Simple Buyer Mining

Getting visitors to your site is easy, if you have cash. Getting the right visitors to your site…well, that takes a little more time and effort.

Over the weekend, Andy Beard pointed me to a Mike Hill video on ‘Buyer Mining’, which was a response to an interview Techcrunch did with two ex Googlers who started the ecommerce company ‘Tell Apart‘. It provides sites with leads based on buyer demographic data. They are so confident in their service and the ability to provide such leads, they only charge a percentage of the revenue increase. Think ‘affiliate style CPA’ here.

Still with me? Cool, because after the video, I’m going to demonstrate how you, as an ecommerce owner and affiliate, can do something similar, for a fraction of the budget.

Buyer Mining

So, what is the ‘buyer mining’ Mike Hill talks about? Well, put simply, it’s looking at key buyer demographics to identify patterns. Some maybe fairly obvious such as the average Tampon buyer being either in the military/Female or both. Others may not be so easy.

Some of the key demographics to grab would include geographic location, age, and gender. In addition, we want to look at the average basket price and whether they are returning customers. Here is Mikes explanation:

Now, the problem is to be statistically reliable, you need a lot of data. In fact, you’ll need more than your average ecommerce store can produce. Still, even with a limited data set, you can make some choices when it comes to lead acquisition.

Simple Buyer Mining

Scenario
Let’s assume you are a small/medium ecommerce store selling red widgets, and one of your lead acquisition methods is Facebook ads. These can be a great for targeting users, but it’s often hard to tap into and they are fairly expensive, unless you are making a decent return.

Broad Matching Demographic, while possible, is often ineffective. So, to make best use of Facebook ads, we need: a location, a gender, and ideally, an age range. Two out of the 3 are very easy for any company to provide retrospectively, and well, 2 out 3 ain’t bad now, is it?

Determine Geographical Location

Virtually all ecommerce systems will store the IP address of the purchase, which can be used to determine location within a few metres in some cases. More likely, it will resolve to a city or a country, at minimum. Be wary, however, as using an IP is not always accurate. For example, my home IP address for many years claimed I lived in London, when I was 200 miles away. At best, it will offer country accuracy with potential city/town. The exception is AOL customers, which some IP systems identify as being in the US regardless of where they are in the world. In addition, you can always take location from the address given at the time of purchase. This will be more accurate, but might only be available for non digital products.

Gender

There are lots of ways to determine gender, and this would be a great place to plug my new tools, except they are not ready yet. Here is a simple way to determine gender in historical data:

From your purchase logs, pull all the first names, and pass them through a gender dictionary to identify female/male names. With the exception of a few Kims, etc. most western names will be one gender or another, and therefore, those running Western based sites should be able to process and get gender demographic quickly and easily. You can do the same for Arabic, but Chinese names are a bit more problematic. Names from some Asian countries and the Indian Subcontinent, however are generally a nightmare to work out this way.

Age

Remember I said 2 out of 3? While you can analyze names to get age,  it is really too unreliable for use. Likewise, browser history is not an option, which leaves only one choice: asking the user for his or her age. Sadly, when looking at historical stats this is not much help. It is also one of the most useful demographics. So much so, I try to encourage all my clients to gather it. For subscriptions and certain products, this is easier to warrant, and obviously you do not want to much cluttering your sales flow, but I strongly suggest you include it even as an optional field.

Facebook Ad fun

Once we have the demographic information, it’s time to do some mining and try to identify trending patterns. Specifically, we want to generate targeted ads.

From experience, the best place to start is with the demographic you should definitely have: location. Depending on your data size, the more you have, the tighter your geographical targeting should be. With this in mind, take your data and split it into geographic regions. For each region, split the data by gender and age, if you have it.  The biggest of these groups will be your first target for Facebook ads. From there, it’s simply a case of setting up one or more ads within Facebook for these groups.

Split Testing

It goes without saying that it’s worth doing straight a/b testing, but it’s worth remembering different groups will react differently, so do split tests for each separate profile group. The results maybe different or you may have a clear winner across all the profiles. Don’t forget to cookie your test groups if they purchase at a later date so you have the split test data. (It’s worth recording regardless to see if a group has a tendency to purchase later.)

Affiliates

I mentioned the above could be used with affiliate marketers, as well as ecommerce owners. This is sort of true, but the ecommerce owner will need identify which of your leads resulted in sales, otherwise you will be grouping blind.

Of course, you can do a blind variation by determining the gender of each lead through CSS history attack. To do this, use the IP for location, split your users up, and test each group to see which is more likely to purchase. However, this could prove to be an expensive option.