How analyzing B2B & B2E SaaS cohorts can lead to better inights.
By Tomer Ygael, Data Scientist — Liquidity Capital
Investment analysts have the tedious task of collecting and analyzing company data to uncover risk and opportunities. SaaS companies in this regard capture plenty of data enriched to determine trajectory and potential.
Analysts have many tools and strategies for assessing companies, one of the key ones is Cohort analysis.
Cohort analysis is a critical element of data analytics in analyzing SaaS companies or any consumer business. The cohort analysis is aimed to track the behavior of a particular group of customers over time. Cohorts are unchanging groups or groupings, e.g., no new customers join a cohort once formed, and customers cannot move from one cohort to another.
In many B2C businesses, we see cohorts behaving in the expected textbook manner. The churn rate from each step is kept, more or less, at a steady rate, creating an exponential decay in cohort retention. If this behavior is observed, it’s relatively easy to discern what is the effective churn rate and Lifetime Value of the company, and track changes in those metrics over time. Churn and Lifetime Value are standard metrics used to evaluate SaaS and e-commerce companies, and are part of Liquidity’s standard scoring and risk assessment model.
In many B2B and B2E businesses, not all cohorts “are born equal”. Moreover, not all cohorts decay constantly e.g. the step-to-step (StS) churn rate is not constant. As a result, one can not deduce much about the future. The challenge is, to understand what will be the pattern — predict how the existing cohorts will behave in the future and what is the most likely pattern new cohorts will follow. How would one assess with confidence those companies’ top KPIs?
In reality most companies, when dealing with this challenge, either use the empirical month over month measure of net ($) churn,
churnMi = 1 -(MRi-NMRi)MRi-1
Where: churnMi is the churn of the i’th month, MRi is the monthly revenue of the i’th month,NMRiis the i’th month’s new customer revenue.
or they come up with a customized manipulation of the data to drive logical results. The month over month (MoM) is not indicative of the future and can be highly impacted by the different behaviors of cohorts, while the customized versions are difficult to assess as its definition is non-standard.
The “High Priests” of data visualizations do not think highly of 3D charts. However, if there is a reason to use a 3D visualization it is this cohort chart. We use it to understand what models we should build to describe the “none classic” patterns. Here we can easily see how the cohorts differ from each other and how their behaviour isn’t following a clear pattern — New MRR, decay rate , revenues are not following the smooth exponential curve.
Image text — Revenue by step (top), the sum of all the cohorts step revenue. Revenue by cohort id (top middle), the total revenue of each individual cohort. 3D plot (bottom), the relation between revenue, cohort id and step.
Observing many companies and their cohorts, Liquidity’s data science team developed a model in which different types of fits are tested vs. the actual data of each cohort. All the fits are tested and only the best type of fit is chosen to describe the future of a cohort. To ensure the most accurate analysis, only the predictive part of the fit is used, creating a merge between actual data and fit. Using these merged cohorts we can create a representative cohort. The representative cohort can take various forms, impacted by the number of clients or New MRR trend.
Looking at the probabilities generated by the fits we can deduce which is the best representative cohort for the company, enabling Liqudity’s scoring model to have a decision point — computing a represnetative Churn and LTV for a company with cohorts that can take many shapes or forms.