Podimo: Championing local content and diverse voices with AI-powered recommendations
About Podimo
Podimo is a local-first premium audio entertainment subscription service that offers original and exclusive podcasts ad-free, an extensive collection of audiobooks, plus podcasts from around the world via RSS feeds. The company creates original content that amplifies the voices of creators from around the world. The app is currently available in seven markets, with ongoing expansion plans.
Tell us your challenge. We're here to help.
Contact usBuilt on Google Cloud, Podimo is creating a personalized, contextualized experience for listeners with an innovative recommendation service for podcast discovery.
Google Cloud results
- 3% lower churn rate for personalized playlists with models trained on Vertex AI compared to manually curated counterparts
- Managed services free up teams to focus on product development, leading to 50% increase in subscriber base year on year
- Notifies relevant team members if infrastructure issues are detected with Cloud Monitoring, so errors are fixed quickly
- Consistently meets average service delivery and response times under 100 ms, ensuring listeners don't experience latency delays
Playlists curated with Vertex AI models perform 15% better
Podimo was founded in Denmark in 2019 and is quickly becoming one of Europe's most popular audio content subscription services. Its goal is to champion local audio entertainment and diverse voices while solving common content discovery challenges for listeners.
"Our primary focus in selecting Google Cloud from inception were the core features we wanted to use as SaaS such as Google Kubernetes Engine. We knew from experience that Google Cloud had these features as well-maintained managed services, freeing up our team to focus on developing Podimo’s offerings."
—Benjamin Biering, Global Director of AI, Podimo"With more than 2 million podcasts on the market, it can be challenging to know where to start," says Benjamin Biering, Global Director of AI at Podimo. "Content platforms often offer recommendations which can be biased towards popularity, but then new or niche podcasts may be overlooked. At the same time, not every podcast is for everyone, and investing time in listening to a new one just to find out it’s not for you can feel discouraging." Podimo is on a mission to solve this discovery problem and remove friction for users through personalization that leverages a blend of human curation and AI.
To support this vision, the cloud-native company, which has been operating on Google Cloud since inception, selected a range of native features to operationalize and fine-tune Podimo's core services.
"Our primary focus in selecting Google Cloud from inception were the core features we wanted to use as SaaS such as Google Kubernetes Engine," Biering explains. "We knew from experience that Google Cloud had these features as well-maintained, managed services, freeing up our team to focus on developing Podimo's offerings."
Building the foundation of Europe’s fastest-growing podcast and audiobook subscription service
Podimo's ecosystem of technologies is primarily supported by Google Kubernetes Engine due to its autoscaling features and easy integration with other Google Cloud solutions, such as Pub/Sub, which supports Podimo's event-based architecture. "The way we communicate data between teams is largely orchestrated by Pub/Sub," says Biering.
"Podimo is Europe's fastest growing podcast and audiobook subscription service, which means we've been making the right decisions so far, not least our selection of infrastructure services that allow us to focus on further improving and expanding our product."
—Benjamin Biering, Global Director of AI, PodimoPodimo stores all media, data, and ML models on Cloud Storage, and serves all media to subscribers using Cloud CDN (Content Delivery Network), which accelerates web applications by using the Google Cloud global edge network to bring content closer to listeners across the seven markets where Podimo is present.
Because customer experience is always front-of-mind for Podimo, one key measure of success regarding this setup is how responsive Podimo's app is for listeners. The company measures this using client-side reporting to check what the average service delivery and response times are for its listeners across Europe. "The industry standard average is of 100 ms and maximum of 200 ms response time," explains Biering, "everything above that is a warning, but we never had to worry about that. We never had latency delays."
Meanwhile, to ensure all is well internally, the Podimo team uses Cloud Monitoring to continually check on the performance and availability of its infrastructure. "With Cloud Monitoring, errors are caught quickly so we can fix them fast," says Biering. They use it to track potential errors in Google Kubernetes Engine and Pub/Sub and are notified of any issues through personal messages via channels, such as Slack.
Using Cloud SQL as a fully-managed database service, Podimo maintains and manages its PostgreSQL relational databases on Google Cloud. This is where the company stores its backend data on MySQL and Postgres servers. "Using these managed services we can leave the infrastructure and database management to Google Cloud and focus instead on developing our application," explains Biering.
With popularity and rapid growth as sure signs of success, Podimo is confident that it has made the right decisions. The subscription service has grown its subscriber base by more than 50% in 2023 alone. That year also showed a surge in global user engagement, with listeners amplifying their time spent on the platform to an average of 20 hours monthly. "This suggests Podimo is Europe's fastest growing podcast and audiobook subscription service, which means we've been making the right decisions so far, not least our selection of infrastructure services that allow us to focus on further improving and expanding our product," says Biering.
Creating personalized experiences for listeners with AI-powered podcast discovery
Next, to maintain this success and take it one step further, Podimo is focusing on AI. It uses Trino on Google Kubernetes Engine to combine databases into a single source as a data warehouse so that AI and Data teams can query through one source. Similarly, BigQuery serves as its BI data lake, used in conjunction with Looker to offer a one-stop shop for business intelligence insights. BigQuery also serves as the database supporting Vertex AI Feature Store for AI/ML models because of its easy integration with other Google Cloud services, such as Vertex AI, which is where ML/AI product development and serving takes place.
This combination supports Podimo's needs for sophistication and scalability when it comes to AI products developed in-house. "Our recommender models used for the different purposes in the app have been trained on Vertex AI and then deployed on Google Kubernetes Engine. We're leveraging Google Kubernetes Engine autoscaling for managing the inference load and messages orchestrating the inference process," explains Biering. "This infrastructure allows us to use different recommendation strategies for each purpose in the app. In the same environment, by only changing the configuration settings related to processing and memory, we can instantiate simpler strategies like content-based, or even a more complex one that uses a deep learning-based model."
Rather than applying a one-size-fits-all strategy, Podimo's recommendation service is composed of several recommender models, depending on the use case and media to be recommended. "We use a number of features from podcasts and users to be able to compute these recommendations," explains Biering. "We prototyped these models on Vertex AI where it is very easy to run a number of experiments using GPU compute in parallel and select which combinations of models and features perform best."
Biering explains that this ability to train different models for serving different recommendations allows Podimo to select the best strategy for each user, so listeners increasingly benefit from the recommendations in the long term. "For example, new users may be more attracted to popular and new content than experienced users. So it's all about trying to meet our listeners where they're at. Minimizing the friction and effort to discover new and highly-relevant content is really what it's all about, so that we can deliver on our promise to make Podimo your preferred audio companion."
Podimo used Google Kubernetes Engine to make these models live, as it grants the team freedom to configure the recommender service specifications as they see fit. Podimo serves recommendations in two ways. First, it computes new personal recommendations to every user daily and makes this data available to the backend. Secondly, Podimo exposes its recommender service as an API so that every collection of items in the app can be re-ranked according to each user's preferences. "In that way, we can start improving the relevance of the content our listeners are exposed to in a much more holistic and pervasive way," Biering explains.
As a result, looking purely at conversion and engagement metrics, Podimo noticed that personalized collections and playlists generally perform 15% better than manually curated, one-size-fits-all counterparts. Additionally, churn rates can be reduced by more than 3% just by exposing users to personalized content.
From a user experience perspective, Podimo increased recommendations diversity by up to 30% in the past year. "We believe this is essential to help our listeners discover new and different content, avoid echo chambers, and keep our listeners engaged and loyal in the long term," says Biering.
Podimo is further investigating how to better contextualize its recommendations. This includes adapting recommendations to the listener's time of day and activity, while also providing clearer and more insightful explanations about why Podimo thinks a given piece of content may be relevant for this user. "This has to do with the objective used to train the recommender model," explains Biering. "We now know that objectives purely based on engagement do not produce the best business results in the long term. I'm a firm believer that giving more control to the users will yield higher satisfaction and ultimately lead to more trust and transparency."
Standing out with relevant recommendations delivered in an improved interface
Against the backdrop of a competitive industry landscape, Podimo believes that the way to stand out and continue growing is to serve relevant recommendations alongside context, transparency, and logic so users understand why they're seeing what they're seeing.
"At Podimo, we envision a one-stop shop service for all things audio, a true companion that meets you where you're at with the right content at the right time. As a data-native startup with solid foundations on Google Cloud, we're in the best position to deliver on that promise."
—Benjamin Biering, Global Director of AI, Podimo"We know from movie and music streaming services that personalization is a must have. Users expect this. But I'm not particularly impressed by how the competition is delivering this," says Biering. "Some big industry players have developed advanced technologies for personalizing every aspect of their services, but it doesn't always result in a superior user experience, particularly if the users are unaware of the logic behind what's being recommended."
Biering believes that what LLMs are bringing to the table, or are at least facilitating, is the possibility to provide users with contextual explanations around the recommendations. "For us, improving the UI to offer this complete package is an opportunity to differentiate ourselves from the competition," he explains.
Next, Podimo is taking its model further with a focus on ever-more exceptional "hyper local" spoken word audio that entertains and inspires. That will include producing more content in different languages, and providing more localized information to more people. "At Podimo, we envision a one-stop shop service for all things audio, a true companion that meets you where you're at with the right content at the right time," says Biering. "As a data-native startup with solid foundations on Google Cloud, we're in the best position to deliver on that promise."
Tell us your challenge. We're here to help.
Contact us
About Podimo
Podimo is a local-first premium audio entertainment subscription service that offers original and exclusive podcasts ad-free, an extensive collection of audiobooks, plus podcasts from around the world via RSS feeds. The company creates original content that amplifies the voices of creators from around the world. The app is currently available in seven markets, with ongoing expansion plans.