A guide to passing the metric investigation question in tech companies

Hey everyone! Inspired by a recent post, I wanted to share a guide for tackling open-ended interview questions in data science. Examples of such questions might include:

  • A c-suite exec tells you day-over-day revenue is decreasing. What do you do?
  • A PM asks you to opportunity size a new product version.
  • A PM comes with mixed A/B test results and needs your help interpreting them.

Disclaimer: While I’m a senior DS, I mostly conduct coding interviews, not these types. This guide is based on my own experiences, and I welcome your feedback to improve it.

These questions usually aren’t about getting the “right” answer, but instead about showing your ability to:

  • Break down complex problems into manageable steps.
  • Take a systematic approach to analysis.
  • Clearly communicate your reasoning.

Framework for Open-Ended Questions:

Example: Exec says revenue is dropping. What do you do?

Step 0: Outline your approach
Start by giving the interviewer a high-level overview of your process.

Example: “First, I’d understand if the issue is isolated or systemic, then break down the metric, de-aggregate, and finally suggest preventive measures.”

Step 1: Understand the big picture
Ask clarifying questions, check related metrics, and assess if the problem is widespread.

Example: “Is this gross revenue? What about active users or subscriptions? This helps gauge how broad the issue is.”

Step 2: Narrow the scope
Look at trends and other possible explanations.

Example: “Did revenue drop this time last year? Could we look at revenue per user to pinpoint whether the issue is user loss or lower spending?”

Step 3: Dig deeper
Now de-aggregate or identify new metrics.

Example: “Let’s split by revenue streams or geography. Ads vs. purchases, US vs. non-US users.”

Step 4: Prevent it from happening again
Show you’re thinking long-term.

Example: “We could add a topline metric for revenue per user and a year-over-year growth view to spot cyclical trends.”

Step 5: Advanced techniques (optional)
If time allows, showcase deeper knowledge.

Example: “We could use causal analysis with tools like DoWhy for root cause analysis in similar future problems.”

Opportunity Sizing Example:

PM asks you to size the potential of a new product version.

  • Step 0: Outline your approach.
  • Step 1: “Is this for all users? Have we launched a similar product before?”
  • Step 2: Identify key metrics like revenue per user or engagement.
  • Step 3: Compare with historical launches, use effect sizes from previous experiments.
  • Step 4/5: Suggest experimenting for future launches, or apply causal modeling techniques.

Final thoughts:
There’s no perfect formula for open-ended questions. Interviewers might steer the conversation, so be flexible! Please share any resources or ideas to enhance this guide.

Looking forward to your thoughts!

4 Likes

Additionally, I would advise you to think about the different types of adjustments. To characterize the possible extent of the problem, I’ve heard the term TROPICS, which stands for:

Duration

Region (for instance, outside modifications such as laws in a certain city)

Additional features and items (product modifications at your business)

Platform (such as desktop, mobile, operating system, or browser)

Industry and rivals (for instance, did Apple Music include a feature that may have persuaded customers to move from Spotify if its listenership declined?)

Cannibalization (does your company’s new product cause a decline in an existing one?)

Segmentation (other types not specifically covered above)

3 Likes

It sounds like this interview will be annoying. The interview process is flawed if a qualified candidate need assistance to navigate it.

There is an unwritten script that applicants are expected to follow rather than being explicitly told what you are looking for. Although you won’t reveal the script to them, they should be aware of it.

2 Likes

I appreciate you sharing. For this question, how would you use the DoWhy package?

1 Like

I adore the concept of establishing a framework and connecting it to the business environment. In order to connect with business objectives, ask the correct questions and keep an eye on the wider picture when dealing with open-ended revenue challenges.

Use previous launches to forecast important KPIs for opportunity size, and always try for tangible outcomes. Additionally, use tools like AgentQL to streamline big data processes so that you may concentrate more on analysis than data collection.