product analytics/ sense 跟踪记录贴

How would u solve a problem in performance in a theoretical web app?
  

LC skyline problem

one Interview:

1) Design a Washer and Dryer - You need to ask the interviewer questions about who the users would be. When I asked him, he said "The same sort of people who would buy a nest thermostat (Younger, more tech savvy and somewhat affluent).
2) How would you measure the success of Apple's WWDC event (This year vs Last Year). Note: The key here is to discern the difference between the event itself being better vs Apple releasing cooler stuff this year thereby getting more press coverage etc.
3) How would you improve Youtube

Onsite:

1st Interview:
1) Latest trends in technology
2) What are the network implications of a handheld device meant for elementary schools kids
3) If Google has invented a way to make air travel 4x cheaper and 4x faster, what would you do with this technology?

2nd Interview:
1) Design an app for the dmv
2) Explain recursion to my grandma
3) You have an ecommerce site in Spain and you've just localized this and now see that traffic has reduced. What could be the reasons.
4) Give me 3 features you like about your favorite google product and 3 that you would improve.

3rd Interview: 1) How would you improve throughput of an airport by 100% 2) You've done all this stuff to improve throughput , what are the implications for the airport itself 3) How much storage space is required to host all the images of Google street view. 4) You see starbucks' on either side of the road sometimes, why do you think Starbucks Corp does this?

understand how much increase would get to +100%. Also need to know bottlenecks needed to achieve +100% throughput.

Breaking down throughput: avg # of people through airport per time period (day)
* The utilization rate of the airport by planes (can I fit more planes traffic into airport)?
* Utilization rate of planes (is there opportunity to get more butts in seats)?
* Airport plane capacity (if there's high utilization of planes and airport, do we need to build more terminals to accommodate the traffic)?
* Airport human capacity (is the airport's capacity to process passengers a constraint)?
* Seasonality (are there days/months where the airport is under utilized, can we improve upon that)?

If we optimize everything, can we get to 100% increase in throughput? What's the revenue vs. cost impact of doing so (would the incremental increase in revenue cover the cost? how quickly could it cover the cost?)

The goal cat be known beforehand since you don't know who the user is. Need to start there first- What the best guess for who this user is? I would use any one of several IP traffic analysis services (E.g. Alexa, Comscore) to get a demographic profile of the user. The service takes into consideration things like time of day the user is accessing the service, location (Based on IP lookup), type of browser (based on UA string), if available 3rd party cookies etc to form a profile that will tell you with a n atached confidence interval if this person is male, 30-45, interested in hiking etc- Can be pretty granular.

Second- Once I have that profile, map that onto heuristics that have been predetermined based on Youtube's biz goals and prior user study- E.g. Single males 30-45 are most likely to watch sports videos. This demographic is more likely to pay for exclusives/ More likely to click on associate videos. Make 4 more similar inferences.

Finally- Use those 6 inferences to pick from a virtual carousel fo videos (E.g. Latest Vs. Most liked Vs. Most views etc)- Idea is to mix it up.

Let's start with the problem ? - get started with YouTube ?

Goal - is to get them to view a video ? be a sticky customer ? convince them that this is a replacement to their TV ?

if someone is logging into YouTube for the first time it could be one of two people

> they have used YouTube before
> they haven't used YouTube before and have started by signing up first and then using YouTube
> they have just up with google elsewhere and are just logging in on YouTube with their google credentials.

In short.
> we may not have their YouTube history but might have their other Google history
> external signals (web/mobile, time of day, region, country etc).

Content
> Let's think about type/format of content
   > Typical YouTube content ( movie trailers, songs, funny video etc)
   > Custom content
      - Introduction to YouTube
      - What's on ? (similar to the way they have with Cable where a channel continuously spins what's new and latest)
   > Short form and long form content.
> Age Group (suitable for a particular age group)
> Such type of content is across various categories ( entertainment, science, technology)

----

For a moment think about how does history of YouTube help you ?
> Show recently viewed.
> People who viewed this also viewed y (recommendation)
   .. basically you get better at predicting what a user might like and improve the probability of them viewing your recommendation.

----

Let's say we come up with various options on which 8 videos to show

Option 1 : Top videos across closest matching affinity group (let's say user is based in Kansas) - what's the top video watched 8 videos in Kansas for today

Option 2 : Custom / Personalized content (What's on , Introduction) + 6 top videos

Evaluate various option based on user persona, goals .

...

there isn't going to be a right answer to this question so the best guess is to make an educated guess well enough so that you can run your first experiment then be ready with other experiments so you can pick the best one

▲
0
▼
A very simplistic answer is success is when I meet the goals I set for the product. however a more elaborate answer would::

Typically I would think that success should equate success for my stakeholders or my customers. lets breakdown who the stakeholders are in this process.

My Org/Company - products are built to further company goals so success should translate to goals of the company
> Revenue goals
> Acquisition goals

Buyers of my software - products should solve a problem for them (helping them design their process)
> Usage, Repeat purchase, NPS # of processes designed, # of complaints are all metrics to measure health of software use
> Qualitatively - through feedback etc one can get a sense of if the buyer feels value in our product and that's a key ingredient of success.

Team success
> at the end of the day your team is one that helps deliver on the product. the team could deliver on the product but then feel so burnt out or have a terrible experience that they may not want to work on the next iteration - although this does not contribute to product success it's not something
> other way to look at it is to see did we launch it in time within budget etc ?

Resellers
> if the products are offered by resellers, you can track their individual metrics as well.

---

From a financial perspective a measure of success can be

> are we tracking per the business plan ?
> what's our acquisition cost ?
> how much did we spend to build the product?

Operational perspective

> are we incurring more supply chain, servicing costs.

the key in my view is that success is not a single # - although one can always roll up to a # that can be a proxy for all others but an appreciation of how and what success means for everyone in the value chain is important to help answer the question what is success.

Acquisition, Activation, Retention, Referral, Revenue << framework is also a good way to think about answering this question

记录自己的学习 感觉这个方法不错！

calalia 发表于 2016-10-18 03:40
晕。好像最近Apply出错超级多 self management 备受质疑啊
还是要端正态度

谢谢分享，超级全面啊

如何刷pm的题呢

信息好详细。