How to assess · For hiring teams

How to Assess Data Analysis Skills When Hiring

The test formats that actually work for Data Analysis, what a strong answer looks like, sample questions and a scoring rubric you can use as-is.

The short answer

Assess Data Analysis with a task, not a conversation: take-home analysis on a real, messy dataset, live data exercise (screen-share with a spreadsheet or sql console), ai-scored case study (e.g. a cohesyve subjective assessment) or structured chart-interpretation scenario. Score it against written criteria you fix before you see any submissions, and weight the criteria that the role actually depends on.

  • Asks clarifying questions about the business question before touching the data
  • Inspects data quality first — nulls, duplicates, outliers, suspicious distributions — instead of trusting the file
  • Chooses the simplest method that answers the question rather than the most impressive one
  • Quantifies uncertainty and states assumptions explicitly instead of presenting point estimates as fact

Paste a job description; Cohesyve generates a role-specific assessment and rubric. Ten candidates free, no card.

When you hire a data analyst, you are not buying a list of tools on a resume — you are buying judgment about which numbers matter and the discipline to get them right. A candidate can list SQL, Excel, and Tableau and still draw the wrong conclusion from a messy table. The only reliable way to know is to hand them realistic, imperfect data and watch what they do with it. This page shows hiring managers how to assess data analysis the way the job actually works: clean the data, find the signal, and explain it to someone who will not read a single line of code.

Why Data Analysis is worth testing

Resumes reward tool familiarity, but the job rewards reasoning under ambiguity. Two analysts with identical stacks can deliver opposite recommendations from the same dataset. Testing surfaces who validates assumptions, spots data quality issues, and translates findings into decisions — versus who produces a pretty chart that does not answer the question. A bad hire here quietly steers real budget and product decisions wrong for months before anyone notices.

What strong Data Analysis looks like

  • Asks clarifying questions about the business question before touching the data
  • Inspects data quality first — nulls, duplicates, outliers, suspicious distributions — instead of trusting the file
  • Chooses the simplest method that answers the question rather than the most impressive one
  • Quantifies uncertainty and states assumptions explicitly instead of presenting point estimates as fact
  • Distinguishes correlation from causation and resists overclaiming from observational data
  • Communicates a clear "so what" and recommended action, not just a description of the data
  • Sanity-checks results against intuition and flags when a number looks too good or too strange to be true

Ways to assess Data Analysis

Take-home analysis on a real, messy dataset

Send a 2,000-50,000 row CSV with intentional flaws (duplicate rows, inconsistent date formats, a few impossible values) and a business question. Ask for cleaned data, 3-4 insights, and a short recommendation. Cap effort at 2-3 hours.

Pros

Mirrors the actual job; reveals data-cleaning instinct and the ability to find a "so what" without a proctor watching.

Cons

Hard to verify authorship; strong candidates may decline unpaid multi-hour work; grading is subjective without a rubric.

Best for Mid and senior analysts where end-to-end judgment matters more than speed.

Live data exercise (screen-share with a spreadsheet or SQL console)

Give the candidate a dataset and 45-60 minutes to answer a question while sharing their screen. Ask them to narrate their reasoning as they clean, aggregate, and interpret.

Pros

You see real working process, troubleshooting, and how they react when the obvious approach fails. Authorship is certain.

Cons

Test anxiety penalizes otherwise strong people; time pressure rewards speed over the careful validation the job needs.

Best for Confirming hands-on fluency and reasoning when a resume claims heavy daily analysis.

AI-scored case study (e.g. a Cohesyve subjective assessment)

Generate a unique, dynamic case from your own job description: candidate interprets a dataset and writes their recommendation in an open-ended response that an AI scores against a structured rubric. Every candidate gets a different case, so answers cannot leak.

Pros

Async and consistent across candidates; cheat-proof because each assessment is unique; scores reasoning and communication, not trivia. A fraction of the cost of live screening, with no question-bank maintenance.

Cons

Open-ended scoring still benefits from a human spot-check on shortlisted finalists; less suited to deep tool-specific debugging.

Best for Screening a large applicant pool fairly before investing interview time.

Structured chart-interpretation scenario

Show 2-3 dashboards or charts and ask what the candidate would conclude, what they would distrust, and what they would investigate next. No tools required.

Pros

Fast (20-30 min), isolates analytical judgment from tool mechanics, and exposes overclaiming or chart misreading.

Cons

Does not test the hands-on cleaning and querying that consume most of the real workday.

Best for Early phone-stage screening and roles heavy on stakeholder reporting.

Cohesyve

Run a Data Analysis assessment on your next opening

Cohesyve generates a unique Data Analysis task per candidate from your job description, with the scoring rubric attached. Questions are different for every applicant, so they cannot be shared or looked up.

What to test

Data cleaning and validation

Whether the candidate finds and handles the quality problems that wreck real analyses before they trust a single number.

Identify and resolve duplicate rows and inconsistent date or currency formats in a provided fileSpot impossible values (negative ages, future order dates) and decide how to treat themReconcile two tables that should match on a key but partially do notExplain how missing values were handled and why that choice will not bias the result

Exploratory analysis and insight generation

Whether they can move from a vague business question to specific, defensible findings.

Given a sales dataset, surface the three drivers of a revenue change last quarterSegment customers and explain which segment to prioritize and whyIdentify a surprising pattern and propose what additional data would confirm it

Quantitative reasoning and statistics

Whether they apply the right basic method and avoid common statistical traps.

Decide whether a difference between two groups is meaningful or likely noiseExplain why a metric average is misleading and what to use insteadSpot a case of Simpson’s paradox or survivorship bias in a result

Communication and recommendation

Whether they translate analysis into a decision a non-technical stakeholder can act on.

Summarize a finding in two sentences for an executive who will not read the appendixRecommend a concrete action and state the risk if the data is wrongChoose the single chart that best supports the recommendation and justify it

Tool fluency (SQL / spreadsheets)

Whether they can actually extract and shape data, not just talk about it.

Write a query joining orders and customers and aggregating revenue by monthBuild a pivot summarizing a metric across two dimensionsUse lookups or a window function to compute a running total or rank

Sample Data Analysis questions

Here is a 10,000-row orders CSV with a business question: "Why did revenue drop in March?" Walk me through your first 20 minutes.

Mid

Look for Strong answers start with data validation (date coverage, nulls, duplicates) and confirm the drop is real before explaining it. Weak answers jump straight to charts.

This column shows an average order value of $4,200 but most orders look like $40. What is going on and what would you do?

Entry

Look for Recognition that a few extreme outliers or a unit/currency error skews the mean; proposes inspecting the distribution and using median or trimming.

A teammate concludes that users who use Feature X churn less, so we should push everyone to Feature X. What is wrong with that reasoning?

Mid

Look for Names correlation-vs-causation and selection bias; suggests an experiment or cohort comparison rather than acting on the observational claim.

Two tables should join cleanly on customer_id but 8% of rows do not match. How do you investigate?

Mid

Look for Checks for type mismatches, trailing whitespace, casing, nulls, and whether the unmatched rows share a pattern — rather than silently dropping them.

Given this dashboard, what would you trust, what would you distrust, and what would you check before presenting it to leadership?

Senior

Look for Healthy skepticism about source freshness, denominator definitions, and date ranges; distinguishes a real trend from a reporting artifact.

Write SQL to return monthly revenue and month-over-month growth for the last 12 months.

Mid

Look for Correct date truncation and grouping, a window function or self-join for the prior month, and handling of the first month’s null growth.

You ran the numbers and the result is exactly what leadership hoped for. What do you do next?

Senior

Look for Increases scrutiny rather than declaring victory — re-checks logic, looks for a confound, and seeks disconfirming evidence.

Explain to a non-technical VP, in two sentences, why our conversion rate "improved" but actual sales fell.

Mid

Look for Clear, jargon-free explanation (e.g. denominator shrank) and the implication for the decision — not a recitation of formulas.

Red flags

  • Trusts the file as-is and never checks for nulls, duplicates, or outliers
  • Presents point estimates with total confidence and never mentions assumptions or uncertainty
  • Confuses correlation with causation or overclaims from a small or biased sample
  • Cannot explain a finding in plain language to a non-technical stakeholder
  • Reaches for the most complex method available instead of the one that answers the question
  • Produces charts that look polished but do not address the actual business question
  • Gets a surprising result and reports it without sanity-checking against reality

Scoring rubric

CriterionWeightWhat strong looks like
Data quality and rigor25%Proactively finds and resolves data flaws, validates before analyzing, and documents how edge cases were handled.
Analytical reasoning25%Chooses an appropriate method, avoids statistical traps, and reaches a defensible, well-supported conclusion.
Insight and business relevance20%Surfaces a non-obvious finding that maps to a real decision, not just a description of the numbers.
Communication20%Explains findings clearly to non-technical readers and recommends a concrete action with stated risk.
Tool fluency10%Extracts and shapes data efficiently in SQL or spreadsheets without fighting the basics.

Mistakes hiring teams make

  • Testing tool trivia ("which function does X?") instead of reasoning on real data
  • Using a clean, perfect dataset that never reveals whether the candidate validates inputs
  • Grading only the answer and ignoring the process and assumptions that produced it
  • Over-weighting chart aesthetics over whether the analysis answers the question
  • Giving a multi-hour unpaid take-home that screens out strong, in-demand candidates
  • Skipping a shared rubric, so different interviewers reward completely different things

Roles that need Data Analysis

Data AnalystBusiness AnalystMarketing AnalystOperations AnalystFinancial AnalystProduct AnalystBI AnalystGrowth Analyst

Common questions

Should I test SQL, Excel, or a BI tool specifically?

Test the tools your team actually uses daily, but weight reasoning above syntax. A candidate who reasons well learns your stack in weeks; one who is fast in your exact tool but cannot validate data or interpret results will mislead decisions. Use the live exercise to confirm fluency, and the case study to test judgment.

How do I stop candidates from using AI or sharing answers on a take-home?

Static take-homes leak quickly. Cohesyve generates a unique, dynamic assessment per candidate from your job description, so no two people see the same dataset or question and there is no answer key to circulate. Because reasoning is scored against a rubric rather than a fixed key, it holds up even when candidates have AI tools.

How long should a data analysis assessment take?

Keep screening under 60 minutes and any take-home under 2-3 hours. Longer tests measure free time and persistence more than skill, and they cause strong candidates to drop out. Save deeper, end-to-end work for paid trial projects once someone is a serious finalist.

Can junior candidates pass the same assessment as senior ones?

Use the same scenario but adjust expectations by level in your rubric. Juniors should clean data and find correct basic insights; seniors should also question the question, quantify uncertainty, and recommend action. Scoring by level lets one well-designed case fairly evaluate a mixed applicant pool.

What is the single biggest predictor of a strong analyst?

Skepticism toward their own data. The best analysts assume the file is wrong until proven otherwise and treat a too-good result as a reason to dig deeper, not celebrate. If a test only rewards getting a number out fast, it misses this entirely — design it to reward validation and clear reasoning.

Cohesyve · Skill assessments for hiring

Test Data Analysis before the first interview

Generate a role-specific Data Analysis assessment from your job description and see who can do the work before you spend interview time on them.

1,500+

assessments completed

50%

faster time-to-hire

90%

completion rate

5 min

from JD to assessment

No credit card · 10 free candidates · Plans sized to your hiring volume

See Cohesyve in action

Free 30-min walkthrough

See it on your role