(Money ÷ Accounts × 30) × (100 ÷ Churn)
easy hard
hard
Slide 21
Slide 21 text
Trends > Absolute numbers
It doesn’t matter if your churn rate is 2% or 15%, it
only matters that it goes down over time (does it?)
Slide 22
Slide 22 text
Raw data
(billed amount per day, last 3Y)
Slide 23
Slide 23 text
30-day moving average
(billed amount per day, last 3Y)
Slide 24
Slide 24 text
Days where billing didn’t
work or where we had to
catch up with lots of
payments
Slide 25
Slide 25 text
These spikes are not
meaningful and distract
from the overall trend.
Slide 26
Slide 26 text
Locally weighted scatterplot
smoothing (LOESS)
(billed amount per day, last 3Y)
Slide 27
Slide 27 text
“LOESS is a non-parametric regression
method that combines multiple
regression models in a k-nearest-
neighbor-based meta-model.”
(Gesundheit!)
TL;DR: LOESS gives you meaningful trends
https://github.com/jasondavies/science.js/
Slide 28
Slide 28 text
Real world example:
raising our prices
Slide 29
Slide 29 text
We increased prices
between 60% and 250%
Lifetime customer value up
conversion rate down
Slide 30
Slide 30 text
To increase conversion
rate, we introduced a
lifecycle email sequence
(after some tweaks it
worked beautifully!)
Slide 31
Slide 31 text
Segmentation
Find places to improve by plan, by number
of users, by number of support emails, by
amount of data, by referrer, etc.