Skip to main content

How to Present Past Projects

Premium

During your interview loop, you’ll most likely be asked to discuss a past project through a prompt like:

  • “Tell me about a significant data science project you were involved with.”
  • “Tell me about a time you used data to persuade a stakeholder to make a different decision.”

In this lesson, we’ll guide you through how to ace this interview question by covering:

  • The purpose of past project presentations
  • Strategies to choose a project
  • How to structure your answer
  • Other manifestations of this question

The image below summarizes this lesson, so you can use it as a quick reference when preparing for your interview.

Past Project Framework

Purpose of past project presentations

With this question, the interviewer is trying to assess many things, which we’ve distilled into the CUTES acronym:

  1. C (complexity). Understand your level of experience through the complexity of your past work
  2. U (understanding). See how well you understand your fundamentals and how those help you make tradeoffs in your work
  3. T (teamwork). Learn how you work with others
  4. E (execution). Learn how you make decisions
  5. S (skills). Assess your data science specializations (machine learning vs. causal inference vs. product analytics) and how well they tie into the role you’re applying to

How to answer

Now that we know why this question is being asked, let’s discuss what makes an answer not just good, but great.

Ideally, you should plan your answer well before your interview, but you can also ask for 30 seconds to map an answer out live.

The main things to plan for include:

  1. What project you’ll choose
  2. What experiences you’ll emphasize within that project
  3. How you’ll structure your answer

Practice giving this answer to a mirror, a friend, or an interview buddy. You can also video record yourself and play it back to see where you stutter or get off track.

This question is usually asked over the phone or in person, so presentation decks are almost never expected. They may even make the answer seem too rehearsed.

Strategies to choose a project

Use the CUTES framework above to choose a project that fits your new role well. For example,

  • For a highly cross-functional role, identify an experience where you worked with stakeholders.
  • For a core ML role, describe a project that required significant machine learning to succeed.
  • For an early-stage startup, emphasize your ability to execute quickly when weighing tradeoffs.

What experiences to emphasize

Similar to how you’d choose a project, you should emphasize the experiences in the project that are most relevant to your new role. For example,

  • For cross-functional data science roles, highlight how you collaborated with stakeholders to interpret the data, the product manager to align on goals, etc.
  • For a core ML role, emphasize the model you chose, the tradeoffs between different models you considered, how you chose which one to go with, and how you evaluated your decisions.

How to structure your answer

A good answer includes the “why” (goal), “how” (method), “who” (stakeholders, users), and “so what” (impact).

A great answer always starts with the “why,” making it very clear up-front why the problem in question matters. This is crucial context for someone who has never worked in the setting you’re describing.

After describing the “why,” state the ultimate impact, and then discuss the “how” and the “who.” However, keep in mind that this isn’t a rigid rule. When deciding whether to discuss “who” or “how” first, talking about the “who” first can help keep things brief, since the “how” is the longest section.

“I was recently a data scientist on the new initiatives team at Uber. There, we wanted to understand if advertising could be a viable source of revenue for the business. This was important because our leading company target was to increase profits through new business lines. At the end of this problem, in just 3 months, we were able to estimate the new annual revenues from this business which helped us decide to launch it in 3 markets. It is now the fastest-growing segment of Uber’s business. Let me tell you how we went about it.”

Below, we’ll discuss how to approach each step of your answer:

  • Describe the problem
  • Describe the stakeholders
  • Describe the bulk of your work
  • Describe the impact

Describe the problem

For the rest of your answer to make a strong impact on the interviewer, they need to empathize with the “why” and understand the problem you worked on. To accomplish this, give CPR:

  1. Context: First, mention the setting in which you encountered this work. For example, “I was recently a data scientist on the prime memberships team at Amazon.”
  2. Problem: Second, mention the problem. “In that setting, we discovered that membership renewals were much lower than expected.”
  3. Relevance: Third, describe why this is an important problem. “This was an issue because renewals account for X% of Y (e.g. revenues, total membership), which made this a critical problem to address.”

A common pitfall here is to rush into the “how” of the answer, such as the models you built, the implementation details, etc. Avoid making this pitfall by starting with the “why.”

Describe the stakeholders

Usually, the “who” matters most for specialist data science roles. For example, research roles on a health AI team, where working with doctors might be a big part of the job. If that’s not the role you’re applying for, keep this part brief and invite follow-ups.

“In this work, we were trying to solve a problem for users. To do that, I partnered with a PM, driving the feature, 2 full-stack engineers who focused on product development, and the privacy team to ensure this approach was feasible in country X. Is there a specific partnership you’d like to hear more about? If not, I’d love to talk about how we solved the problem.”

Describe the bulk of your work

Explaining the “how” can be the hardest to talk about because you’re condensing weeks or months of work into a few minutes of explanation. It can be easy to describe more than needed or go into details where it doesn’t help.

For example, your work might have involved figuring out what data to use, building pipelines to pull that data from different sources, speaking to stakeholders to make sense of it, cleaning it, processing it, maybe even handcrafting features, creating test sets, building a model, testing the model, building a different model, productionizing the model, and monitoring it.

A common pitfall here is to try to cram every detail into your answer. Instead, select the 2-4 points you really want to emphasize. These points can be less “shiny” aspects of the work, like data cleaning. If you do so, connect it points to why it mattered and what made it challenging. Then, briefly mention the other details, making it clear you’re open to specific follow-up questions.

It took a number of other steps to get there, like aggregating the right data, making sense of it, and functionalizing it. However, the bulk of the work involved building the right model, deploying it, and monitoring it. I’ll focus on this section first. What do you think?

Try not to talk for more than 4-5 minutes without checking in with your interviewer. Making this a habit will help you clearly focus on describing segments of your work, and also make the interview more engaging, all while making it unlikely that you fall into any rabbit holes that your interviewer can’t follow. After each segment, check in with something like,

“Next we can discuss X aspect of this work, or we can dive into any other details you were curious about. If there’s anything I just said that you’d like to dive into, we can do that too.”

Great “how” answers often include:

  1. A meaningful north star metric. Connect the north star to the business goal. For example, “Because we cared about retaining memberships, our goal became to predict cancellations with as high accuracy as possible”
  2. Tradeoffs that were made and how. Connect the tradeoff decisions to the north star. For example, “In order to solve the problem, we built 3 different models: a baseline logistic regression, a gradient boosted tree, and a random forest. Since the latter had the highest accuracy…” Consider describing the effort you made to build the model, the amount of data needed, the explainability, and the potential for bias.
  3. The scope of the data and/or the tech stack. What to emphasize will vary based on the role. More senior roles will emphasize the stakeholders involved in getting the data, rather than the tooling. Less senior roles should address the questions below:
    1. Was the data unstructured or structured?
    2. How many data points did you have?
    3. How did you build the necessary pipelines?
    4. How did you productionize your model?

Describe the the impact

Describing the impact, or the “so-what,” is an effective way to close strong. It can also be a good way to start strong after describing the problem and the context, i.e. giving CPR, and then describing how impressive this work was.

When describing the impact, be sure to:

  1. Restate the problem. “We were trying to solve X because Y”
  2. Describe the quantitative impact on the north star. “Our work helped reduce Y by Z% in T time, saving the company D dollars”
    1. Be sure to include a baseline or piece of context here. “For context, D was 30% of the entire spend, or Z% was 50% better than it used to be.”
  3. Connect the model or analysis results to the actual business problem. “As a result of this work, users said A, our business unlocked B, etc.”

How long your answer should be

To determine the length of this answer, assess whether it is the main question for the entire interview (~30-45 minutes) or one of many questions in a given interview round. Usually, you can get this sense pretty early on, as the interviewer lays out the agenda.

When giving the short version of this answer, focus on the “why,” “how,” and “so-what.” Briefly mention the “who” and invite follow-up questions. As a rough rule of thumb, spend 30% of your time on the “why,” 40% on the “how,” and 30% on the “so-what.”

Other manifestations of this question

This lesson outlines how to answer a specific project you worked on, but it applies well to other behavioral questions.

For example, to answer, “Describe a time you convinced a PM to take a course of action using data,” instead of using CUTES to choose a project, you would choose an experience that matches the question, and then follow the same approach, this time emphasizing the stakeholder relationship with the PM, instead of, say, the model you built.

Summary

  • Assess what kind of experience your interviewer would like to hear with the CUTES framework (complexity, understanding, teamwork, execution, skills)
  • No matter what problem you choose to present, give it CPR. Give Context, describe the Problem, and explain its Relevance.
  • Aim to cover the “why,” “who,” “how,” and “so-what” of your work. While giving this answer, especially in discussing the “how,” check in with the interviewer frequently to ensure they’re engaged.
  • Close with a statement of why this work mattered (the “so-what”). Describe the metrics, north star(s), and impact created.
Next Lesson