Design a product for Podcast recommendations.

Candidate: In today's more and more people are producing podcast content, thus it becomes increasing difficult to sift the entire collection and find the podcast that the users would enjoy. I assume the goal is build product that let users enjoy the podcast recommendations.

Interviewer: okay

Pain points

1. Users are unable to discover the great podcasts on their own that they would enjoy

2. Users spend lot of time sifting through large number of catalog to find the podcast to find the next one

3. Without personalization engine, when users search for a podcast, users run a risk of bumping into a click baity /racy/violent/hateful podcasts content that leave the users dissatisfied

3. Without ranking algorithm, users might be exposed to less podcasts from less trust-worthy and less reliable sources that make unproven/misleading claims

Solution

In order to solve these problems, I would build a recommendation algorithm that decides what podcasts to show to on homepage, listen to next podcast, and search.

What podcasts are recommended to user would be based on the following factors:

Podcast performance: Initally, I would surface a padcast to small number of users, then I would use click through rate, average listening duration relative to the podcast total duration, to decide how widely it will be distributed to other users

Personalization signals:

Since there is not peronal data availble yet for the first users, I would show what is popular in the user area.

Once the users started to use the app, we can get personalization signals to understand the user preferences from the topics that they click, pod cast makers that they click/follow, podcasts that they like, podcasts that they dis-like, podcasts that contribue to longer listening duration, podscasts that users liked based on the survey etc.

Related podcasts: If there are existing set of users that watch podcast a then b, I would use that as signal that podcast b is related a, and use that in the recommendation engine

Seasonal factors: I would factor-in the seasonality in the personalization algorith, and recommend the podcasts related to skiing during winters and podcasts related to camping in fall

Borderline content:

I would demote the content is racy/violent/hateful in the recommendation list

Trustworthy vs. Misleading content:

I would promote podcasts from trustwothy and independent organizations and I would demote podcasts that involve in mis-information compaign.

Design of the podcast recommendation app

1. I would create a homepage that includes the recommended podcasts

2. Once the user selects the podcasts there would be option to like/dislike/share/comment

3. I would deploy the personalization algorithm to recommend the 'listen next' podcast

4. I would use the recommendation algorithm to suggest the podcasts when users search for the podcast

5. I would create a playlist that users can use to binge on podcasts

6. I would create a feature that allows the users to send additional signals to personalization engine on the fly, this feature would include set of name cards that is shown on homepage and up next section. The name cards would include a) podcasts from same creator b) podcasts related to same topic c) recent podcasts d) mixed etc.

This feature would allow users to send explicit feedback signal such as do not recommend this channel, do not recommend this podcast, etc.

7. I would send a user survery occasionly to understand whether the user enjoyed the podcast that they just listened

Succes metrics

1. #users that listen to podcasts weekly, months

2. #hours that users cumulatively spend listening to podcasts

3. #average number of weeks a typical podcast users listen to podcast in a year

4. #ad supported or subscription revenue from users.

Leonardo Fernandes · Answer 2 · 2022-01-03T23:29:21+0000

1) Clarification

Is this product supposed to be inside the Facebook App?
- Yes.
Is there a focus on what type of content or geographical location?
- No, there is not such focus at this moment.
Is it only for recommendations or is it possible to listen to podcasts?
- Your call.
Is it for web or mobile devices?
- Preferably mobile devices.

Great, so just rephrasing the statement. We are trying to design a product for podcast recommendations inside Facebook App and focused on mobile devices.

2) Goal

I believe that this new product is very well-aligned with Facebook's goal, which is to bring people together all over the world. By recommending and talking about podcasts, people with the same interests can interact with each other and foster even more the community.

Because it's a new product, I would like to focus on acquisition and engagement. How does it sound?

So, here's how I would like to approach this question:

- Talk about some user groups;

- Choose one group to focus and mention some of their pains;

- Based on that, we are going to brainstorm some solutions;

- Then, prioritize one or two solutions;

- Finally, define some success metrics;

3) User groups

So, I am going to focus on the listener side, not the creator one. I could think about some types of user groups.

Power Users
- These people listen a lot of podcasts. They probably already use other tools to find podcasts, such as Youtube, Iphone Podcasts or Spotify. They tend to spend hours per day listening to podcasts.
Regular Users
- These people listen to podcasts by squeezing them in their schedule, like going to work or taking their children to school. They do not necessarily listen to podcasts everyday and focus more on specific topics inside each episode.
Common Users
- These people rarely listen to podcasts, generally before going to bed or on weekends.

Based on those groups, I would pick Regular Users, because I believe there are opportunities to leverage Facebook platform to bring them to Podcast's world. Does it make sense?

4) Pains

Regular Users - Pains

Hard time finding content to listen and/or watch
- Do not know how to search and where for instance
Do not have time to watch an entire episode
- Squeeze the podcast in their routine is complicated
Do not know where the interesting content starts within the episode
- They do not want to watch the entire episode, therefore want to watch just 10-15 minutes
Have difficulty to differentiate good and bad content
- Hard time reading reviews, and understanding what a good podcast would be
Search for interests according to their life goals
- Let's say they are focused on transitioning careers, saving money, investing, trips, etc

Looking at this product, I believe we could tackle those pains, because they are linked with one another. Let's see how it goes, what do you think?

5) Solutions

After taking some time here, I could come up with one more detailed solution

An app that users would insert some key information about their interests, and an powered-AI algorithm would recommend them the podcasts
- Impact high: it's what the biggest companies use to recommend content, such as Spotify, Netflix, etc.
- Cost high: it's very hard to design an efficient algorithm to recommend content
An app that users would ask for recommendations by using like a "survey" and the community would recommend them the podcasts
- Impact medium: to have a content, the users need to take action and hope that the community answers him/her
- Cost medium: not so difficult to organize this "survey" and community
An app where the user would select a specific topic and/or section, and he would be able to listen for 5 minutes of that content and decide if he wants to listen more. It's like a "Tinder" for podcasts, and instead of "Like" would be "Listen More", something like that. Creators would cut 5 minutes of their podcasts as they upload to the platform, and it would like the entrance door for listeners.
- Impact high: Even though it's not so well sophisticated as the AI one, this idea offers the information users need. AI could be the future for this one.
- Cost medium-high: It's a lot of information to put together, but it's very possible

Based on those impact X costs, I would go for the third idea, and also because it's different. It should look like the following:

a) Users create a profile at the "Podcasts page";

b) Users complete some information on their profile

b1) Interests (work related, hobby related, sports, etc etc)

b2) Time available (5 minutes per day, 10 minutes per day, 20 minutes per day, etc)

b3) Language to listen the podcasts (native, english, etc)

c) Users are directed to the main page where podcasts sections would be shown

c1) Interests first sorted by time

c2) What friends are watching

c3) The best podcasts by review

d) Users select a section or topic they want to listen to. Let's pick Politics for example.

e) Users would be directed to a page like "Tinder" where they could listen to 5 minutes of that first podcast.

f) Users will be able to see "reviews", "comments", "reactions", "high-level summary", "what friends have watched".

g) Users can "Listen more" or "Go to the next one"

h) Users can "React", "Comment", "Share" that podcast in their Newsfeed/Stories

Some key points/assumptions

- Figuring out the time for creators would be paramount (5 minutes, 3 minutes, 30 seconds?)

- Users need to be able to take action by voice, since they generally are doing other things while listening podcasts

- It would open a new stream of revenue from ads and positioning!

- Recommendation system would be simple and basic at first, based on interests mainly. After some time, we can improve that to an AI algorithm and leverage data from users' connections and actions across Meta's products.

6) Metrics of Success

I could come up with some metrics based on our goal

- Number of profiles set up among users

- Level of engagement among users (reactions, comments, shares, reviews, etc)

- Avg session time spent on the product

- Avg watched time per user

7) Trade-offs

One trade-off that comes to my mind is that these types of users might not engage fully because of their habits, but they indeed are watching contents and interacting with podcasts. So, the level of engagement should be considered when analyzed.

Prayansh Ratan · Answer 3 · 2021-12-26T23:48:42+0000

Clarify scope -

What kind of product are we trying to build?
Is it like a newsletter that goes daily/weekly to subscribed users with top 10 recommendations?
Is it like a Twitter bot that tweets the links for these podcasts?
Is it an app or a website or some plugin that plugs to Spotify/ Apple Music/ Google Podcast chrome extension or something? If it’s an app what kind of devices are we targeting?
Will this make popular recommendations or personalized recommendations?

Going ahead with the assumption that this product is an app for iOS users to make personalized top 10 podcasts recommendations for them based on their interests

Goals -

So as per the assumptions, the goal here is to create an iOS app to make personalized podcast recommendations according to user interests

Users -

This application is useful for users of any age. Let me explain this -

Kids - Kids probably don’t have the expertise or the resources to find the right podcasts for them. There are very few lists that curate these materials that are ideal for kids. They need some application that tells them what is appropriate and educational for them.Millennials -

Millennials don’t have time to search and listen to every podcast to choose great ones in this fast-paced life, with their colleges to jobs they are occupied with. They want a ready-to-go solution to just open an app and listen to recommended ones.

Adults - Adults have jobs, and families to take care of. Some of them are parents. They don’t have time to go through every podcast and choose the right ones for them. They just want to start listening, that’s it!

User Journey -

The user logs into the app. If they are a first-time user then they need to sign up, using their email id/ phone number. They can one-click sign up using their Facebook/ Gmail.

If the user is not a first-time user they can simply log in using their email/ Gmail/ Facebook.

Once the user is logged in, if they are a first time user they are asked for their favorite genres, to connect their contacts, Facebook, Spotify, etc to recommend them accordingly

Assumptions -

The assumptions here are that the users are well versed with technology (at least enough to use apps) and they do understand their genres and are able to select their preferred genres.

Another assumption is that they have friends who share similar kinds of traits.

Pain Points -

Not having time to listen to every podcast and select to listen in the future.
Not knowing where to find these podcasts or which service to use to get these recommendations.
Parents worry about their kids listening to podcasts that are inappropriate for them.
Not knowing what’s trending or what other people are listening to, especially friends and family.
Sharing your recommendation as playlists/albums.
When learning a new language, one doesn’t know about good creators in that particular language, so recommendations are important.

Prioritization - Usually the pain points are prioritized, but since this is already a niche product and has a limited number of features, we can implement these. Had it been asked to design a product for podcasts or a general recommender system, we could have prioritized these pain points as to which pain points to solve.

Solutions -

Let users connect their Spotify/Apple Music/Google Podcast accounts to pull their most listened to artists/ genres and make recommendations accordingly.
When signing up with the app ask users to select up to 5 genres to get podcast recommendations from.
If the user doesn’t know their preferred genres, artists then show them the top 10 podcasts globally/ locally, and similar podcasts to what they already listen to by pulling their podcast app data (reference point 1)
Let users add languages of their choice. Users trying to learn another language won’t know who great artists in other languages are.
Let users connect their socials (Facebook, Contacts, Twitter, etc) to recommend podcasts from their mutuals.
Let them select what mutuals to get recommendations from (while connecting socials all data gets pulled in which means if you have saved a cab driver’s number back from 2013, the recommendations can be made from that, which makes no sense at all so let them select)
To share/recommend podcasts to friends on the platform by giving a nudge like “your friend [x] has recommended [y] to you.”
A module which contains all the trending podcasts, genre-wise trending podcasts, and global top 10, if in case the users want to explore podcasts outside their favorite genres
Parental control for the accounts of users who are less than 18 years of age allows parents to restrict the recommendation made to just the ones that are educational and appropriate.
Optional anonymous profile option (private profile) for users less than 18 years of age.
Ability to curate the recommendations in a bundle and share with friends on Facebook, Twitter, etc.
Ability to add a recommendation to favorites so that the recommendation algorithm learns more and provides better recommendations over time.

Prioritization - I’ll prioritize these features according to their Impact and Complexity and assign priorities for the same -

Feature	Impact	Complexity	Priority
1	high	Medium	p0
2	high	low	p0
3	high	low	p0
4	medium	medium	p2
5	high	high	p0
6	medium	low	p1
7	high	medium	p1
8	medium	low	p0
9	high	medium	p0
10	medium	high	p2
11	medium	medium	p0
12	medium	high	p1

The features have been prioritized according to their impact and their complexities. The p0 features are a must to have this app working in the right manner, p1, and p2 features can be added later to make the app more useful and provide more control to the users.

Success Metrics -

Success metrics for this app can be something like DAU/ MAU to understand whether users are using this app or not

Another metric can be something like that users listen to at least 20% of the recommendations made by the app.

Summary -

We have to design a podcast recommendation app. We analyzed the ideal user group for the product, understood their pain points, the potential solutions for these pain points, prioritized these pain points according to their impact and complexity, and defined metrics to measure its success.