These questions are open ended case studies. The goal here is to have a dialogue with the student (interviewee) about methodologies and techniques to solve the problem.
You are working for a large software company that provides a calendering application to large businesses. The businesses use the application to schedule all meetings between employees.
Your company would like to learn more about how the customers are using the application, and in particular, are interested in how users with different profiles interact with the application. As a small part of this project, you must identify which users of the application within a company are managers.
Assume you have access to all the scheduled meetings at a large company across multiple years. Each data record you have available represents one meeting scheduled at this company, and contains the following information:
- A user id for the person who scheduled the meeting.
- User id's for all the people invited to the meeting.
- A timestamp for the time the meeting began.
- A timestamp for the time the meeting ended.
- A flag for whether the meeting is recurring (i.e. an instance of a repeating meeting that recurs at regular time intervals).
- Information on the recurrence of a meeting (i.e. is the meeting weekly, monthly, quarterly...).
You have no other information; in particular, you have no identifying features about the users of the application besides the meetings they were invited to or scheduled.
Discuss how you would design a solution for identifying the users within the organization that are managers.
You are contracted with a large public library (which has multiple locations within a US metro area) to help identify checkouts of books that are at a high risk of not being returned. You have five years of historical data in three tables (this data will be pulled for you on the day the contract begins, and will be current up to that day):
| column | type |
|---|---|
| user_id | int |
| join_date | date |
| branch_id | int |
| name | string |
| ... | ... |
| column | type |
|---|---|
| book_id | int |
| author_id | int |
| genere_id | int |
| publish_date | date |
| name | string |
| ... | ... |
| column | type |
|---|---|
| user_id | int |
| book_id | int |
| checkout_time | timestamp |
| return_time | timestamp |
The Users and Books table contain demographic style data on the users and books associated with the library, feel free to make assumptions about what exact information these tables may contain. The return_time field in Checkouts table is null if the book has not yet been returned.
Discuss how you would develop a predictive model to help inform the library about what books are likely to go missing. How would you deliver this data to the library? How would you evaluate whether your solution meets the libraries needs?
You are a data scientist a medium sized bike sharing company in a large city. Your company distributed bicycles throughout the city, which users may checkout for short periods of time (multiples of ten minutes).
After being in the market for a few months, a few problems have emerged:
- Bikes sometimes disappeared, and need to be replaced.
- Over time, bikes tend to congregate at lower elevations in the city, as they are easy to ride downhill and difficult to ride uphill.
You are tasked with giving advice on how to approach both problems. You may assume that you have gps data on trips, i.e. where the trips begin, end, and some points in the middle of a trip if it is longer. Here are some questions:
- How would you detect whether a bike has disappeared?
- How would you decide when a bike needs to be replaced?
- How would you redistribute bikes that have congregated in lower elevations, and when would you take action?
In addition, your boss wants you to always keep the following issues in mind, as they are constantly relevant to the business.
- How would you use this data to inform the quality of bicycles used by your company?
- How would you decide the overall size of the fleet your company should employ.
- How would you decide how much to charge for each 10 min ride?
You are a data scientist at a large car insurance company working on optimizing their pricing strategy. Customers at this company pay for six month contracts; while the contract is in effect the company agrees to pay for all auto insurance losses incurred by the customer (for both accidents where the customer was and was not at-fault). After a customer buys a contract at a given price, that contract cannot be re-priced, the insurance coverage is locked in for six months. After the contract expires, your company will automatically offer the customer another six month contract, possibly with another price.
Your goal is to design a pricing strategy that will, on average, allow the company to stay profitable, and possibly, to accomplish this is an optimal way.
Discuss what data is important to collect on the insurance customers to accomplish this task. What models are important to build, and what quantities are important to estimate? How will these, possible multiple, models work in concert to form a full pricing strategy?