In Sir Arthur Conan Doyle’s classic novel “The Lost World”, there is an early passage where journalist Ed Malone has had his proposal of marriage to Gladys rejected. He asks her for a reason, and she replies that while he is a good man, she is looking for something more – a man who has tasted adventure, and laughed in the face of danger. He despairingly asks her where he is supposed to acquire these attributes – he is after all a humble man of the press – and she answers simply that “There are heroisms all around us”. Needless to say, several months later, our hero was on a remote plateau in South America, running for his life from flesh-hungry dinosaurs . . . !

I was thinking about this the other day, while on the receiving end of a frustrated rant from my chairman. He is passionate about the intelligent use of customer data to drive effective email marketing programs. Make no mistake – so am I, but there is only so much you can learn about response behaviour if all you have to work with is basic open and click data, and I said so. Which was when I got a response very similar to that of the unfortunate Malone – along the lines of “there is always data waiting to be derived, you’ve just got to use some lateral thinking to uncover it.”

All of this was at the back of my mind earlier this week while I was looking at e-marketing that I’ve received over the past month. As you might imagine, I have a professional interest in receiving as many marketing emails as possible, and there were 575 for this period, of which slightly over 20% had ended up in my junk folder. Now, for clarity, we are only talking about marketing emails that I have provided permission to receive, and I’m excluding the plethora of fake Viagra, fake Rolex, fake lottery win notifications, etc. from this discussion.

We use Barracuda as our corporate spam filter ( along with another 100,000 corporate users around the world ), and its settings are all still the same defaults that it came out of the box with. Nonetheless, it’s pretty efficient – yesterday ( 01st April ) 1.89 billion emails were processed by Barracuda users around the world, with 1.71 billion being rejected – roughly 90% ! Our own stats are roughly comparable – on average, we receive around 4,100 inbound emails per day, of which approximately 3,600 are blocked at server level ( these are generally the ones of the “grow your penis size” variety ). Of those that do get through, approximately 10% get “tagged”, meaning that they will deliver to my address, but will be re-directed to my junk folder, and it’s these emails that I’m most interested in for the purposes of this discussion. That is to say, genuine marketing emails that have been mis-classified as spam – false positives, in other words.

So what I did was to copy the marketing email data from both inbox and junk folder into a spreadsheet, which I then ran some data analysis against. And, lo-and-behold, I started to see some really interesting trends emerging, as follows:

• In terms of activity, my marketing email traffic builds up consistently over the course of the week. Monday delivers around 14% of my total weekly activity, building up to 25% on Thursday ( the most popular day of the week ). Only 8% of total activity arrives over the weekends.
• On average, 22% of the permissioned marketing emails that I receive end up in my junk folder. However, there are notable variations within this overall number. On Mondays and Fridays, this number drops to around 15%, while on Thursdays and Saturdays it is as high as 30% - there are definitely some no-go areas for email marketers to be aware of !
• Time of day also flags up some material variations. I split the day up into “mornings” ( 08h00 to   13h00 ), “afternoons” ( 13h00 to 18h00 ), and “night” ( 18h00 to 08h00 ). “Afternoon” comes up trumps in terms of most favourable treatment from the junk folder. Only 16% of all marketing emails received during this period ended up in my junk folder, compared with around 24% for the other time periods.
• I then extended my test to look at subject lines, and categorised them as “short” ( less than 33 characters ), “medium” ( less than 66 characters ), or “long”. Here, “short” was the clear winner – only 11% of these emails ended up in my junk folder, compared with 26% for “medium” and 23% for “long”.
• I also wondered about the impact of using special characters ( “£”, “%”, “!”, etc. ) in the subject line. Many marketers will substitute “£” with “GBP” in the belief that it will result in an improvement in spam filter treatment. So I applied this logic to my sample data set, and – interestingly – there was no material variation at all between those marketing emails containing special characters in the subject line and those that don’t, in terms of inbox vs junk folder placement.
• However, for that other ongoing debate in e-marketing circles – the use of the word “Free” in the subject line - the results were notably more clear cut. Where “Free” did form a part of the subject line, junk folder placement was 30%, compared with 20% where it didn’t. Junk folder placement rose to a remarkable 46% when “FREE” was deployed in block capitals !
• The final element of my informal analysis concerned frequency – are spam filters influenced by how often marketing emails are received from a sender ? So I split my data out by “1 – 4” ( ad-hoc to weekly ), “5 – 8” ( more than weekly to bi-weekly ) and “9 & above” ( more than bi-weekly ). The first and third categories showed junk folder placement of around 75%, but for the middle category        ( “5 – 8” ) this number was only 11% - perhaps a good steer in terms of how often consumers want to be interacted with ?

I’ll be the first to hold my hands up and say that this hasn’t been a particularly scientific analysis. It is obviously influenced by the type of e-marketing that I sign up for ( although it’s pretty wide-ranging ), and we are only looking at the behaviour of a single spam filter package ( albeit an extremely well-known and widely used one ). One could also argue that there is not a direct causal relationship between the factors that I have examined on the one hand, and junk folder placement on the other. Spam filters don’t only look at email content – they are also influenced by other factors such as sender reputation metrics, broadcast volumes, and blacklisting notifications.

However, it doesn’t take a huge leap of logic to suggest that while the use of “Free” in the subject line doesn’t directly result in junk folder placement, consumers may be implicitly more pre-disposed to complain about the type of emails which contain “Free” in the subject line, which then filters through     ( excuse the pun ! ) in the form of poorer sender reputation metrics, which in turn increase the likelihood of junk folder placement.

That’s not really the main thrust of this article, anyway. The greater point that I was looking to make is that email marketing has always been extremely popular for the rapid test-and-learn potential that it offers. While e-marketers may view a simple email list as providing extremely limited scope for any form of more detailed behavioural analysis, this is actually not true at all. Over and above the usual sources    ( click-stream data, web behaviour data, transactional data, etc. ) there is masses of less obvious, but extremely rich data, that can be used to complement an email program. Sometimes it just takes a slightly more creative thought process to identify that data. To slightly mis-quote Sir Arthur Conan Doyle – “There are insights all around us . . . !”

Guy Hanson
The Database Group