So, what’s your favorite product?

A classic product management question — one that helps interviewers filter candidates in the best possible way. I have personally spent hours on this question — read blogs and spoke with product managers in an attempt to hit that sweet spot. But all that research led me to the same set of answers:

· There’s no right or wrong answer here.

· It’s a test of your creativity, structured thinking, competitive awareness, technical abilities (to some extent), and leadership (yes, whiteboarding in front of the interviewer makes a difference even on a video call).

· You don’t need to go out of your way to choose a unique product to distinguish yourself. In fact, it may backfire if the interviewer has no clue about the product and you couldn’t explain the product vision effectively in the first place.

· It’s absolutely essential to justify why the product you chose is your favorite. It’s obvious that YouTube will have more takers than Splitwise for a multitude of reasons including but not limited to scale, target audience etc. But would everyone’s reason of choosing YouTube as their favorite produce be the same? Highly unlikely.

Okay, let’s walk through a question. But before we dive in, please know that this is completely my perspective and in no way any industry standard. I’ve had questions from many people asking how this should be answered and this is just an answer, not a way to guide you towards a particular approach.

Enough said. So, what’s your favorite product?

My favorite one has always been YouTube. It’s a unique social platform. At times, my daily life takes serious toll on me, but I still won’t have enough time to sit through an entire movie. I need momentary relaxation and that’s what YouTube is for — short snippets of videos that allows me to consume exactly what I want. I’m a huge stand-up comedy fan, and YouTube is world’s biggest repository of user-generated stand-up comic content that doesn’t last more than 20 minutes each. Great way to enjoy, relax, and back to work.

Perfect. Can you suggest any improvements?

Sure, but first let’s make sure we’re in the same page. YouTube has several models — Premium, Music, TV, and the ad-supported version. I have been consuming the ad-supported version long enough to comment on it although I have a fair idea on what the others are meant for. Let’s move ahead with the ad-supported version for the time being because it has ~1.5B users. That scale can let us do wonders with the product.

Now let’s understand what “improvement” means. I can think of numerous goals that we can set and then work towards it. We can increase user engagement, we can try to retain more users thus reducing churn rate, we can try expanding to newer geographies, or we can take a direct shot at revenue. But engagement would be more interesting than others for a few reasons:

· YouTube is a social platform built to engage users. More engaged users automatically decrease churn.

· More engagement could come in multiple ways — spending more time per video or hopping through multiple videos. Either way, it improves YouTube’s recommendation engine and in turn helps with caching content. (If you don’t know what caching is, look how and why YouTube caches content across its CDNs). That, in turn, saves space by eliminating irrelevant caching and improves delivery.

· More engagement improves the Ads engine too, effectively making it more relevant and less of a friction. This “could” entice users to turn off their ad blockers and helping YouTube generate more revenue.

So, by hitting engagement, we’re essentially improving the other metrics too.

Now that we’ve established we’ll be optimizing YouTube’s engagement, let’s see where we should test it out first. Given the scale, we won’t launch anything globally in the first iteration. We need to narrow it down to a city if possible, but a country is still fine.

We can segment based on YouTube’s user base: US, Europe, India, APAC (Excluding India), ROW. India has a huge user base comparable to any country, and thus deserves to stand alone.

To prioritize among the regions, we can look at metrics that tie back to the original goal. Let’s say we prioritize based on:

· Number of users.

· Total watch time per user.

· Number of users using ad blocker.

· Churn rate.

This would help us point to one region. Try doing a very crude estimation here with the following parameters:

· Total population

· % of population with internet

· % of population that can afford video streaming (i.e.; YouTube users)

· Estimated watch time per user, ad blocker usage, churn rate.

Let’s say we narrow it down to the United States. Now the question is — what’s our engagement improvement aim and within what time frame? A qualitative discussion with your interviewer should help set a goal here.

So finally, we have established a goal: Improve YouTube’s user engagement within the United States by 20% (assume) and achieve it within a quarter (assume).

Let’s look at the YouTube’s user clusters now. On a high level, YouTube has direct consumers, content creators (individual) and businesses. We can choose to improve for any one of them, but should we? What value does increasing engagement for business bring to YouTube? They would upload their content and expect end users to either purchase (for paid content) or consume (for free content). Same applies to content creators although we can argue that the individual content creators would have followers that would engage more with the platform if we help content creators streamline their delivery. True, but to what extent? Are they enough to achieve the goal? Again, we can prioritize based on follower base and engagement per follower if we choose to go ahead with the content creator persona OR choose the consumers if they are in higher numbers with improvable engagement time. Again, doing a mix of qualitative and crude quantitative discussion helps.

Let’s say we go ahead with “Consumers” for the above reasons. We now dig a level deeper and see the individual user persona here:

· Kids

· Teenagers

· Young professionals

· Soccer moms

· Older people

Although it might seem that I have segmented solely based on age, I’m also thinking about their available time to spend on YouTube, their interests, their knowledge of ad blockers, their probability to buy things if the Ads engine is improved, and their expected CTR on ads.

We can use the above table in multiple ways, one of which is doing sort of a conjoint analysis. Put a number between 0 and 1 to each attribute to the left and the total should add up to 1. Next, put values to each category (i.e.: 0.1 for Low, 0.4 for Medium, 0.5 for High). Leave out categories that you can’t do a quantitative analysis with (e.g.: Age). Then add them up and choose the persona with the highest score.

Let’s say Teenagers score the highest. (Another way to prioritize is to do a qualitative analysis — talk through each attribute and say why you’d choose one over another. E.g.: We can expect Teenagers to have a higher CTR if they see relevant ads but are very susceptible to churn and are least brand loyal. So, we need to hit the bull’s eye if we chose to go with them).

Now that we have a persona, we work through their user journey on YouTube. What do they do on YouTube?

· Watch movies

· Watch original content

· Engage through likes/Shares

· Scan the recommended videos and watch if they find anything useful

· Search for content using the search bar

· Add videos to watch later

· Hovers on the thumbnail to get a sense of what the content might be like

What are some of the pain points they suffer during their journey on the platform?

· They don’t know the quality of the video upfront (no aggregate reviews on the recommendation panel)

· They don’t know whether it’s a clickbait or a real video

· Let’s say they’re interested in clothing and accessories and see one in the video. They don’t know how and where to buy it from.

· They don’t know the name of the accessory they’re interested in.

· They don’t know how famous/great the content creator is, and how relevant their videos are until they click on it.

Strategyzer is a great tool I use to come up with the above laundry lists of journey, pains and gains. Below are some snapshots:

Now, we can’t solve for all. We must choose one. How do we do that? There are multiple ways. Think from the user’s perspective what makes more sense to them? What are they most passionate about? What other things (apart from video) can they potentially hire YouTube to solve for them?

This is where a deeper understanding of the persona helps. Let’s say from user research, we have understood that our target profile (it’s a further segmentation within the “teenagers” persona) shop heavily online (from browser history, GPay usage etc.), are always looking for unique items to purchase (search history), and spend at an average $600 on shopping a month (Gmail receipts, android data mining etc.). What does that tell us?

Let’s form a hypothesis here: We should solve for the teenagers interested in purchasing clothing and accessories online since that not only improves user satisfaction, thus engaging those users more to the platform, but potentially opens a channel for future mainstream revenue.

Now that we have a definitive problem statement, let’s brainstorm some solutions:

· Make accessories and clothes available for purchase along/beneath the videos.

· Promote stores that sells merchandise related to the video being watched with the video.

· Run a frame by frame analysis of the video, use image recognition to identify the clothing and accessories used in each frame, and allow users to choose whether to buy or not.

How do we know which solution to opt for? We again look at a bunch of parameters:

· Originality of the idea

· Implementability (else the engineer will probably crush the idea)

· Threat of substitutes

· Potential market size.

Now, while talking about implementability, we shouldn’t just choose a solution that’s easy to implement rather should be ready to walk through the interviewer on how we can implement a seemingly impossible solution.

A general analysis would let us know that the third solution is better than the other two if all the factors are considered. Let’s get a bit deeper into the solution: Imagine that you’re watching a video when you like a shirt worn by your favorite actor. You click on it and you get the shirt open in a new window that lets you customize the product. You do it and share it in a marketplace (let’s say YouTube controlled) which is demand-initiated than supply initiated meaning that you post the product you wish to purchase, and suppliers/manufacturers bid for it to either sell you right away or manufacture and sell it. We can add a social element to it where we post multiple products we want to purchase, and our connections help us choose the best one, which then is taken up by a manufacturer who wins the bid.

Why would it be important to YouTube?

· It engages users not only with the video but now with a social scene

· Can diversify revenue streams (not just Ads)

· This increased revenue stream (let’s say commission on each order placed) can help reduce YouTube TV and Premium prices which can then see better adoption.

· The bidding system (the lowest bid wins the order) is in user’s favor.

Do we have trade-offs here? Obviously!

· What if users can’t even figure out how to use the feature?

· What if the frame by frame video analysis is not efficient enough to create the exact replica of the item the user is looking to buy?

· What if the sellers collude and instead of going down, the sellers bid high and share profit amongst themselves offline?

We should be careful during the product implementation phase to make sure we have adequate business logics implemented to stop these from happening.

How do we define success for our product feature?

The idea is to tie them back to our initial goal:

If our current engagement time per user in the US was 3 hours per day, we should strive to achieve 3.6 hours or more per day within a quarter. What should we measure?

· Number of users clicking on the video to initiate the feature

· Number of users placing an order

· Time taken from initiation to placing an order

· User adoption rate

· User churn rate (users who initiate the feature, customize it but don’t place the order)

· Average turnaround time per seller

· Average seller rating.

If we assume 200M US users are on YouTube with 3 hours per day engagement time and the adoption rate of the feature is 50%. Then those 100 million users need to spend 4.2 hours per day on an average by the end of the quarter for us to call the project a success!

There, we have it. A seemingly long answer to the most common product management question. Will you have enough time to go through every checkpoint mentioned here? NO! Can you do every quantitative calculation mentioned here? NO! But it’s always worth mentioning (and passively checking with the interviewer) that if we had enough time, I’ll explore this and whether the interviewer wants you to enter that zone.

Also, if you still have some time left — talk through implementation. How do we process so many videos frame by frame (frame compression using CNN, MapReduce to parallelly process and aggregate), where to store (SQL vs Object Store), how to store (hashmaps), time complexity (should we look at all the frames vs adopt a Heuristic method), what to cache etc.

Image source: Product Manager HQ