On Retention
If you have been reading this newsletter, you know that most of the posts are market/strategy centric. To complement those, I’m also adding a new series of posts focused on how to take decisions inside an org, particularly from a product management perspective, since both are adjacent to each other. However, if you don’t prefer these (or are looking for other content), please share your feedback here.
People often talk about chasing retention, the holy grail of metrics. However, retention is a very broad term, often not understood well.
Firstly, what is retention? Retention is a measure of repeat behavior of users, which inherently is the value you are providing to your users. If they are paying for that value, you measure revenue retention i.e. are users coming back and paying? And if they are getting it for free, you measure usage retention. You also measure both because high usage often leads to higher revenue.
Retention has to be measured on the elemental unit. This unit is a measure of the core action of a product. For an ecommerce company, that is an order. For Uber, that is a ride. For Instagram, that is a post. And for Whatsapp, that is a message.
So what retention do you look at? Weekly, monthly, daily?
Once you know the core action, define the natural frequency of that usage. Is it a daily behavior - like eating food, consuming content or communicating? or a weekly product like ordering groceries? Or a monthly product - like paying bills? So for a product like Instagram or Whatsapp, daily retention matters more. A person ideally consumes content everyday and hence the platform would like to measure what % of users do so. But for paying credit card bills, you would want to see what % of users come and pay bills every month and try and increase that number.
By the lure of habit formation, most products would want to operate at a daily frequency ideally. However, this is tougher to do for products whose inherent use-case is monthly or weekly. The risk there is obvious - user might forget to use your app in a month. In that case, products use engagement levers to hook people on to the app so that people dont forget.
E.g. Flipkart launching video content. Buying (non-grocery) products is definitely not a daily use-case, but content is, hence adding videos makes people come to the app.
However, this is risky because unless there is a coherent benefit of doing this, such properties usually end up dead. Which is why Flipkart links its video content to Supercoins.
While the core action retention remains the main lever, as a product becomes large, you also start looking at different actions to see if they are optimised. E.g. For Spotify, it's not just about music listening retention, but also about podcast listening, playlist making and an overall app retention as well.
Improving Retention
There are different types of retention curves in products, broadly bucketed into
Smiling
Flattening
Declining
The question then arises - which part of the curve should you focus on?
The general framework is - if you have a leaky bucket (i.e. declining curve), fix that first. Leaky bucket means users are not finding value over a sustainable period of time, which manifests in retention curves not plateauing in long term.
Once you have a plateauing or a smiling upwards curve, you start to focus on shifting the entire curve upwards.
Now we know where to look, but how do you actually move the numbers?
If you reason out from what retention stands for - the core value of the product - it means that if you improve the core value it should be easy to improve retention. True, but not as easy.
Let's solve this through two approaches
Qualitative
Quantitative
Qualitative Process
It basically requires a study of whatever your users are doing on the platform and then improving it. The most basic thing to improve would be a bug that is hindering the user flow, then we move up from there.
Bug —> the product is not working as per specs. Big red flag if this is the reason for low retention, which means you're not shipping quality products.
UI —> the product is working as per specs, but users are not feeling the interactions as intuitive to use or understand.
UX —> the product is working functionally, but the flows are not correct. E.g. A multilingual content platform not asking for language upfront and then instead showing the wrong language content.
Feature —> you've not built what users actually want. This requires a fundamental relook at what you are building and why.
The qualitative process, in increasing order of importance, requires really good user understanding. If it's not there, you wont know if the experience you are providing is good or not. Here's where user anecdotes particularly help.
Quantitative process
As a PM, you apply your data chops to get numbers to tell you what to do.
E.g. You break down retention further into different user segments and lets say you find three user segments of varying retention.
Segment A is 5% size and has 90% retention.
Segment B is 30% size and has 70% retention.
Segment C is 65% size and has 50% retention.
Now, you try to find patterns that differentiate A, B, C.
One mental model to identify such patterns is to see what are the measures of core actions (remember from earlier part of the post?) or adjacent actions are correlated with high retention. One such famous pattern for Facebook was that users who added 7+ friends in 10 days had significantly higher retention than others.
If this is identified, you have a clear actionable on what you want the user to do to build his/her retention on the platform. So in FB's example, it would be adding callouts to add more friends etc.
Another way to look at this quantitatively is through funnels. You see the entire flow from the first page to the core action, and see where users drop off and why. Some dropoffs are inevitable, but if you have unreasonably high drop-offs, the pages are not designed to communicate to the user why they should go ahead to the core action.
To be clear, both these qualitative and quantitative processes have to be applied in conjunction, not in isolation.
If you are starting qualitative first, you start with an anecdote or observation, but still validate it with data to see if it makes sense.
If you are starting quantitative first, you start with data, but still form a hypothesis of why the data is a certain way and then try to prove/disprove it.
This process repeats itself continuously as you start, prove/disprove things and then move forward. Over time, if you have made enough right decisions along the way, you have a chance of seeing a retention gain.