# BCG Gamma DS Interview

Hi, got contacted for a DS interview by BCG Gamma. How much of the interview process revolves around machine learning/advanced modeling vs traditional cases? Best ways to prepare? Thanks!

• GrubHub / Eng
I did this interview very recently. It was case-based and he asked very interesting questions, here are my notes:

Case Interview : client is facing a problem and we want to fix it w/ my experience, Design experiment .

Context: fast moving consumer retailer: eg 7/11 or target.
Local promotions for items on sale w/ discount
Client is trying to evaluate the benefits of coordinated national marketing
What is the optimal discount that should pursue

All promotions at the client run weekly, meaning they start Monday and end on Sunday
No canablization across products, all product sales are independent
Price elasticity is linear. If you raise the price, demand doesn’t really drop.

Question: Exploratory w/ Linear Regression to show price elasticity for each different product. Predict price elasticity.

The first one was to formulate a linear regression model to predict units sold given price. Then you can use coefficient on the predictor price as an indicator of price elasticity. Meaning for a unit increase of price you’ll see w increase in Sales. Where w will probably be negative or very small positive. It would be small positive if the item is inelastic to price, for example, gas. For example, if gas price increases, volume will not decease as much as luxury products for example.

He then posed a harder problem which I wasn’t able to solve, but given that we have a model for each product to predict units sold, and a budget for a whole store, how much should be price each item?

I formulated it as a linear combination of prices and elasticity: \$Product1 * w1 + \$product2 * w2 + … + \$product_i * wi <= BUDGET. How do you choose the prices now? This is a constrained optimization problem.

THEN, he posed a very interesting question, albeit a bit flawed for the interview context, about RF regression. We take that same data for the linear model and fit it using a RF regression. He then said the client complained that the fit isn’t monotonic, or in other words, it has a flat spots and this is wrong. WHY?

This is b/c the DT fitting routine recursively splits the decision space. In this case, ti just finds the median/mean of the points at the split and draws a straight line. regtree2.png . Therefore if you don’t have enough estimators in your RF, you will see the flat spots!
Mar 16 6
• OP
Thanks for the response! What level were you interviewing for if you don’t mind me asking?
Mar 16
• Just a nitpick, but a monotonic function may have flat spots. A bijection has an inverse.

I'm not sure what hardcore ML is or how regression doesn't fall under data science, what is the line between analytics and DS?
Mar 16
• Oracle stuffed🦄
Difference is whether you need to read papers and think about experimentation, not whether you use the latest hot set of algorithms. Thinking DL is DS is the mistake IBM made some years back
Mar 16
• Microsoft dpw83
Real data science (mostly titled as applied scientist) pays more TC and the supply/demand curve is better. Any monkey can do regression... You learn that in high school!!!
Mar 16
• So real data science jobs are purely defined by pay, interesting. Yes, most people can execute algorithms and get good results, there are some finer.mathematical results that are relatively.intereating that don't have very much practicality.

So, what technical areas qualify as real data science?
Mar 16
• Microsoft / Data rets
BCG cares more about your looks. If you look skinny, upper class, they're way more likely to hire you.
Mar 16 0
• Financial Engines / Data frugalguy
I interviewed and failed. All interviews were pretty much cases but data intensive. Sample cases - you are working with a oil company. (1) you are helping them price oil (in real-time ) for all their gas stations in Texas . How will you approach it. What data you want to see. (2) citi bike in new york city. How you define the strategy to rearrange bikes between locations. (3) working with a grocery chain, how you will define their loyality program. What data you want to capture. How do you define a model to issue coupons to increase sales. Hope this helps.
Mar 16 5
• OP
Thanks for the response! I heard of people getting a take home challenge, did you as well? And which level were you interviewing for?
Mar 16
• Financial Engines / Data frugalguy
Yes. First round is take home challenge. And phone screen to discuss the take home project.
Mar 16
• New olgias
Was the take home challenge on Hackerrank?
Apr 24
• Financial Engines / Data frugalguy
No. They send a dataset. And ask to develop a recommendation for specific questions using data.
Apr 24
• New olgias2
Ok, thank you. I have one coding interview with BCG Gamma soon and it will be on Hackerrank. I think it is somewhat new, in fact I cannot find info about it. Could you tell a bit more about the dataset you received? which industry? did you apply some basic ML to it? any tips are appreciated.
Apr 25
• Financial Engines / Data frugalguy
For preparation, try practicing cases. Like you would for any consulting interviews.
Mar 16 0
• Microsoft dpw83
Curious too. Tried to break into MBB a while back
Mar 16 0