Analyst interviews at Pakistani banks, telecoms, and e-commerce companies mix three ingredients: SQL exercises (often written), Excel scenario questions, and a business-thinking case. Tools get you shortlisted; the case round decides the offer. Here are the recurring questions from each round.
INNER JOIN returns only matching rows in both tables; LEFT JOIN returns all left-table rows with NULLs where the right has no match; RIGHT JOIN mirrors that; FULL OUTER combines both. The favourite practical probe: "customers with no orders" — LEFT JOIN orders then WHERE orders.id IS NULL. Being able to write that pattern fluently, plus knowing join duplication risk (one-to-many inflating counts), covers most interview SQL.
WHERE filters rows before grouping; HAVING filters groups after aggregation. "Branches with total deposits above X" needs GROUP BY branch HAVING SUM(deposits) > X — filtering on an aggregate in WHERE is the error interviewers plant. Follow-up often asked: execution order (FROM → WHERE → GROUP BY → HAVING → SELECT → ORDER BY), which also explains why column aliases can't be used in WHERE.
A cleaning checklist, stated in order: understand the schema and business meaning of each column, profile the data (row counts, distinct values, null rates, ranges), fix structural issues (duplicates, inconsistent categories like 'LHR'/'Lahore', date formats), decide null-handling per column with business logic rather than blanket deletion, and document every transformation so it's reproducible. Naming reproducibility is what elevates the answer from junior to hire.
Detect via distribution checks — IQR fences, z-scores, or simply plotting. Then the key point: never remove automatically. An outlier is either an error (fix or drop with documentation) or a real extreme (often the most important row in the dataset — a fraud case, a whale customer). The interviewer is testing business judgment over formula recall.
The classic case question; structure beats speed. Verify it's real (tracking change? definition change?), segment the drop (platform, region, user cohort, acquisition channel), align timing with internal events (releases, pricing, campaigns ending) and external ones (competitor, seasonality, connectivity), then quantify the dominant driver and recommend. Thinking aloud in this order — data validity first — is precisely what's scored.
It answers its user's decision question in under ten seconds: the headline KPI with trend and target, supporting breakdowns one level down, and drill-through instead of clutter. Practical points that land well: build for the consumer (a branch manager needs different granularity than the CFO), avoid vanity metrics, and put filters where users expect them. Referencing a dashboard you built — with who used it and for what decision — turns this into your strongest answer.
A strong resume is what gets you into the room. Build one free — ATS-friendly templates, unlimited edits, no account needed, first download free.