Jump to Content
The opinions expressed herein are my own personal opinions and do not represent
my employer's view in any way.
newtelligence dasBlog 2.2.8279.16125
Email Me (Tim Gaunt)
© 2013 Tim Gaunt.
In my last post about Cohort analysis I briefly introduced what a cohort graph is, in this one I'm going to go into a little more detail about how you can use it. I'm planning on releasing some code so you can add it to your reporting suite in another post but if you can't wait until then, I've thrown up a quick online cohort generator at: www.quickcohort.com -just dump your data into it and click graph.
A cohort chart can give you an idea of customer loyalty -and- an indication of potential problems in their lifecycle.
The key metric it gives us is customer loyalty. This is also one of the most frequently ignored metrics in every company (mostly because people aren't clear on how to measure it) yet it's probably one of the most important at the same time.
That depends very much on your business and as a result the graphs will likely look very different. It's probably easiest to compare what a couple of business types -a retailer and an online magazine- might consider loyalty.
As you can see, although there may be some cross overs (Visits to the store and Logins are probably quantified by a similar metric) the "what" depends very much on your business but the longer the duration between the first and last engagement in all scenarios above is how you can demonstrate loyalty.
Unlike most tables which you read left to right (a row at a time), you'll probably get more value out of a cohort chart from reading it by a column at a time. This will enable you to spot possible problems in the user's lifecycle.
Problem points in the user's lifecycle can often be spotted where the colours change in the same column. The greater the difference between the shades of colour, the bigger difference is between the two months. In a perfect world, 100% of the customers from Month 0 will still be using your site in Month 12 but life is rarely like that.
Looking at the chart below of user logins over time, the eager should spot that there are three months which appear to have issues: Month 3, Month 5 and Month 9:
If you're not sure on what you're looking for, you're spotting those columns which have a similarly shaded background which then lightens in the next month (or in the case of Month 9 is completely unshaded.
As the chart above was logins over time, a quick glance over this cohort chart this would suggest the following to me:
When reading a cohort chart, you can generally discount the last cell of each cohort as it's the current month.
Logically, to be a repeat customer, you have to make at least two purchase from the retailer so the duration we're interested in (the month) is the period between the first purchase and most recent purchase.
First Date (Month 0): First Purchase Last Date (Month x): Most Recent Purchase
Check out the cohort chart below and see what you can interpret.
Remember, Month 0 represents the first purchase -all customers appear in this column. Looking over the cohort chart, of the 69 customers the retailer had in October 2010, 39% (27 customers) were still around in November 2010 (month 1), 21% (15 customers) were still around in December 2010 and so on.
None of the 69 customers who made their first purchase in October 2010 are still a customer 12 months later (although if your customers tend to make a purchase near the end of the month, the customer who made a purchase in September 2011 may still make a purchase).
So looking at that chart, 30-40% of customers would make a second purchase 2 months after their first purchase. Of the older customers (those which first purchased before March 2011) 10% would make another purchase 5-6 months later.
There are a few interesting things with the chart above, another is the sudden drop off in month 3 for those customers who first purchased in May/Jun 2011, similarly, the customers who signed up in Mar/Apr 2011 also stumbled in the same month, that could indicate some form of seasonal trend or change in marketing.
With a little background you will be able to get a much better insight into the meaning behind some of the numbers. If for instance you had changed your marketing routine around Jun 2011 this might explain the difference in numbers. It might be that your business is very seasonal (in which as you'd be better to look at a 24 month chart rather than 12 month).
Keeping a record of what you were doing around the different months is important, for instance you might start a pay-per-click campaign. Everyone's happy because you notice an increase in sales (so an increase in Month 0) but are they a valuable customer (a repeat purchaser) or a one-off? Cohort chart analysis will quickly highlight this to you as the increase in Month 0 will be reflected in Month 1.
Although time will tell, it would appear that a lot of customers make another purchase about 2 months after their first. This could be co-incidence or it might be that you're selling a product with a small sample accessory (e.g. a free pack of chalk with each chalkboard) and that sample pack runs out after a couple of months. Alternatively it could be a fault in the product e.g. you sell hinges and the oil runs out after a couple of months so they're buying more grease. By adding a little context to the data you'll likely get even more interesting stats out (we certainly have in the past).
Things start to get really interesting when you start comparing two cohort charts for the same customer base and period against each other.
First Date (Month 0): First Purchase Last Date (Month x): Most Recent Login
What's interesting when you compare the two charts is most of the customers who have made a purchase are still returning to the site (over 30% are still logging in in Month 11). This would suggest there's not as greater a problem with customer retention as there is with sales.
This could be because your store sells seasonal products but you keep customers engaged, it might just be a co-incidence but it should drive investigation.
In isolation it's helpful but only really gives you a top level view on a customer's lifetime with you, it's when you're able to combine this data with knowledge of your business, sales statistics, marketing strategy and information such as a customer's LTV (Lifetime Value) that it gets really interesting and useful.
Using a cohort chart and average sale value, you can use it as part of your sales forecasting and predict what your company's sales will be going forwards e.g.: if the Average Order Value is £10 and your average first 5 months looked like this:
You'll then know that of the 1,000 or so customers which sign up in the current month your revenue is likely to look something like this:
How have I got to those numbers? Well, of the 1,000 customers that purchase this month, 50% will make another purchase on or after month 1 and 20% will make a purchase on or after month 2. With this in mind, of the 1,000 customers from Month 0, 30% will make a purchase in Month 1 so the calculation is as follows:
([Number of customers] * [Percentage Returning in Month]) * [Average Order Value] = [Expected Sales]
(1,000 * 30%) * £10 = £3,000
There are a few assumptions with doing it like this -for instance, customers who purchase monthly will only be counted once etc but this is still a good start.
Using a cohort chart with sales data becomes very powerful as you're able to get a really good insight into whether campaigns are generating worthwhile leads or just generating traffic to the sites. If you're interested in reading more about that I'll overview it in another post as that gets pretty heavy on number crunching.
If you find that customers tend to drop off after a set number of months then it might be worth setting up some form of customer engagement which is triggered just before this point e.g. an email upselling a product that complements theirs or asking them to get in touch with feedback.
Don't forget to follow me on Twitter.