Over the last year or so, there has been news about how YouTube’s recommendation algorithm, which is designed with the goal of maximizing users’ watch time, pushes users who spend enough time on the platform toward progressively more controversial or extreme content. For the purposes of this exercise, let’s assume that this is an unintentional “by-product” of YouTube utilizing user watch data in a content agnostic manner.

Here are some potential solutions and their associated trade-offs: Update content policy and let users be first line of defense. Right now, YouTube has guidelines for certain kinds of content that are not allowed on its platform, which it could expand to include unverified content like conspiracy theory videos. There are at least two main challenges in this regard: first, in order to be an effective update of its existing reporting mechanism, YouTube would have to educate users about what it’s looking for. This could result in an uptick of reports across the board and it could take some time to see an impact on unverified content. Second, YouTube would be proceeding further down the road being an arbiter of the content uploaded to its platform and it may see backlash from some users (a la Facebook).

Update recommendation algorithm: There are two ways I can imagine YouTube might go about this:

• Focus primarily on reducing the number of complaints. First, build a data profile of a user who is likely to complain about controversial content. Then, build a profile of a controversial video. Then, update the recommendation algorithm to avoid recommending videos in the latter category to users in the former group. This requires a degree of fairly involved data work and retooling of the recommendation algorithm, although YouTube’s data team may have the foundations for these profiles already.
• Fundamentally alter the algo: Stop baiting users into staying longer through some kind of progressive escalation; only feed this content to users who deliberately search for certain, category-specific things (like “craziest conspiracy theories”). Anticipate and then closely monitor how this affects watch time. This could represent a significant strategic pivot and would likely require intensive research into alternative methods of getting users to stick around.

Do nothing. YouTube’s goal is still to maximize watch time and it is willing to accept and manage the fall out. As this requires no change to YouTube recommendation engine, this is obviously the easiest solution to implement from an engineering standpoint.

Prioritization: The above approaches are in priority order. Assuming for the sake of the exercise that YouTube cares about addressing this problem, I would proceed from user reporting of content to each algorithm update in succession after tracking the impact of each previous initiative. In the first case, some layer of programmatic logic would likely have to be implemented to separate the “wheat from the chaff” of customer flags to make the volume of flagged videos manageable.

Customer feedback is one of the important ways to keep track of product current health and identify features and strategy for future use

The approach would be to analyze the feedback to identify the potential causes for negative user feedback

Data will be captured through multiple resources such as Appstore Ratings, Personal Interviews, surveys, etc

I will start by analyzing data to get answers to the following questions

1. Is it on Mobile app, website
2. Is it specific to a device
3. Is it specific to an OS
4. Is it specific to a certain region
5. Is it been reported by a certain age group or during a certain time
6. is it related to user experience
7. Is it related to Online harassment
8. Is it related to content

Going Back to Youtube Mission Statement

Our mission is to give everyone a voice and show them the world. We believe that everyone deserves to have a voice and that the world is a better place when we listen, share, and build community through our stories.

My Priority of above fixtures will depend upon two factors, impact on engagement and retention metrics from negative publicity and Youtube mission statement

I will divide the above issues into technical and nontechnical Buckets

Points 1, 2,3,6,8,can be taken into technical bucket and rest can be taken to Semi-Technical Bucket

Out of technical buckets features

Priority will be given to User experience as bad user experience hampers the mission statement and

Impacts negatively Retention and Engagement metrics the 2 important metrics that determine the health of the product

Now User experience will be further breakdown to narrow down the concern

is it

9. UI Issue
10. UX Issue
11. Backend issue
12. Security Issue
13. Performance-related

If it is extremely urgent then we will collaborate with the scrum team and will go for the hotfix, else

will create an epic: Improve the user experience

Brainstorm with the scrum team to come up with user stories linked to epic

Collaborate with scrum team to find the Optimal solution

Will share user pain points in Current user journey with designers if possible will also share wireframes or sketches

Will share Functional flow and acceptance criteria with developers and QA

Will assist in answering all the queries team have

Once the solution has been build

Will do A/B testing with 30 % of overall users

Half will be shared new version and half the previous

Optimizely or some other tool can be used for tracking

Criteria for Success Retention rate after new update compare to before the update

3 Possible scenarios

Scenario1 feature is well received, will appreciate team and incorporate new learnings and launch this feature to the whole user base

Scenario2: Feature is moderately successful, will evaluate and brainstorm to consider long term impact before releasing

Scenario3: Feature failed, will take ownership of the failure , will take the learnings from the failed project and will start all over again

YT’s mission is to give everyone a voice and show them the world. So when a user is sharing their voice with us in the form of a feedback, I would receive it with open mind. I would first determine what the review is about: 1. app/website’s performance 2. quality of content they experienced 3. complain about YT being addictive and causing them to lose time 4. difficulty in using the app/website (search, navigation, recent changes negatively affected their experience) 5. complain about other YTber’s behavior on the platform (harassment on YT platform or on other platforms like Twitter/Insta/Facebook).

Then I would prioritize it based on the severity of its potential impact on the User. #5 being the highest priority. Here are few options I am considering for #5: 5.1 immediately start looking into the case, 5.2 resolve it based on the platform policies, 5.3 reach out to law enforcement if needed, 5.4 reach out to other platform’s teams focused at stopping harassment and 5.5 reach out to the user either with updates on our action or to seek more details or just to empathize while we are working on the case.

Next priority will be #2 addressing quality of content – even though every content uploaded on YT is reviewed by ML algo for copyright infringement and now they are expanding it for hate speech and child safety, currently they rely on community to flag content not appropriate and have their team review it and take action. I believe there is a process in place to acknowledge the report of bad content and a follow up on action. If the user’s complain is not solved, I would arrange follow up with user. If their reported content is still on the platform and didn’t seem to violate the policy then we can explain them the policy again and thank them for their effort to keep YT content clean.

Next priority will be #3. I am not sure if YT has any feature of disabling the Autoplay after certain minute of watch in a session but I would consider such feedback for a future improvement to help people who might be addicted to YT watching and feel bad about wasting time… In this case I would add such complaints in the list of incidents when people are complaining about being addicted to the YT and los of productivity. Other area where such feedbacks can be helpful is for Recommendation Algo. At min I am going to acknowledge the issue and keep such users informed about our action on this.

Next will be #4 and #1: I believe the performance issue will be surfaced through my regular health check metrics if it is wider. If it doesn’t exist already I can set an auto reply to such users with acknowledgement of complain and if we are aware of any issue or we believe it’s related to other factors like their internet connection and/or device. And for #4 with each feature change there will be strong focus on user feedback and complaints around it, so I believe such complains reflecting on recent changes will surface through that channel. If the complain is not associated with any recent feature changes then it becomes a source for user experience research and improvement initiatives including our help section. At min I would communicate with the user thanking them for feedback and our commitment to improve their experience on the platform along with links to help videos.

In cases where I have proposed reaching out to users – I would need efficient tools to respond to such users’ complains which means I will also prioritize tools for auto response to users based on their complain, surface the issue that needs more attention like further review, human intervention etc. and schedule follow up call as needed.