a mobile device. Google Analytics had no referrer information and no campaign tag to tell it what to do, so it dumped the visits in the great bucket which should be called “unattributed” or “I haven’t a clue” but is know for historical reasons as ‘direct’. This does not just apply to Google Analytics, it will be true of any system which is using the referral data from the web browser to track the source of the visit. The ‘direct’ bucket is one of the standard features of web analytics.
Solutions to inaccuracy in tracking social media generated visits?
You can and should use nicely tagged-up campaign Tweets or Facebook links back to your site is only really a slight variation on the traditional outbound direct promotion. Tools such as Awe.sm and Hootsuite enable this to be done automatically.
But the real strength of the social web is when your customers refer you to their friends and the word spreads from person to person simply on the basis of the quality of your products or services. That amplification of the reward for providing good service, products and value is the power of the social web. And it’s the part which is least likely to show up in your numbers. In these cases, there likely will be no tracking parameters.
So, it seems that much of the discussion on this subject is revolving around the inability to track the return from social campaigns which companies have initiated themselves. There are solutions for that aspect already. But that kind of social marketing is very close to traditional on and off line promotion.
The aspect of on-line social activity which I think may be more significant is the ability for these new channels to amplify unsolicited mentions and praise of your products or service. Since you didn’t begin those conversations, you can’t tag them and you can’t easily see the benefit in your site’s data. We just don’t know.
The announcement of the new social reports in Google Analytics is extremely welcome. These will provide a great way for getting insight into how well people are reacting to what your are offering them. That’s hugely valuable. It seems a much better measure than calculations of engagement based on page views, time on site or viewing specific micro-conversions, because it is based on a specific indication of liking something or wanting to share it. There’s an element of intent involved in such an action.
But that data is still about what’s happening on your site. The great promise of the social web is that it will be a means for new people to learn about your products or services through the amplification of referrals. Unless I’ve got this wrong, we’re still largely in the dark about the scale of customer acquisition through social marketing.
Common sense says social marketing matters. I’ll go with that for now. But I’d like my numbers too. Any ideas where I can find them?
More resources on tracking Social Media in Google Analytics
During 2011 I’ve collected resources that show approaches to managing this challenge. I’m sharing them here in the hope they may be helpful:
- [July 2011] There is yet another excellent discussion here http://techcrunch.com/2011/07/14/twitter-drives-4x-as-much-traffic-as-you-think-heres-why/[Opens in new tab]
- [August 2011] Twitter are rolling out use of their t.co URL shortener, which has the effect that more Twitter traffic will appear as referrals from their t.co domain.
- Avinash Kaushik has started a discussion about this on Google+, complete with a link to download a suitable GA Custom Segment: https://plus.google.com/105279625231358353479/posts/LsHgmGicHQq [Opens in new tab]
- Tom Critchlow has a detailed post on the subject on the Distilled Blog, which includes an interesting comment thread. The discussions there also consider whether/when Twitter may apply the t.co approach to URLS which have been shortened by other services like bit.ly or owl.ly – http://www.distilled.net/blog/social-media/twitters-t-co-link-shortening-service-is-game-changing-heres-why/ [Opens in new tab]
- You may well find it interesting to examine the configuration of the ‘Social’ group in GA’s default Multi-channel Funnel Groupings. I’ve written a post about how to explore and customise those groupings here: http://www.cxfocus.com/index.php/google-analytics-tips/multichannel-funnel-reports-group-brand-generic-search/
- [December 2011]In November I got another opportunity to see if the situation had changed.Once again, someone prominent Tweeted about something on this site and caused a spike in traffic which was very marked and coming from a ‘known’ source.
But there was a difference in the attribution this time. 68% of the traffic was now correctly identified as being ‘referral.’
The clue can be see in the form of that top referrer: t.co is credited with nearly 80% of the referred visits.
As explained in the August update above, during the period between the original spike described in the main post and this new one, Twitter had imposed a policy of redirecting links through their own shortening system. These links are now being reported as referred from t.co.
This means that GA reports like the traffic sources reports and multi-channel reports are now giving you a much more accurate idea of the level of visits from Twitter. t.co is already set up as part of the default ‘Social’ channel is multi-channel, for example.
But the situation is still far from perfect. Those ‘direct’ visits are still suspiciously high. But this is a huge improvement in the space of a couple of months. There’s a lot of work going in to this and things are moving fast now.