← All work

Parcelry

Subscriber churn

A survival model showed who left in the first 60 days. A welcome offer for that group kept more of them.

+9 pts

more customers kept

Role

Analyst, with the growth team

Tools

Python, SQL

When

2021

The problem

A third of new subscribers cancelled within their first year, and most left in the first two months.

Marketing was spending to replace them faster than the product team could find out why.

Sample retention curves over 24 months, before and after

The result

A year later, more of the offered group were still subscribed, and the offer became part of every new customer’s first month.

The model now refreshes monthly and feeds the growth team’s weekly list.

How I got there

I built a survival model on the first 60 days of order history in Python and picked out the customers most likely to leave early.

With the growth team I ran a six-week test of a welcome offer for that group only, with a held-out control.

Create a free website with Framer, the website builder loved by startups, designers and agencies.