Case Study Analysis for Interviews: From Raw File to Recommendation
The raw Excel file looks harmless until you open it: blank cells, duplicate customer IDs, five spellings of the same city, and a revenue column that does not match the dashboard. The difference between an average candidate and a strong one is not who knows more formulas - it is who can turn that mess into a decision a business leader can act on.
- Start with the decision, not the data. Ask: what recommendation must this analysis support?
- Use a 6-step flow: clarify objective, audit raw file, clean data, segment, diagnose drivers, recommend action.
- Separate symptoms from causes. Falling revenue is a symptom; lower conversion, lower traffic, lower AOV or stock-outs may be causes.
- Use MECE buckets. Break problems into non-overlapping, collectively complete drivers like Traffic x Conversion x AOV.
- Quantify impact. A recommendation without size, trade-off and risk sounds like opinion.
- Communicate answer-first. Say the recommendation, then 2-3 reasons, then evidence and caveats.
- Common trap: presenting charts instead of a decision. Charts are evidence; the recommendation is the answer.
Big picture: a case-study analysis is a conversion process. You are converting a raw file into business judgment. The file gives signals, but the recommendation comes from structure: objective, data quality, driver diagnosis and decision logic.
The Core Method: The 6-Step Case Analysis Workflow
The mistake most students make is opening the dataset first. In a business case, the dataset is not the starting point; the decision question is. If the business question is “Should we expand into Tier-2 cities?”, your analysis is different from “Why did Tier-2 performance fall last quarter?”
The Two Mindsets: Analyst Who Reports vs Analyst Who Recommends
Raw-file cases punish passive analysis. You do not get credit for saying “sales declined by 8%” unless you explain why, how much it matters and what to do next. The interviewer is testing whether you can move from descriptive analytics to decision analytics.
Definitions You Can Say in One Breath
- Case analysis: A structured method to convert evidence into a decision with assumptions, trade-offs and risks made explicit.
- MECE: Mutually Exclusive, Collectively Exhaustive - categories do not overlap and together cover the whole problem.
- Hypothesis: A testable explanation for what is happening and why it may be happening.
- Barbara Minto's Pyramid Principle: Start with the answer, then support it with logically grouped reasons and evidence.
- Recommendation: A specific action choice supported by evidence, impact estimate, risk and implementation logic.
The Driver Tree: How to Find the Real Problem
A driver tree prevents random chart-making. For example, if revenue has fallen, do not immediately blame marketing. Revenue can fall because fewer people visited, fewer converted, customers bought less per order, discounts rose, products were unavailable, or returns increased.
Metrics That Make the Case Measurable
Use metrics in two layers: first check whether the file is trustworthy, then measure the business problem. Good numbers vary by industry, but the direction and benchmark logic matter more than memorising one universal target.
A Small Worked Example: From Numbers to Recommendation
Assume a toy Indian e-commerce category file for one month. The file shows 100,000 sessions, 2,500 orders, average order value of ₹900, gross margin of 45%, and variable non-COGS costs of ₹260 per order.
The interviewer is not looking for arithmetic alone. They want the leap from calculation to judgment: “The issue is not AOV or gross margin; the next diagnostic should compare conversion by acquisition channel and device.”
Case Study: Lenskart - Turning Omnichannel Signals into Expansion Logic
Lenskart shows how a company can use customer, channel and location signals to make better omnichannel expansion decisions in a trust-heavy category.
Situation. Eyewear is a difficult online category because customers care about fit, prescription accuracy, lens quality and after-sales service. A pure online funnel can generate interest, but many customers still need physical reassurance before purchase.
The move. Lenskart built an omnichannel model combining digital discovery, physical stores, eye testing, assisted purchase and vertically integrated supply. In a case-study setting, the raw file could include pin-code demand, website visits, store walk-ins, conversion rates, repeat purchases, prescription categories, service requests and delivery performance.
The analysis logic. A weak answer would say, “Open more stores where demand is high.” A stronger answer asks: where is online demand high but conversion low, where service coverage is weak, where average order value supports store economics, and where nearby fulfilment can maintain customer experience?
Outcome or lesson. The strategic lesson is that the primary driver is reducing category friction - fit, trust and service - supported by digital demand generation, store presence, supply-chain control and repeat-customer economics. In an interview, this is exactly the kind of multi-driver explanation that sounds like business judgment rather than spreadsheet commentary.
The Recommendation Pyramid: How to Present the Answer
Once your analysis is done, do not narrate your entire journey. Present the answer in a pyramid: recommendation at the top, 2-3 reasons below it, and evidence at the base. This makes you sound structured even when the raw file was messy.
How AI Changes Case Study Analysis
AI does not replace business judgment, but it changes the speed and breadth of raw-file analysis. The best candidates use AI to accelerate hygiene, hypothesis generation and communication - then personally verify the logic.
- Faster data understanding: Tools can summarise columns, detect missing values, suggest joins and flag suspicious outliers. You still decide whether the anomaly is an error or a business event.
- Hypothesis generation: LLMs can suggest driver trees for revenue decline, churn, store expansion or pricing cases. Your job is to make the tree MECE and relevant to the company.
- Recommendation drafting: AI can convert analysis notes into an answer-first memo, but you must check whether the recommendation is supported by actual numbers and caveats.
Use ChatGPT or Claude with a small anonymised dataset summary: “Here are the columns, data quality issues and KPI movements. Build a MECE issue tree and list five analyses to confirm the root cause.” Then use NotebookLM to load the company annual report or investor presentation and generate likely interview questions from its strategy and risk sections.
Interview Relevance
“You are given a raw sales file for an online retailer. Revenue is down this month. Walk me through how you would analyse the file and arrive at a recommendation.”
Use this sentence when you are stuck: “Before I jump into charts, I will first define the outcome metric, check whether the file is reliable, and then decompose the change into MECE drivers.” It buys time and signals structure.
Common Mistake
The single biggest error is confusing analysis with recommendation. Candidates describe cleaning steps, pivot tables and charts, but never make a decision. It costs them because managers do not hire analysts to admire data; they hire them to improve decisions. Fix: end every major insight with “therefore, I would recommend...” and name the action, impact, risk and next test.
What to Revise Next
This is the final lesson in the course, so do a capstone rather than another concept. Pick one business case - revenue decline, market entry, customer churn or store expansion - and practise the full chain once: raw-file questions, cleaning assumptions, driver tree, KPI table, 3 insights and a one-page recommendation.