Predictability

Imagine a school of fish. The movements of any particular fish are inherently unpredictable. We could (and do) gather data on the individuals in the school and can make predictions as to the movements on the individual. But we're entering a world of probabilistic modeling. It's a world of variance.

But the movement of the school is far more predictable than the movements of each individual component. When we project lifetime value, we're predicting the trajectory of the school.

Cohorts

The backbone of lifetime value analysis is the lifecycle cohort. The lifecycle cohort is the group of customers who first engaged with your business at the same time. By collecting customers into cohorts, we're effectively standardizing many of the factors that influence customer behavior. For example, the state of the product, market, economy, competitive activity, campaigns and marketing, were all similar for a given cohort. We can then analyze this cohort as though it is a single customer and easily compare across cohorts and time periods.

Predictability of Cohorts

Where the behaviors of individuals can be hard to predict, the behaviors of cohorts are usually quite predictable. The key is to produce useful cohorts that can be tracked over time. We want to study our cohorts not just from the customer level (e.g. the count of all of the customers who joined in a given period) but at the value level (e.g. the revenue contributions). As the cohort declines and customers churn out, we may see increases to the value curve as the less engaged customers fall out.

‍

Hakuin delivers lifetime value estimates that businesses use to plan and test

‍