Just like you have editors, debuggers, profilers, libraries, frameworks and other things, your friendly marketer has a bag of tools that they use to do their job. One of these tools is the so-called “cohort-analysis”.
Think about the people coming to your web site. When your friendly marketer ran that facebook competition, she attracted a lot of people who had never been to the site before (“first time visitors”).
Your marketer will be very interested in seeing what those new visitors are up to, so it makes sense to look at them as a group! Your marketer calls those groups “cohorts”.
A couple of weeks or months down the line, she will look at the “January facebook iPad competition” cohort, or maybe the “christmas 2012 first time buyers” cohort. She will want to know whether they ever came back to the site, whether they bought more, consumed more content, engaged, or whatever your key goals for the site are.
She will compare a cohort with the rest of the visitors to see whether specific marketing activity nurtured valuable leads or brought new, valuable customers.
All of this is very important for her, so if you know how to collect data for a cohort analysis, you will make her very happy.
In principle, implementing for cohort analysis is very straight-forward. All you need is a way to retain the original first date.
But there is one issue: original first date of what?
Let’s look at one example and come back to that question. So, let’s assume our friendly marketer is interested in cohorts of first time visitors she managed to attract via her different marketing efforts, her campaigns.
The cohort is directly linked to these campaigns and to the first time the people come to the site. In order to pull a report or build a segment that covers the “January facebook iPad competition” cohort, we must know when our visitors came to the site for the first time, and maybe whether they came from that campaign.
We therefore need to store the data of the first visit and make it available for reporting.
Luckily, conversion variables (“eVars”) have a setting that makes this really easy: set allocation to “Original Value (First)”.
What that does is show the first ever value that was passed into that eVar in reports, no matter what was passed in later.
In essence, you can just put a timestamp into that eVar on every hit. The system will make sure that only the first one will be reported on.
How exactly you pass a timestamp into the eVar is up to you. On this blog I just use “YYYY/MM/DD”, which means the report looks like this:
Of course you will use SAINT classifications to make the report easier to read and understand. I just aggregate into months, years and weekdays, but for your marketer, you should also create a classification that assigns dates to campaigns.
Pulling the classified report will hopefully lead to some answers or more questions, like in my case:
Like: what is it with those 16 people who came in March? I think that is easy: I’m one of them. And what about the searches? 24% were done by people who started reading the blog in November 2011? That’s odd.
People who started seeing in June 2013 saw an average of 3 pages, which is half a page more than almost everyone else!
There is some potential for research there, I’d say.
Back to the question raised above: which date do you capture?
The first visit is not the only important event in the customer life cycle! It makes a lot of sense to look at cohorts based on others, like first sign-up, first order, first installation of app, first comment posted, or others.
The easy answer is that for every date that your marketer would want to build a cohort for, you need one eVar.
The system provides up to 75 eVars for each report suite (we’ll explain what that is in more depth next week), and you will use some of these for other stuff, so it is impractical to do more than a handful of cohort dates.
That handful will already be extremely useful for your marketer, though. Just make sure you discuss them with her first. You want to track the relevant ones.