According to Brian Balfour, CEO of Reforge and ex-VP Growth at Hubspot, retention is a silent killer. High-potential companies backed by elite investors failed despite insane growth metrics. In 2013, Homejoy, a.k.a the fastest growing startup funded by world-renowned Y Combinator crashed only 18 months after an impressive $38M round. Same old story for the unicorn that had acquired 1m users between 2011 and 2012 and ended up bought three years later…. for $15m (in other words: peanuts). In both cases, the post-mortem diagnosis was poor user retention.

This is precisely why we need to talk about user retention. Balfour considers it as “the one growth metric that moves acquisition, monetization and virality”. Today, the very essence of retention are social media behemoths such as Facebook and Whatsapp on top of the list. But even for smaller companies with fewer users, retention should be considered as a top priority and not be underestimated — as often — over acquisition.

Sounds understood that in its early days. The social media platform for music-sharing with 8M users built its product and strategy around high retention goals. We met Rémi Louf, Chief Data Science Officer at Sounds with a background in research, statistics and data science. His result-driven approach and appetence for growth-hacking are very welcome to take a deep dive into retention.


The first rule of retention is to have a clear value proposition from the beginning. When you build a social media platform around music, your priority must be shares and interactions — not streaming. “Spotify has tried to go social but it didn’t really work. Streaming music is a solo experience: it does not matter whether your friends are with you or not”, Rémi explained.

This is why the music player is not a central feature within the app. Music is a trigger for social interactions. Rémi added: “Snapchat and Instagram rely on photograph to connect people while Whatsapp and Messenger use text. At Sounds, we connect people with music”.


Sounds’ main screen is a feed that you scroll to see what people shares, and interact with their content with classic features (like, comment, share). The possibility to share content on the major social media platforms such as Facebook and Instagram is a huge acquisition lever. Why? “Thanks to the Sounds watermark added to every piece of content coming from our app. It works wonder with Instagram: still our main channel of acquisition today”, Rémi added.

Retention is the only growth metric that matters, according to Rémi. “You can buy acquisition, not retention”, he added. Plus, acquisition efforts without a solid retention boils down to fill a bucket full of holes. When you take 10k users acquired and a 80% monthly retention, you only have 700 left at the end of the year. But with a 90% monthly retention you would have only needed to acquire 2500 users to get the same end result. Hence the importance of creating an exceptional experience for new users who open the app for the first time after downloading. Retention is intrinsically connected to product development.


Many platforms (especially with social media) share the same goal: becoming an app that people use on a daily basis. At Sounds, the killer metric is called n-day retention: it is the % of people that were there on day one that came back on a specific day. Rémi added: “The goal is to have a curve that stabilizes and then goes up again (sometimes called a ‘smiley curve’), which is an indicator that users start coming more often as they discover the app.”. The other main metric they use is called “stickiness”: it measures the % of users on a given day that used your app the day before. “N-day retention is somehow about the future, while stickiness is looking at the past: they are complementary metrics”, Rémi explained.

Sounds’ strategy against churn is to focus on user experience in order to create daily habits. “Thanks to our data, we know that a user who has shared 10+ pieces of content is 80% likely to become a daily user”, Rémi explained. Ideally, you would need to cluster users depending on what they do in the app in order to understand what habits increase retention. Then, you should look at retention and stickiness metrics for each of these clusters.

If you don’t fancy clustering techniques, you can try something easier. Let’s take new users who shared 10 pieces of content in 5 days. Among them, how many are still active users on a given day? Conversely: among the active users on this specific day, how many did 10 shares in 5 days when they began on the platform? The next step is to find the actions that will optimize the result at the intersection of these two sets. For this result is the path you want your users to take.


The Sounds data team relies on two types of data:

Implicit data is extracted from the analytics database: every action by the users is tracked and recorded. Quantitative results from A/B tests are simple but essential for decision-making and product development.

Explicit data leads to qualitative insights directly provided by users. Again, the hack is simple: Sounds has a chatbot (Julie) that the team members use to chat with users. “Beyond metrics, it is essential to talk with your users to get different kinds of feedback on your product; metrics give you the state of affairs, qualitative feedback tells you why. At Sounds, a whole day is dedicated to customer support per week,” Rémi added.


Product development always requires teams to set up their own processes. At Sounds, each update is an experiment, and the team systematically spends time formulating its goals and the hypothesis behind it. To avoid future bargaining, the team debates the metrics that will be used to determine its success, and the result to achieve in order to implement the feature. Only after this, you code. “We usually roll our tests on 10-20% of the users. At the end, we look at the data: if you have reached the initial goal, it is validated. Otherwise, it doesn’t matter whose idea it was. If the results are not conclusive, we ditch the feature. No arguing.”, Rémi clarified.

A common mistake made by teams around the world is to consider that data is for data teams only. Rémi uses a great tool for product analytics and data visualization called Amplitude. “We are a data-driven team: everyone has access to the data it needs on customised dashboards. Developers often come to me with interesting insights!”, he added.

On the tech side, Sounds has got a lot of legacy code written in PHP because it made sense a couple of years ago. Lately, the startup decided to refactor everything by building microservices in Go. Rémi precises: “The feed and messaging features are microservices: we intend to do so for everything”. Analytics still relies on Amplitude, a Saas company specialised in analytics, but are soon to be migrated to a in-house solution built on top of BigQuery.


Sounds is a small startup with only tech & data experts. A short-term goal is the creation of squads: agile pluridisciplinary teams working on specific projects during sprints. Companies like Spotify managed to scale this groundbreaking product development methodology with the help of agile coaches. At Sounds, the idea is to gather backend, iOs and Android developers with a data analyst. One squad would operate on a specific block of the app. Everyone in a team has a growth mindset, and one squad will be dedicated to the onboarding and growth. “We learned a lot about growth and network effects from James Currier, seasoned entrepreneur and Managing Partner at NFX, and Stan Chudnovsky, partner at NFX and Head of Product & Growth at Facebook Messenger, after getting in the NFX Guild,” Rémi explained.

Whatsapp is often cited as a model for small teams and Sounds is no exception. The startup intends to stay below 30 people to preserve its culture and agility. The team has set the bar high when it comes to retention, with a lot of benchmarking of the other social media platforms’ performance. The results achieved so far are impressive: “Facebook and Whatsapp are unrivaled on retention. But we are close to that of Instagram: we actually have a higher stickiness”, Rémi revealed.