The secret’s out (in a small way) and Google is happily
sharing a top line look at the processes and algorithms that go together to
make up what the company describes as “One of the largest and most user facing
applications of ML (machine learning) at Google”, namely the Priority
Inbox
.

So how can we use this knowledge to ensure better inbox placement in
Gmail? And is this news to us anyway? 

For many years, those within the email industry have watched
the growth of engagement metrics used in the filtering of emails. This was
mostly restricted to the observation of results coupled with reading between
the lines of ESP best practice guides and feedback from sender support. 

We also had the reputation gurus like Return
Path to help steer senders down the right road of relevant emails and good
reputation, but we have never before had an ISP lay bare any part of their
filtering processes for all to see.

So why has Google done it?

I think the answer is that Google is pretty pleased with
itself, and wants to show how much thought goes into its user
experience.

I think it also wants to show the sending community and marketers that it is not the bad guy. Google is trying to demonstrate the time and effort it puts into ensuring that, if an email is really important to someone, it
gets to the top of the pile!

It is also sharing with the wider ISP community too;
sharing ideas that might ultimately help to improve email account filtering
across the board. 

I know there are those
that decry this sort of priority filtering, but with the volumes of opt in
emails due to increase considerably over the next few years, steps need to be
taken to ensure the viability of the email channel.

How does Gmail’s Priority Inbox learn what’s important?
And how can marketers use this information?

Gmail’s Priority Inbox uses certain features to decide what
emails are important to the recipient,

There are two levels of learning from features. The first is
global, allowing the product to be effective out of the box, as it uses
aggregated information from the global pool of data. The other and more
influential is user level behaviour and manual labelling.

The features themselves include social, content, thread and
label.

Social Features

“Social Features are based on the degree of interaction
between sender and recipient”. It looks at the amount of emails the recipient
has received and compares this with how many are opened or clicked. Which is
good? Isn’t it?

How can marketers use this feature to improve email
marketing? It should be fairly simple to segment out people who are no longer
opening your emails, and reduce the mailing frequency to these users, or even
try another channel. The message here is only to send emails to people who
are still listening to you.

Content Features

Content features look at what the email contains (e.g.
headers or terms in the subject line) and measures this against how the
recipient has responded to similar content in the past. This feature also looks
at the speed which the user responds to the email based on the content
included.

What does this mean for marketers? Well, if you are a good
mailer, and get high response rates from your campaigns, your content
signatures will be more positive than negative.

It sounds very much like a sort
of Bayesian process (haven’t got space to explain, but worth a Google) and
again this should reward the good mailer, who segments well and gets good
response.

Thread features

This looks at an email thread (who was it started by , did
the recipient reply). A marketing message will not generally be replied to.
This could be used by Gmail to differentiate what a personal email looks like
as opposed to a commercial one. Ideas to use this? Thinking caps on!

Label features

These are the labels the recipient places on the messages,
and are by far the most powerful of all the features. This is due to the
weighting applied to the different metrics, which places anything  the user sets as being the most significant
element when choosing how to prioritise your messages in the future.

Why’s this of interest to marketers? Well, if you’ve any
chance of being labelled by a user, then you have to ensure you email is
relevant and of interest, and something they would want to label.

Does it work?

As I mentioned before, the objective of Priority Inbox is to
improve the user experience and this seems to have been borne out in research
carried out on Google employees.

According to Google, Priority Inbox users spent 6% less time
reading email overall and 13% time reading less important email. Google found
they were also more confident to bulk archive or delete email. This almost
sounds like the Google employees who used Priority Inbox, were more efficient
at work too!

This is the face of email for the future. The objective has
moved from trying to “not” look like spam, to becoming one of sending important
or interesting emails to recipients; ones they welcome and ones they read.  

If you would like to read the paper, here is the link to the Google blog