How Calm is using machine learning to keep us all mellow

Gary Flood Profile picture for user gflood June 4, 2021
Meditation, wellness and sleep app Calm is using Amplitude to increase engagement and customer retention.

Image of the Calm app
(Image sourced via Calm)

With a mission is to make the world "happier and healthier" by helping people prioritise mental health and claiming to be the world's most popular sleep, meditation and relaxation app Calm exults in being what the Centre for Humane Technology dubs "the world's happiest app".

Designed to help users manage stress, sleep better and live a happier, healthier life, the app is available in more than 190 countries. But for all the apparent talk of "calm science," there's a lot of deep IT and data-driven marketing behind the way the privately held, nine-year-old company gets such high satisfaction rates for its meditation app.

Even more than that; there's a lot of machine learning going on behind the scenes at the operation, which achieved "unicorn" status with a valuation of over a billion dollars quite some time back. That's in the shape of its use - from its very first days - of a machine learning-driven marketing database provided by a company called Amplitude, which claims to offer a digital optimization service that can enable apps to perform better through AI.

One small change

Specifically, insights are gathered from analysis of "trillions" of customer actions, which are then translated into business recommendations for customers like Calm and delivers results based on behaviours. And in Calm's case, its Head of Lifecycle Marketing, Sue Cho, told us, by making one small change to its in-app daily reminders, customer retention was almost immediately improved by a factor of three.

Cho explained what she sees as her contribution at the company. She says: 

The role of lifecycle marketing is to convert new users, engage existing users, retain the users we have, and win back everyone that's churned. I consider my job part psychology, because it's really the study of human behaviour and how I can influence human behaviour to help with business metrics.

The specific business problem Cho says the software was installed to try and deal with was to find a way to accurately correlate which member behaviours were associated with engagement and retention. Then, as the company expanded its app and offerings from meditation into new areas including sleep, mindfulness, and then into wellness and education, management felt it needed a way to understand which specific behaviours of the app inside all these new categories was leading to the highest engagement and retention. Cho says: 

Our most engaged users come back year after year. But since we're a yearly subscription product, we really want to know what specific behaviours and sessions they are specifically doing that are associated with them coming back and paying for the app.

It seems that the vendor sealed the contract by showing how easy it could do this, at scale, when the supplier's CEO built a chart in just a few seconds that showed exactly which behaviours were correlated with retention. Even better for Calm's leadership point of view, the analysis showed a clear association with daily, recurrent content. Cho explains:

We saw people who did our daily Calm were 66% more likely to renew than people doing regular meditations and our famous sleep stories: the vendor showed us that in a beautiful bar chart in just a few seconds. But that's something that would have taken us working with the data scientist for at least two weeks to figure out. It was just so obvious that we needed this.

‘Holy moly, our daily content is what's driving engagement!' That insight informs the content team on what they should spend their investments on and what they should produce more of, while from the marketing side, it allows us to understand what types of content people are going to resonate with customers.

Mapping the Sunday Scaries 

A very strong start, but what does the use of the system look like in practice today, now it's part of daily workflow? Calm's user data is directly fed into the third-party machine learning system, including all purchase and behavioural data, Cho told diginomica. That data then allows her and her team to run complex what-if questions and ad hoc analyses very easily, she claims:

This is probably the most perfect product I've ever used - and they didn't pay me to say that, bypassing the need for building complex SQL queries. I'm not a data scientist, but the product is so intuitive and easy to use it pretty much allows me to select what activity is associated with the business KPI.

It's not just marketing that uses the system; Customer Experience and Engineering also employ its help, she states. The latter, for example, turns to it whenever a new release is pushed out against internal quality metrics so that step doesn't have to be done through coding. Cho says:

Across the company, we all get to save a lot of time. But as it's such an intuitive tool we gather a lot of insights from them. One was that what is bringing people into our app is actually our sleep stories, which are kind of bedtime stories made for adults. But the tool also told us that while that may be what brings people in, again, it's the daily content that makes them stay. I don't know if we could have arrived at that conclusion as quickly as we did without this software.

Another pattern the machine learning surfaces is what Cho describes as the cycle of Calm use. She explains:

You may have heard the term ‘the Sunday Scaries' - we see that every week through our data. Toward the weekend, usage drops starting about Wednesday, and Saturday is our worst day of engagement. Then on Sunday, we see a massive spike going up in usage and it peaks around Monday and Tuesday: people are freaking out Sunday night that it's the end of the weekend, and they need to get rest and they're stressed out.

Other usage patterns the analysis brings up is a regular daily cycle too, with a big bump early in the morning when people are doing the daily Calm meditation followed by a second but larger bump toward the evening around 10pm, when people are trying to wind down and go to sleep.

A definite COVID-19 usage bump

In terms of next steps with the system at Calm, Cho says that she's interested in the vendor's roadmap of moving from being a product team tool to being more of a marketing analysis analyst service. But we also asked if there had been any specific impact of COVID-19 on use of the system at Calm, and we got this rather interesting insight into how we've all been doing this past crazy year. Cho says:

From a business perspective, COVID-19 absolutely had a major impact on Calm, and for the good: people were stressed and anxious, and companies didn't know how to support their employees as they shifted to work from home. We saw numbers triple during that period of time, and our B2B business just went bonkers.

Plus, not only did organic signups and referrals increase, but we also saw an increase in guest gift card purchases, which we only see during the holidays, showing people were really trying to reach out and help each other.

That was pretty cool-and that's why I like to say Calm is using real data to influence better habit formation for the better of mental wellness.

A grey colored placeholder image