This is an interview. Meta I was interviewed by the WhatsApp team for a DS position in the Business Product direction. The interviewer was from the WhatsApp team. This direction is inclined to product analysis, which pays more attention to user behavior, growth strategy, experimental design, etc., and requires a higher understanding of Product Sense. The overall process is divided into two rounds, one for Product Sense and one for Execution, and today I'm sharing one of the Product Sense interviews, which was very in-depth, with the feeling that I was not brushing up on the questions but rather chatting about product logic. The analytical thinking, assumptions, experimental methods are push very deep, in the middle of the programhelp voice reminder, or a few points may really break.
Product Sense: Structured but deep questions.
| module (in software) | element |
|---|---|
| Type of interview | Meta Product Sense Electric Surfaces |
| Interviewer background | WhatsApp Team DSA |
| length of interview | 45 minutes. |
| Interview highlights | User behavior analysis, data-driven product decisions, experimental design logic |
| interview difficulty | ⭐⭐⭐⭐ (requires strong product reasoning + depth of analysis) |
Interview Process: Stuck vs.
First question: How do you see the need for a group call?
I gave some high level points at first, like:
If user a often plays videos with b and c separately, it may be an alternative behavior to group;
If these calls all occur within a short period of time, it's all the more reason why they need the group call feature;
But the interviewer immediately followed up with, "How do you define a short period of time? Is it 5 minutes or 30 minutes? Any thoughts on how to model it?"
I was stuck for a while, I had a bunch of ideas in my head but couldn't make sense of them. Luckily, programhelp reminded me that I could use a "sliding time window + weighting graph" to describe the structure of the user base. I immediately picked up on that and added that I could:
- A weighted network is constructed with users as nodes and the weights of the edges represent call frequency and time proximity;
- Find frequently recurring combinations (e.g., using community detection);
After the interviewer listened to the obvious recognition, but also asked what can be done if there is no network tool, I went on to say that you can use pair count + clustering way to approximate can also determine the needs of the group.
Second question: if you could get more data, what would you want?
I'd say it can be expanded in two ways:
determine the nature (usually of error or crime)(Qualitative): user survey, feedback, focus group discussion to see if there are any voices;
quantitative(Quantitative): e.g. whether these users are present in a group chat, their chat history, whether they often synchronize their calls, etc.
I was more prepared for this part and answered it fairly well.
Q3: How do you design experiments before going live? What metrics to look at after go-live?
In this section he clearly says "pre-launch experimental design". I clarified and then said:
Use A/B test, but be aware of the randomization unit, which may need to be made clustering randomizationBecause there may be spillover between friends;
It is mentioned to control the novelty effect and look at the indicators when observing the statistical significance + Business impact.;
The interviewer went on to ask: "How do you do the clustering you mentioned?" I was a bit stuck in this place and couldn't figure out how to express myself, but programhelp reminded me that clustering can be done from geographic location, so I immediately added that clustering can be done according to region, or group chat activity level.
Don't hang on to your job if you don't have a clue about the interview
Overall, there is no standard answer for this round of Product Sense interviews, but you must speak up:
Why look at it this way;
How to validate by data;
Whether scalable or not;
What are the limitations.
I got stuck a few times in the middle of the process, but fortunately I had programhelp's remote voice assistance to help me sort out the logic in time, especially when explaining complex methods (e.g. clustering randomization), the voice reminders are really crucial!
If you are also preparing for Meta / Stripe / Pinterest / TikTok etc. product analysis post, don't just brush up the questions, practice the logic of thinking + disassembling + experimental design. If you need OA ghostwriting or interview help, please come to us! programhelp!
