The data analyst title is one of the most stretched in Indian staffing. The same three words appear on CVs of someone who runs pivot tables in Excel at a BPO, someone who builds Tableau dashboards nobody reads at a mid-size ITES firm, and someone who owns end-to-end measurement strategy at a Bangalore fintech. Same title, three very different jobs, three very different salary bands. Most clients hiring through a staffing agency want the third version — a partner who can translate a fuzzy business question into a measurable analysis and defend the result against skeptical stakeholders. Most CVs claim it and deliver the first. The strongest filter is live SQL. Window functions, CTEs, and joins beyond INNER JOIN separate the analyst who learned on the job from one who did a 6-week Coursera course. The guide below tells them apart in 45 minutes using a SQL test, a dashboard review, and a short stakeholder scenario.
Window functions (ROW_NUMBER, LAG, running totals), CTEs for readability, joins beyond INNER, basic query optimization. The hardest skill to fake and the most common fake on Indian analyst CVs. Live SQL at 20 minutes filters 70 percent of candidates who claim it.
Tableau, Power BI, Looker, Metabase, or Superset with multiple dashboards shipped to non-analyst consumers who used them. Ask for screenshots or a walkthrough. Dashboards "built for practice" do not count — production dashboards survive stakeholder feedback.
Understands median vs mean and when to use each, what a confidence interval means in plain English, correlation vs causation, why an average can mislead. Basic literacy, not PhD. Stops them from shipping an analysis saying "signups doubled" when one enterprise customer onboarded 200 trial users.
Translates "grow revenue" into "revenue per monthly active user segmented by acquisition channel over rolling 12 weeks" without hand-holding. Not "I pull the data when asked." Separates a senior analyst (75K+ per month in Bangalore) from a junior report writer.
Writes a one-page TL;DR at the top of every analysis with answer, confidence, and next step. Ask for a sample. Analysts whose deliverable is a 47-slide deck with no executive summary are not ready for stakeholder work.
Useful for analyses beyond SQL — cohort analysis, funnel stitching, fuzzy matching. Not required for most roles but a clear premium for product and growth analyst positions.
Has designed, run, or analyzed an A/B test. Bonus if they understand peeking, multiple comparisons, novelty effect, selection bias. Rare in the Indian market outside product companies; strong premium.
E-commerce, fintech, BFSI, healthcare, logistics. Reduces ramp by weeks. Candidates from Flipkart, PhonePe, Razorpay bring domain vocabulary that takes new hires six months to develop.
Modern teams use dbt with Snowflake, BigQuery, or Redshift. Familiarity with dbt models, tests, and lineage signals recency. Most legacy IT services analysts will not have this.
Keeps SQL, dbt models, or notebook analyses in version control with meaningful commit messages. Elevates the candidate from "report writer" to "analytics engineer" in client eyes.
Walk me through an analysis in the last year that changed a business decision. The question, the data, the result, and what changed.
What to listen for
Real causal chain from question to decision. Many analysts produce dashboards no one uses. Strong: "My PM wanted to launch feature X. I showed the segment was 3 percent of MAU, not 30. We deprioritized." Weak: "I built a dashboard for sales, I think they use it."
A stakeholder walks up and asks "are signups going up?" Walk me through exactly how you respond before opening any tool.
What to listen for
Clarifying questions before touching data. Which definition of signup — email capture, verified email, activated user? Which time period? Which segment? Straight-to-query is a yellow flag for junior thinking.
Write a SQL query right now: top 10 users by revenue in the last 30 days, handling the case where one user has multiple payment records per day.
What to listen for
Handles date filter, aggregation with SUM and GROUP BY user_id, ORDER BY desc, LIMIT 10. Strong candidates notice the multiple-records-per-day trap and ask "do you want distinct transactions or do duplicates count?" Weak candidates write SELECT * and panic.
You spent a month building a dashboard that nobody uses. What do you do?
What to listen for
Goes to stakeholders and asks what questions they actually answer today, rather than blaming them. Investigates whether the dashboard answers the right question. Considers killing it. Strong candidates treat unused dashboards as signal that the analysis was wrong, not that the audience is lazy.
Tell me about an analysis you published that turned out to be wrong. How did you find out, who told you, what did you do?
What to listen for
Specific example with specific impact. Honest acknowledgment, named person who caught it, concrete remediation. "I have never been wrong" is disqualifying — analysts at scale are wrong regularly and good ones learn to surface their own errors faster.
An A/B test shows your new feature increased conversion by 12 percent with p-value 0.03. When would you not trust this result?
What to listen for
Small sample size, peeking before completion, multiple comparisons if many metrics tested, novelty effect, selection bias, Simpsons paradox across segments, seasonality. Strong answer mentions at least three unprompted.
Favorite chart type for showing a trend over time and why. Least favorite for any purpose and why.
What to listen for
Visual literacy — line charts, area charts, sparklines for trends with opinions about log scales, dual axes, rolling averages. For least favorite, pie charts with more than four slices, 3D anything, rainbow palettes. "Pie charts" without nuance is yellow flag above junior.
Your PM insists a metric went up. You think it went down. Walk me through how you resolve this without a political fight.
What to listen for
Looks at both definitions side by side, finds the disagreement (usually different populations or time windows), writes a reconciliation memo, involves the PM in choosing the canonical definition. Escalates only if the PM refuses to engage.
Score each candidate against these weighted criteria. Total: 100%.
| Criterion | Weight | Signal |
|---|---|---|
| SQL and data fluency | 30% | Writes window functions and CTEs from memory under timer. Comfortable with messy joins, NULL handling, date arithmetic. Knows when to use HAVING vs WHERE. |
| Business framing | 25% | Translates business questions into measurable analyses without hand-holding. Asks clarifying questions before querying. Names stakeholders and decisions their analyses informed. |
| Communication | 20% | Explains a chart to a non-analyst in 30 seconds. Writes a clear one-page summary at the top of every analysis. Adjusts depth for audience. |
| Statistical literacy | 15% | Skeptical of results, especially their own. Understands basic experimental design pitfalls. Recognizes when sample size is too small to conclude. |
| Tooling depth | 10% | Has shipped real dashboards used by named stakeholders. Can discuss trade-offs between Tableau, Power BI, Looker, Metabase. |
CV claims "SQL expert" with certifications listed but cannot write a three-way JOIN with a date filter under a 10-minute timer
Cannot name a single analysis that influenced a real decision by a real person
All experience in Excel or Google Sheets — has never connected to a database or queried more than 100,000 rows
Claims "data scientist" but cannot explain a confidence interval or correlation vs causation in plain English
Dashboard portfolio is visually noisy with 20+ tiles per dashboard, inconsistent colors, no clear narrative
Upload Data Analyst CVs and let AI score every candidate against the same 42-point evidence rubric.
Try CVPRO Free