Skip to main content

What are Decomposition Questions?

Decomposition (decomp) questions are the closest thing to an FDE's actual job that you'll face in an interview. Palantir pioneered the format, and more companies who are building out their own FDE function are mirroring it e.g., OpenAI and Databricks.

A decomp question asks you to design a solution for a stakeholder, using the resources you're given, within a time bound.

"Given a dataset of taxi trips with fields like fare, locations, times and distance, propose a solution that helps drivers and could ship within a week."

This prompt is a textbook decomp question because it hits all three elements you need to identify before you start solving: the stakeholder, the resources, and the time constraint. Here's what each one looks like and why it matters.

Stakeholder(s)

“drivers”

A decomp question will state the stakeholder(s) you're helping with your solution. This is the most important element in the question, since it's what points you toward the pain points you should address and shapes the solution you end up designing.

The stakeholder isn't always named explicitly. For example, "A major city wants to reduce 911 emergency response times" names the city, but the city isn't the stakeholder. The emergency response units are. Getting specific about who your stakeholders are helps you identify more concrete pain points and design better solutions. Don’t be afraid to confirm your assumptions explicitly with your interviewer.

Resources

"dataset of taxi trips with fields like fare, locations, times and distance"

What resources are available for you to use in your solution? Sometimes this comes in the form of data sources (e.g., traffic data, weather forecast) or technologies (e.g., AI).

If the resources aren't explicitly mentioned, ask the interviewer during the interview. We'll show you how to do this in the next lesson.

Time bound

"ship within a week"

Tight deadlines are something FDEs face on deployments, and the time bound tests your scoping skills. Can you scope a minimum viable product (MVP) that addresses the customer's pain point within the constraint you're given?

Since you won't be able to ship the entire solution, only an MVP, you're also expected to sequence out the next few iterations, in much less detail than the first.

Examples

The taxi prompt spells out all three elements cleanly, but not all prompts are this tidy.

"A major city wants to reduce 911 emergency response times. They have call data, traffic data, and ambulance GPS data. How would you approach this?"

Here, the stakeholder takes some thought to identify (it's the emergency response units, not the city), the resources are a set of three data sources instead of one dataset, and there's no explicit time bound at all. That’s your cue to ask the interviewer what timeline they have in mind before you start designing.

"Build a system that assigns analysts to investigations based on expertise and availability."

This one reads more like a technical spec than a product prompt. There's no dataset named up front and no deadline mentioned, but the stakeholder (analysts and whoever staffs investigations) and the resource question (what data would you need to make this assignment decision) are still there.

Practice

Try identifying the stakeholder, resources, and time bound (if any) in each of these before moving to the next lesson: