Thread by @dunkhippo33

Today's tweet thread is on cohort analysis. What are they? Why should you care?

tl;dr cohort analyses is a good way to know if you are actually getting better or worse with your business.

Most ppl look at revenue. Or product improvements. But cohorts are the way to go.

>>

1) What is cohort analyses?

It's a way to put your customers (or users) together in mini-groups and see if as a group your company is getting better or worse over time.

2) So for example, here's what a cohort analysis graph looks like:

https://twitter.com/patrickc/status/1130523323899686912?s=20

3) How do you read this?

If you double click on the graph, you can see that after 1 month, for the group of customers who started w you on May 2018, 92% of them were retained.

And by month 2, 75% of the original group that started w/ you in May 2018 was still with you.

4) Eventually, by the end of 1 year, 73% of the original group that started in May 2018 is still retained.

That information is ok in itself, but it's much more interesting as you compare May 2018 with subsequent cohorts.

5) For example, for the cohort that started in Aug 2018, something went really wrong. Retention was bad from the get go after 1 month (only 78% retained) and 68% retained from that group by month 9.

6) This is interesting, because you can go back and see if there was anything that you did differently for that Aug 2018 group.

Did you target customers differently in your marketing compared to before?

Did you change the UX?

7) If you lump all your users together and take average numbers over time, you won't actually know if changes you are making are making things better or worse. But Aug 2018 clearly shows that that cohort was significantly worse.

8) Sometimes significant changes in cohorts occur due to seasonality or changes in an economy (such as COVID!).

But sometimes they are due to changes that YOU have made in your company. And that's what you want to tease out.

9) You can do cohorts in terms of retention percentages such as shown here. But sales for the cohort overall can be another good way to do cohorts.

Why?

10) % of ppl retained is often not the best metric. The fact of life is that you are going to have some churn. And if you churn out your worst customers, that's ok.

What you really want to know is whether your dollars earned in a given cohort is going up or down.

11) For ex, the best SaaS cos still have some churn, but their most loyal customers are upsold to effectively (increase # of seats / price package etc).

So the cohort as a WHOLE actually increases revenue over time *even though the # of customers may be declining over time

12) In SaaS, it's common to use monthly cohorts as the time period. But if you don't have a monthly subscription prod, you'll want to use a diff interval.

For ex, it could be daily users for an app. Or weekly subs for a weekly newsletter. Or quarterly spend for mktg budgets.

13) For more info on cohort analysis, we did a webinar on it:

14) In short:

-Set up cohorts early in your startup to know whether you are getting better or worse
-Select a time period that makes sense for your biz
-If $$ is your KPI, # of ppl who churn is less relevant that dollars churned