Last year I was interviewing for an Analytics Manager role in telecom. I passed the HR screen, behavioural test, the hiring manager call, meeting with my future manager and colleagues. Then they sent me an email.
"We'd like you to complete a case study. Please find the dataset attached."
I had less than 24 hours to send back a presentation.
The data was messy. Based on the column names I could tell it was a raw export straight from the database. No instructions on what to analyse. No template. Just a file and a deadline.
This round is where most candidates fall apart. Not because they can't do the analysis. Because they don't know what the round is actually testing.
Today let’s talk about the Case Study Round.
Before we proceed, a small ad. Your clicks on ads help me to cover hosting fees. Last month I made enough to pay for the hosting + a cup of matcha. Thank you!
Facts. Without Hyperbole. In One Daily Tech Briefing
Get the AI & tech news that actually matters and stay ahead of updates with one clear, five-minute newsletter.
Forward Future is read by builders, operators, and leaders from NVIDIA, Microsoft, and Salesforce who want signal over noise and context over headlines.
And you get it all for free, every day.
What the case study round actually tests
It is not testing whether you can clean data. It is not testing whether your model is perfect.
It is testing three things:
Business thinking — can you look at a dataset and identify what actually matters? Not every column is interesting. Not every trend is worth showing.
Communication — can you turn your findings into something a non-technical stakeholder can act on? An executive summary. A clear narrative. Slides that don't need explaining.
Judgement — what did you choose to include, and what did you leave out? This tells them more about you than the analysis itself.
The technical work is the entry ticket. The presentation is what gets you the job.
Step 1 — Read the data before you touch it
Before writing a single line of code or opening a pivot table, spend 20–30 minutes just reading the data.
Open it in Excel or run a quick df.head(), df.info(), df.describe() in Python. Ask yourself:
What does each column represent?
What time period does this cover?
What is one row? One customer? One transaction? One day?
What looks wrong immediately — nulls, negative values, duplicates, dates that don't make sense?
Write these down. You will be asked about them in the presentation.
In my case there were duplicates, and some values were in a wrong column (as if the data shifted when it was exported).
Step 2 — Clean, but don't present the cleaning
Yes, clean your data. Remove duplicates. Handle nulls. Fix data types. Filter out obvious errors.
But do not put data cleaning slides in your presentation.
Nobody wants to see a before/after table of your null values. Executives don't care. What they care about is that your numbers are trustworthy.
What you do instead: in the Q&A, when they ask "did you clean the data?" — and they will ask — you explain clearly what you found and what you did about it. That's when the cleaning work pays off.
During the Q&A they asked exactly that question and I walked them through everything I found. It showed I'd done the work without wasting their time upfront.
Step 3 — Find one strong insight, not ten weak ones
Most candidates try to show everything they found. Resist this.
Pick your strongest finding — the one that would change a business decision — and build your narrative around it. Supporting insights can come after, but there should be one clear "so what" that runs through the whole presentation.
Think of it like a newspaper article. The headline comes first. The detail follows.
My approach is to dig for all the insights, and once I have the full picture - I decide what will be included in the main slides, and what goes into the backup.
Step 4 — If they ask for a model, keep it simple and explain it
I was asked to build a predictive model.
I built a simple logistic regression first, explained what it was doing and why, showed the key features driving the prediction, and talked about what a more complex model could add and when it would be worth it. I wanted to try XGBoost too, but I was running out of time. That’s why decision was to talk about future development and iterations
They are not evaluating your model. They are evaluating whether you understand what you built.
Tools: Python (scikit-learn) is the safest choice for speed. If you're more comfortable in R, use R. If the dataset is small and they haven't specified, Excel with a regression add-in is completely fine.
Step 5 — Match the company's brand before you design a single slide
This is the thing nobody tells you.
Go to the company's website. Find their latest annual report or investor presentation — most public companies publish these. Download it.
Now you have their exact colour palette, their font choices, their slide style. Build your presentation to match.
When I did this for my interview, my slides looked like they could have come from the company's own marketing team. That is not a small thing. It signals that you pay attention to detail, that you think about your audience, and that you already see yourself as part of the organisation.
How to find brand colours: open the annual report PDF, take a screenshot of a branded slide, and upload it to coolors.co or imagecolorpicker.com to extract the exact hex codes.
What to include in your presentation
Keep it short. 8–12 slides maximum.
Slides 1 - 2 — Executive summary. One paragraph. Key finding, what it means, what you recommend. Write this last, put it first.
Slides 3–7 — Your analysis. 3–4 key findings with charts. One insight per slide.
Slide 8–9 — Model (if asked). What you built, what it predicts, key drivers, accuracy metric in plain English.
Slide 10 — Recommendations. What should the business actually do? This can also be moved to the beginning, right after the Executive Summary.
Slide 11 — Next steps / limitations. What would you do with more time or data?
On the timeline
I had less than 24 hours to submit the presentation, then almost a week before the actual presentation date — apparently because of calendar conflicts on their side.
If this happens to you, don't waste the extra week. Use it to stress-test your findings, prepare for the hardest questions they could ask, and rehearse your narrative out loud. The gap between submission and presentation is where you can genuinely pull ahead.
Most people treat the case study as a test, but it is a preview of the job.
You are not being evaluated. You are showing them what their Monday morning will look like if they hire you.
That mindset changes everything — how you structure the work, what you prioritise, how you present.
Keep pushing 💪,
Karina
Need more help?
Just starting with Python? Wondering if programming is for you?
Master key data analysis tasks like cleaning, filtering, pivot and grouping data using Pandas, and learn how to present your insights visually with Matplotlib with ‘Data Analysis with Python’ masterclass.
Building your portfolio?
Grab the Complete EDA Portfolio Project — a full e-commerce analysis (ShopTrend 2024) with Python notebook, realistic dataset, portfolio template, and step-by-step workflow. See exactly how to structure professional portfolio projects.
Grab your Pandas CheatSheet here. Everything you need to know about Pandas - from file operations to visualisations in one place.
Data Analyst & Data Scientist
