How to assess · For hiring teams
How to Assess Data Modelling Skills When Hiring
The test formats that actually work for Data Modelling, what a strong answer looks like, sample questions and a scoring rubric you can use as-is.
The short answer
Assess Data Modelling with a task, not a conversation: model a domain from a brief, diagnose a broken model, ai-scored assessment (e.g. cohesyve) or metric definition exercise. Score it against written criteria you fix before you see any submissions, and weight the criteria that the role actually depends on.
- States the grain of every table and can say what one row represents
- Separates facts from dimensions and knows which measures are additive
- Chooses surrogate keys where natural keys are unstable and can explain why
- Handles slowly changing dimensions deliberately, with a chosen strategy per attribute
Paste a job description; Cohesyve generates a role-specific assessment and rubric. Ten candidates free, no card.
Data modelling is the decision that everything downstream inherits. Get the grain of a fact table wrong and every report built on it double-counts; choose keys badly and joins fan out; ignore how a dimension changes over time and history quietly rewrites itself. It is also the skill most often assumed rather than tested, because every data professional has built tables. This page covers how to assess data modelling for analytics engineering, data engineering and BI roles: grain and keys, dimensional design, handling change over time, and the judgement to model for the questions the business actually asks.
Why Data Modelling is worth testing
Modelling mistakes are the most expensive in data because they are the hardest to fix once populated and depended on. A warehouse with ambiguous grain produces confident, contradictory numbers, and every analyst learns to distrust it. Testing shows whether a candidate can state what one row means, and that single discipline predicts most of the rest.
What strong Data Modelling looks like
- States the grain of every table and can say what one row represents
- Separates facts from dimensions and knows which measures are additive
- Chooses surrogate keys where natural keys are unstable and can explain why
- Handles slowly changing dimensions deliberately, with a chosen strategy per attribute
- Models for the questions that will be asked, not for the shape of the source
- Prevents fan-out joins by design and can spot one in a query
- Documents definitions so a metric means one thing everywhere
Ways to assess Data Modelling
Model a domain from a brief
Describe a business — subscriptions with plan changes, orders with returns — and ask for the tables, their grain, keys and how history is kept. Forty-five to sixty minutes, on paper or in a diagram.
Pros
Cons
Best for Any modelling role.
Diagnose a broken model
Provide a schema and a query that double-counts revenue because of a fan-out join and an ambiguous grain. Ask the candidate to find and fix the modelling cause.
Pros
Cons
Best for Mid and senior roles.
AI-scored assessment (e.g. Cohesyve)
Generate a data modelling task from the job description — a domain to model, a schema to critique, a change-over-time question — with a rubric. Each candidate receives a different variant; the reasoning is scored alongside the work.
Pros
Cons
Best for Screening an applicant pool fairly before interview time is spent.
Metric definition exercise
Ask them to define a metric such as active customers precisely enough that two analysts would get the same number.
Pros
Cons
Best for Analytics engineers and BI leads.
Cohesyve
Run a Data Modelling assessment on your next opening
Cohesyve generates a unique Data Modelling 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
Grain and keys
Whether tables mean something precise.
Dimensional design
Whether facts and dimensions are separated well.
Change over time
Whether history is preserved correctly.
Modelling for questions
Whether the model serves the business.
Sample Data Modelling questions
What does "grain" mean, and why does it matter?
EntryLook for What one row represents; everything about correctness and joins follows from it.
Revenue doubles when you join orders to shipments. Why, and how do you fix it in the model?
MidLook for One-to-many fan-out; aggregate to the right grain first or model shipments as their own fact.
A customer changes region. Reports for last quarter should show the old region. Model it.
MidLook for SCD type 2 with effective dates, surrogate key, facts referencing the version at the time.
When would you use a surrogate key over a natural key?
MidLook for Unstable or composite natural keys, history tracking, source changes; the cost is an extra lookup.
Define "monthly active customer" so that finance and product get the same number.
SeniorLook for Precise activity definition, time window, timezone, inclusion rules, and where the definition lives.
Red flags
- Cannot state the grain of a table they designed
- Models tables as copies of source systems
- Has no strategy for dimensions that change
- Does not recognise a fan-out join
- Metric definitions live in people's heads
Scoring rubric
| Criterion | Weight | What strong looks like |
|---|---|---|
| Grain and keys | 35% | Every table has a stated grain and correct keys. |
| Dimensional design | 25% | Facts and dimensions are clean; additivity is understood. |
| Change over time | 20% | History is preserved deliberately. |
| Fitness for questions | 20% | The model answers what the business asks. |
Mistakes hiring teams make
- Testing SQL syntax rather than modelling decisions
- Not asking about grain — the single most telling question
- Accepting a model that mirrors the source
- Skipping change-over-time scenarios
- Assuming years of SQL means modelling skill
Roles that need Data Modelling
Common questions
Is dimensional modelling still relevant with modern warehouses?
Yes. Cheap compute forgives some denormalisation, but grain, keys and history handling still determine whether numbers are right. The discipline matters more than the exact pattern.
What is the best single data modelling question?
Ask for the grain of a table and what one row represents. Candidates who answer precisely tend to get everything else right.
Can I assess modelling without a warehouse?
Yes. It is a design skill; paper and a diagram are enough. Confirm SQL fluency separately if the role needs it.
How long should a modelling assessment take?
Forty-five to sixty minutes for a domain-modelling exercise.
Cohesyve · Skill assessments for hiring
Test Data Modelling before the first interview
Generate a role-specific Data Modelling 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
From the blog