Kill the zombies! Segmentation and engagement metrics for better email deliverability

If you follow the email marketing industry, you know that engagement is quite the buzzword lately.  But Engagement isn’t new at all. It has been a part of the filtering mix for quite a while. ISPs including Yahoo! (Xtra) Hotmail and Gmail are adding clicks, opens and other measures of user engagement to the long list of other engagement metrics that have been in use for a while. All these metrics try to do the same thing — figure out which messages are truly wanted by subscribers.

ISPs are measuring engagement and using it to decide who gets to the inbox, and who goes to the junk folder. In simple terms, the ISP is basically looking at whether or not your subscribers open, click, and in general, “interact” with you. If you send an email that mistakenly goes to the junk folder, then the subscriber moves it back out, you scored some engagement points. If your subscriber clicks your links or hits “reply” to send you a message, you get some engagement points.

Returnpath’s George Bilbrey says to senders:

“Treat inactive subscribers differently: This is probably the biggest change that most marketers need to think about. Mailing to a lot of inactive accounts may actually make your reputation look worse at some ISPs. Segment out inactive users and run a win-back campaign. If you cannot win back these subscribers, you may simply want to stop mailing them altogether.”

Over at Clickz, Jeanne Jennings had this to say about inactive members of your list:

“If these folks really aren’t that into you, they may take the next step and report you as spam. It’s like that shunned suitor who just won’t go away; eventually the victim will consider him a stalker and get a restraining order. Keeping inactive names on your list can open you up to blacklisting and deliverability issues.”

There is an art to deciding who is engaged and who is not.  This will depend on your buying cycle and the types of emails you send. It is good to use an email expert to help you make a matrix for your own business but there are some things you can consider:

Do you have strong calls to action in your emails – so that there is something to click?

Do you have a genuinely relevant and  interesting email stream, sent at least bi-monthly (6 per year)?

If you have a frequent email (weekly or more) do you allow people to control the frequency and type of emails they get using a Preference Centre?

  • From time to time you should dissect your email list to identify who have never opened, clicked or bought something from you. We call them ‘zombies’.  They bring all your metrics down, they impact your engagement measures and they don’t pay their way.  Try to get them to wake up – or kill them off.
  • Next look for who is in a coma – used to engage and now don’t.  Talk to them differently too.
  • Who is on their way out?
  • Who are you best responders?  Make them feel special, use them to spread your word, and keep up the good work!

There is much to this and a good agency can help you do this and come out the other side with a more profitable program.

And worst case is you get to kill a few zombies!

 

 

 

  • Feynman Diagram

    The zombies you talk about can be regarded as unlabeled data (in the language of data-mining, machine-learning, data-classification, etc,…). It means that this data hasn’t been assigned a class label, eg, in a 2-class system, say class A is a group with “likely to buy from you” and class B is a group that is “unlikely to buy from you”. Well, if there are zombies in the data-base, it means that those people haven’t been categorized as either belong to class A or class B. The zombies are unlabeled, they are neither A nor B. How to go around this in terms of marketing segmentation? Well, there’s new algorithms called semi-supervised-classification (SSC), which isn’t widely adopted yet because it is still unknown to those outside of the data-mining & machine-learning community.

    What these SSC algorithms do is to attempt to classify the unlabelled data (the zombies) based only on the available labelled data already stored in the database. It means that after running SSC algorithms on a database that contain zombies, it will lablell the zombies into either A or B and afterwards, there’ll be no more zombies. SSC algorithms is successful in its use by Amazon to promote items to the zombies in its database. There are a few variant of SSC available on the net, where your development team can look it up.

    A popular one is called TSVM (transductive support vector machine). The traditional classification (or segmentation) algorithms of today are based on a database that is fully labelled (ie, database that has no unlabeled data or no zombies). If zombies exist, it is simply ignored or deleted as uninteresting. Well, there may be interesting individuals in the zombies group, but to find out who are those in that group, then SSC must be used. Traditional segmentation can’t do that.

  • Feynman Diagram

    In the future of business, brand loyalty will become very important, so segmentation and engagement will probably have a different form. They will still be there I think, but perhaps in a another form that may be a little different from today for the reasons that machines of the future can crunch customers’ behavioral data in nano-seconds then make automated recommendation. Amazon is already doing automated recommendation at the moment, however the recommendation’s linguistic expression is very basic, such as : “Customers who bought item A also bought item B” or “You may be interested in the following items, B, C, D, …, since users who clicked or viewed this current item A, have also looked at them “… Recommendation of the future, will be more advanced. The recommendation message will be linguistically indistinguishable from that designed by human experts as templates as it is done today. Machines will be able to strung up phrases, sentences, paragraphs on the fly from a corpus of marketing text knowledge base. This will also be personalized too. Personalized recommendation today is still template based. The same template is used for everyone.

    I think that the first speaker in the following panel is correct about brand loyalty as technology evolved in to the future.

    Are We Ready For the Coming ‘Age of Abundance

  • Jericho

    Interesting look into personalisation in the future…. makes sense the robots and algorithms would get more and more advanced… And crazy that today’s concept of ‘personalisation’ is based on the same rules for everyone, so overall really quite a rudimentary level of ‘personalisation’…