Challenge: When users initially subscribe to Blinkist, they are highly engaged, motivated, and are often power users—however, as they progress further down the lifecycle, they often become more casual users or even dormant users. This makes them a high churn risk, since less active users are substantially less likely to renew their subscription.
In order to prevent this and increase YoY revenue, Blinkist needs to ensure those users remain highly engaged by keeping them active in the app. This means, ensuring they continue to discover fresh content (Blinks, Shortcasts and Collections) to engage with. This content needs to be of high relevancy, quality, and we need to ensure the user has context to the recommendation. We had also seen positive indicators for a more regular cadence, shorter form emails, and current events-based content. We want to lean into these positive signals in order to retain users through greater relevancy in content recommendations and creating a higher frequency of usage. Previous tests and user research showed an appetite from users for content related to current events and news topics, as well as that if the recommendation is the right one (i.e. useful, relevant), an openness to receiving more comms.
We anticipated that a campaign which fulfilled this criteria could increase power user activity by as much as 15%, with extremely positive knock-on effects on revenue With over 23,000,000 users in the app, the impact could be extremely positive.
Solution: Using AI in combination with user location to connect books from our library with the real life events happening around the users.
How does this work? Together with the Data Science team and an AI copywriter called GT3, we created a fully automated campaign called the Blinkist Signal. news-based, localised, personalised recommender. Title recommendations would be provided based on a) trending local news, b) the users’ location, and c) users’ behaviour. Then the AI writes the copy used in the campaign to explain and contextualise why this book is relevant to that event.
We would then use connected content to draw data about the title from our backend, and populate the title information into the email such as the cover, the url, and the title. This is set up within a canvas with users’ being sent title recommendations daily based on this logic. Such localised, personalised and unique recommendations would not be possible to our large user base without the use of this technology.
Finally, to make sure the machine and AI are smart, we tested this against a control group and against a less personalised recommendation made each day by our in-house editorial team.
Outcome: The Signal outperformed the human made recommendations by 10% in terms of immediate content engagement. And we also saw +15% users saving the title to listen to later. Our analysis showed this campaign is a clear success in creating high intent for users to engage with our app and content. The campaign is still entering next steps and iterations, but initial analysis suggests it’s especially useful for dormant users—who are showing strong reactivation rates.
Learnings from this campaign will be used to inform our engagement strategy, and show positive indicators for more news based content pairings, regular cadences for content recommendations, shorter form emails, a mixed automated and editorialised strategy, and a user appetite for more automated scaleable content using tools like GPT3 to augment this.
.png){kind=link}