Why pre-screen data analysts before the technical exercise
Analyst pools are large and the resumes are indistinguishable, because the tools are the same everywhere. What differs is judgement: whether someone questions a number that looks too good, whether they know the data well enough to spot a broken join, and whether they will tell a stakeholder that the analysis does not support the conclusion they wanted. A short screen surfaces that before you spend anyone's afternoon on a take-home exercise.
What actually matters when screening Data Analyst candidates
- 01
Technical proficiency
Check SQL fluency beyond basic joins: window functions, CTEs, query tuning on large tables, plus dbt, Python (pandas), and BI tools like Looker, Tableau or Power BI.
- 02
Systems and trade-offs
Probe how they modelled messy source data: handling duplicate events, slowly changing dimensions, late-arriving records, and choosing between a one-off query and a governed metric definition.
- 03
Evidence and rigour
Test analytical rigour: sample sizing, significance testing on an A/B experiment, confounders, and how they validated numbers before a metric reached an executive dashboard.
- 04
Collaboration and communication
Assess how they turn analysis into decisions: stakeholder scoping of ambiguous requests, presenting findings to non-technical leaders, and handling pushback when data contradicts a favoured plan.
Pre-screening questions to ask Data Analyst candidates
12 questions grouped by what they test. Ask the same set in every screen and score answers on a consistent scale, or send them as an async video screen and compare answers side by side.
Analysis acted on
3 questions01Can you describe a time when you had to use data to make a decision?
Listen forA decision that changed because of their analysis, with what the data showed and who acted on it.
Analysis delivered with no decision attached, or findings that only confirmed what was already believed.
02Can you describe a project where you used large datasets?
Listen forReal volumes with the approach that made it workable, and what they had to change when the data outgrew their first method.
Scale claimed with no method, or every dataset small enough to open in a spreadsheet.
03Can you describe your experience with predictive modelling?
Listen forA model with the evaluation approach stated and an honest view on whether it was ever used in production.
Accuracy quoted with no holdout or baseline, or models built that nobody deployed or monitored.
SQL depth
4 questions04What is your proficiency level in SQL?
Listen forDepth shown through a specific query problem, such as a join that duplicated rows or an aggregation at the wrong grain.
A high self-rating with no example, or discomfort with window functions and multi-table joins.
05What is your experience with Python or R for data analysis?
Listen forAnalysis they wrote themselves with a view on when scripting beats a query, tied to a specific piece of work.
Libraries named with no analysis behind them, or code copied from tutorials without understanding.
06What data visualisation tools have you used?
Listen forCharts chosen for the question being answered, with an example of removing a visual because it misled people.
Dashboards built to display everything available, or chart types chosen for appearance.
07What is your experience with data warehousing?
Listen forUnderstanding of how the tables they query are built, including grain and what a slowly changing dimension does to a count.
No idea where their tables come from, or joins made across grains without noticing the duplication.
Checking the data
3 questions08How do you ensure the accuracy of your data?
Listen forSanity checks run before trusting a result, such as row counts and totals reconciled against a known source.
Accuracy assumed because the query ran, or no check against an independent source.
09Can you explain the process of data cleansing and its importance?
Listen forCleaning decisions documented rather than made silently, with an example where a cleaning choice changed the answer.
Outliers removed by default, or cleaning rules applied with no record of what was excluded.
10How do you handle missing or inconsistent data?
Listen forThe reason for missingness investigated before deciding, with awareness that absence can itself be informative.
Missing rows dropped as a first step, or gaps filled with an average without checking why they are missing.
What it does not prove
2 questions11How do you handle large data sets?
Listen forWork pushed to the database rather than pulled into memory, with sampling used deliberately and its limits stated.
Everything extracted to a local file, or samples treated as representative with no check.
12How do you maintain data security and confidentiality?
Listen forA clear line on personal data, including what they will not extract and how findings are shared without exposing individuals.
Personal data extracted to local files, or results shared at a grain that identifies individuals.
How to score responses
Score every candidate on the same four criteria immediately after the screen. At this stage you are shortlisting for panel interviews, not making the final call.
Technical proficiency
35%5Writes complex SQL from memory, names warehouse specifics (Snowflake, BigQuery), and describes dbt models or dashboards they built and maintained.
Systems and trade-offs
25%5Explains schema and metric definition choices with trade-offs, and cites a case where a shortcut later caused rework or reconciliation pain.
Evidence and rigour
25%5Distinguishes correlation from causation with a real example, quantifies uncertainty, and describes QA checks that caught an error before publication.
Collaboration and communication
15%5Names a decision that changed because of their analysis, and describes reframing a vague request into a measurable question.
Analyst resumes list the same three tools and say nothing about judgement. A one-way video screen asks what their analysis got wrong and how they found out.
Try it on HirevireScreening FAQ
Process basics
How long should a pre-screening round for a data analyst take?
Ten to fifteen minutes across eight to ten questions, answered async. Enough to hear one analysis that changed a decision, test their approach to data quality, and check how they handle an inconvenient finding.
Should this replace a SQL test?
No, it goes before one. The screen removes candidates who cannot describe their own analysis, which is a large share of the pool, so the technical exercise is only marked for people likely to pass it.
Evaluating answers
What is the strongest signal when screening a data analyst?
An analysis that turned out to be wrong. Analysts who check their own work can describe how they found the error and what they changed. Anyone whose analyses have all held up has not looked back at them.
How do I judge SQL claims without testing them here?
Ask about a query they had to rewrite. Real answers involve a join that duplicated rows or an aggregate at the wrong grain. Anyone who rates their proficiency high but has no such story has written simple queries.
Go deeper on this role
Sanat has been hiring since 2012 and watching the recruitment industry change up close ever since, and turned that screening process into Hirevire's video screening platform. LinkedIn
