Nov 26, 2013

What Does Your ISP Say About You?

If all you knew about me was my email address, what could you say about me? Put on your Sherlock Holmes stalker hat. Does an address by itself tell you anything at all?

Take my personal address:
john [dot] 4man [plus] 420 [at] gmail

First, let’s look at the prefix. Well, we know that the address contains the name John in it. And we know that John is a fairly common given name for men, so it’s a reasonable assumption that this address belongs to a man. And there’s an age distribution associated with the name John. Today, it’s not nearly as popular as it used to be, so that narrows (slightly) the age range down.

And how about that dot or the plus sign in the name? Maybe that says something about my age or tech savviness that I know those are legal characters in a gmail address. And perhaps the 420 in the address says something about my age range.

We could go on.

Indeed, prefixes say a lot about an email address. Fun fact: if you have "moh" in your email address prefix, you’re 3 times more likely to be male. Why? Because that letter combination crops up very, very often in "Mohammed" and not a whole lot elsewhere.

On the other hand, a prefix containing "grl" makes you 15 times more likely to be female. There’s lots to learn from so few characters.

But what about the domain? Gmail in my case. Gmail is ubiquitous. It’s a free mail provider. So surely it says nothing.

Not so! When we compare email addresses across the top ISPs, the differences are stark.

MailChimp’s top domains

For this post, we’ll look at MailChimp’s five top receiving domains, which for 2013 are Gmail, Hotmail, Yahoo, AOL, and Comcast. Here’s a chart of our sending volume to these domains this year. Gmail is more or less killing everyone.

ISPVolume

Age

Gmail was introduced 7 years after Yahoo. And Yahoo was introduced 6,000 years after AOL. So we’d expect age differences in those who choose these different services. But how do we find out the ages of subscribers in the system?

It turns out that there are already tens of millions of email addresses for these top domains with associated birth date merge tags already in MailChimp. Pulling all that data down, we can construct the age distribution of each domain’s email address owners. In the graph below, I’ve provided the first quartile, median, and third quartile age for adults using each service.

ISPAge

Let’s go through the results. Gmail and Hotmail both have median ages of 31. Being from the States, this initially surprised me until I learned that Hotmail still maintains its popularity amongst the young abroad. Furthermore, in order to nab an Xbox Live account these days, Microsoft requires you register using a Microsoft-owned freemail platform (Outlook, Live, or you guessed it, Hotmail). That’s another source of young’uns.

Yahoo is a bit older with a median age of 34. AOL makes a startling jump up to a median of 43, and Comcast is all the way up at 49 years old.

Why might Comcast be so old? In order to get a comcast address, you have to buy internet service from Comcast, and so, plunking down that cash will naturally shift the age of the service up.

Browsers

Whenever recipients engage with email from MailChimp, we get back the user agent string from the recipient. This allows us to determine the browser or device the reader is using. So how do these browser choices break out by domain?

ISPBrowsers

Let’s walk through it. A couple things should jump out immediately. First, the iPhone browser (Mobile Safari) is the most popular browser used in engaging with email (here I looked at clicks) for all domains except Comcast, because those folks skew pretty old. Not only does iPhone not come in first, but Safari knocks Android completely out of the top 5 for Comcast—these email addresses are heavily desktop based.

Note how Internet Explorer tracks extremely well with our age data, increasing in popularity with the domains that are used by older folks. Chrome, on the other hand, displays the opposite trend—skewing toward the young.

Now, I haven’t touched on the most frightening bit of information here—the enduring popularity of AOL Explorer. Not only do millions of folks in the US continue to pay for and use AOL dial-up, but they continue to actually browse the web with AOL’s browser. This browser was discontinued in 2006.

Interests

So it’s apparent now that the domain associated with an email address has a bearing on age and technology use, but does that say anything salient about the content preferences of the person behind the email address?

Using MailChimp’s Wavelength database, we were able to pull the categories of content where each domain was either over-represented or under-represented. If a disproportionate number of email addresses from a given domain are subscribed to a type of content, that says something about the general interests of folks on that domain. Here are the results for our top 5:

ISPInterests

Note how Gmail users are overly represented in the mobile and software categories, which interestingly enough is not engaging to Hotmail users, even though these domains have similar age distributions. Following along age lines, AOL and Comcast users are not interested in technology-heavy categories like games, daily deals, and social networks.

On the other hand, these readers are intensely interested in politics and the finer things in life, like homes, gardens, and good eats. These latter interests take a bit of money, which gels well with the age range of the domains.

It seems that political engagement in email skews older, which matches well with what we know in general about age and politics.

Segmentation + A/B testing is the best learning experience of all

In your MailChimp account, you have access to all sorts of data: domain, device preference, and geolocation to name a few.

Sure, you may not know each subscriber personally, but if you’d like to craft content differently when engaging a particular audience, you can make basic assumptions using the data from this post to form quick and dirty segments.

You could make some domain and device static segments, and then A/B test content at a segment level to verify that for your list, the trends I just presented hold. Maybe your Comcast readers really don’t care about your Android app, and Gmail readers are too busy fiddling with their iPhones to vote for your candidate.

And if you ever meet a subscriber who prefers using AOL Explorer, learn more. Maybe they were frozen in ice or something.