by Cath Rogers.
Demistifying cohort analysis is the first of a multi part instalment from AirTree Ventures, which also covers unit economics, key metrics to manage your business by, financial modelling and board reporting.
This series isn’t just about making you investor ready but more importantly it can help you to use data to make better business decisions and sensibly minimise cash burn. We firmly believe that with the impending downturn in global tech markets, businesses that best understand this will be far more likely to survive and prosper.
Demistifying cohort analysis – how it can drive better business decisions
At AirTree, we use cohort analysis to help drive insights and inform decisions within our portfolio companies and to better understand the quality of potential investments.
What is a cohort anyway?
A customer cohort could be any defined grouping of customers, but is most often temporal in nature. For example, all customers who first made a purchase or signed up to a service during a certain month, say January 2015 form part of the Jan-15 cohort.
Why should I care about cohorts?
Each customer group or cohort can be tracked and compared with prior and subsequent customer cohorts to generate valuable business insights. The simplest reason for doing this is to understand customer behaviour over time and hopefully to make better business decisions that improve desired characteristics such as total spend, retention or repeat rates.
What will cohort analysis tell me about my business?
The most useful metrics to track will vary according to business model, but cohort analysis can help answer questions like; How long does it take for a customer to make a second and third purchase on average? Is purchase frequency and amount increasing or decreasing over time? What percentage of subscription customers are still active after 6 or 12 months? These are all important things to know when deciding how much you can afford to spend to acquire each customer. Also, if you’re modelling cash burn based on assumed customer repeat behaviour, make sure you track these repeat assumptions relative to actual cohort behaviour, otherwise you risk running out of cash more quickly than you thought.
Why over-complicate things? Cant I just use a simple average at a point in time to get to the same answer?
Looking at metrics across an entire customer base in a given period is a necessary and useful exercise but this can mask underlying trends. For example if businesses are growing rapidly, net churn figures in a given month may look very low or even negative, however it’s possible that churn within each cohort is actually very high and as soon as marketing fuelled growth is dialled down, the business will begin to shrink. Equally, average order value for a given month might be impacted by historic pricing or changes in customer base that will be much more apparent when looking at individual cohorts.
Once I have the cohort analysis, what next?
Whether fundamentals are improving or declining, it’s important to understand why to help zero in on potential reinforcing or corrective action. The next level of insight can come from segmenting the customer base to understand why certain cohorts perform better than others. For example, customers of a certain profile or derived from particular channels. Marketing campaign effectiveness is often measured up until the point of conversion or initial purchase, however customer acquisition costs should be based on the customer’s lifetime revenue and gross margin potential. Deep discounting can sometimes see an influx of customers at what may appear to be attractive acquisition costs, however if customers quickly churn or never repeat this strategy may be ineffective.
Is there a lot of development team effort required to do this analysis?
In most cases, the data can be retrieved with a simple SQL (or other database) query. It shouldn’t take more than a couple of hours of devt time to set this up initially and depending on customer numbers the query may run overnight. Depending on your systems you may be able to extract exactly the data you need or it may need some manipulation in excel. If you get a data dump in excel it should include monthly purchase activity by customer. First sort the data by customer (using a unique identifier) then allocate customers to cohorts based on when they first became customers and analyse from here.
How often do I need to do this?
Optimal frequency of cohort analysis will vary across businesses but on average monthly is a good starting point. At AirTree we like to see monthly cohort analysis from our portfolio companies included in monthly board packs (more on board reporting later in the series)
Download AirTree’s cohort analysis excel template at www.airtree.vc/resources and feel free to post about your experiences with cohort analysis, good or bad at www.airtree.vc/resources or tweet at @airtreevc or @cathrogersvc. We’ll do our best to chime in if we can be helpful.