Data engineer phone screen questions: pipelines, orchestration, warehouses and data quality
On this page
- Data engineer, analytics engineer or platform engineer: place the candidate
- One pipeline, end to end
- Reliability: reruns, backfills and schema changes
- Data quality, governance and cost
- How data engineering resumes overstate the work
- Certifications: issuers and validity
- Logistics, knockout checklist and scorecard
- Questions people ask
A data engineer phone screen should find out which pipelines the candidate built and ran, not which tools sat in their company's stack. Ask for one pipeline end to end: the source, the schedule, the orchestration tool, where the data landed, how much of it there was, who depended on it and the last time it broke. Then ask how they handled a backfill, a schema change from an upstream team and a data quality problem someone else found first. Engineers with real ownership answer with table names, run times and a 6 a.m. failure. Inflated resumes list Spark, Kafka, Airflow, Snowflake and Databricks without a single pipeline they can walk through.
This page is for the people who build data systems. Data analyst phone screen questions covers people who query data to answer business questions, and software engineer phone screen questions covers application developers. If the role trains or serves models, pair this with machine learning engineer screening questions.
Data engineer, analytics engineer or platform engineer: place the candidate
| Role in practice | What the week produces | Question that places them | Common mismatch |
|---|---|---|---|
| Data engineer (pipelines) | Ingestion from sources, batch and streaming jobs, scheduling, monitoring | "Which sources did you bring in, and how did data get from each to the warehouse?" | Analyst who wrote a few scheduled queries |
| Analytics engineer | SQL transformations, data models, tests and documentation, often in dbt | "Did you work on getting data in, or on modeling it once it was there?" | Modeling work described as pipeline engineering |
| Data platform engineer | The warehouse or lakehouse itself, access, infrastructure as code, cost controls | "Did you build the platform other data engineers used, or build on it?" | Pipeline builder described as platform owner |
| Streaming engineer | Event streams, real-time processing, low-latency delivery | "What latency did consumers need, and what did you run in production?" | Consumed from a Kafka topic once, listed as streaming expertise |
| ETL developer (legacy tools) | Jobs in Informatica, SSIS, Talend or DataStage, on-premise databases | "Which tool, how many jobs, and any move to cloud tools?" | Strong on legacy ETL, presented as modern cloud stack |
One pipeline, end to end
This is the most useful ten minutes of the screen. Ask the candidate to pick a pipeline they built and take you through it. Prompt with the questions below only if they skip a part.
| Prompt | What a strong answer sounds like | Red flags |
|---|---|---|
| Where did the data come from? | Named sources: a Postgres production database, Salesforce, a payments API, event logs. | "Various sources." |
| How did it get extracted? | A method with a reason: change data capture, an API with pagination, a managed connector such as Fivetran, files dropped in storage. | Does not know how extraction worked. |
| How often did it run, and what scheduled it? | A schedule and a tool (Airflow, Dagster, Prefect, a cloud scheduler), plus dependencies between jobs. | Ran it by hand when someone asked. |
| How much data, and how long did a run take? | Rows or gigabytes per run and a run time, with any change as data grew. | "Big data," no numbers. |
| Where did it land, and how was it modeled? | The warehouse or lakehouse (Snowflake, BigQuery, Redshift, Databricks) and the model: raw, cleaned, business tables. | Everything in one giant table with no layers. |
| Who used it, and what did they do with it? | Named consumers: finance dashboards, a machine learning feature, a customer-facing report. | Does not know who used it. |
| When did it last fail, and what did you change? | A specific failure: an API rate limit, a late file, a schema change, duplicate rows, and the fix. | It never failed. |
Reliability: reruns, backfills and schema changes
Pipelines that are run once are easy. The skill is in running them every day for years while sources change underneath. These questions separate the two.
- "If a job ran twice by mistake, what happened to the data?" Strong: the job was designed to be safe to rerun, for example by overwriting a date partition or merging on a key, and they can say how. Red flag: duplicates, discovered by a stakeholder.
- "Tell me about a backfill you ran. How far back, and how did you check it?" Strong: scope, method, run time, and a reconciliation against the source. Red flag: never reprocessed history.
- "What happened the last time an upstream team changed a column without telling you?" Strong: how they found out, the fix, and a contract, alert or schema test they added. Red flag: learned about it from a broken dashboard, and nothing changed afterwards.
- "How did you know a pipeline had failed, or silently delivered wrong data?" Strong: alerting on failures and on freshness or row counts. Red flag: users reported problems first, every time.
- "Did your pipelines have a delivery time someone depended on?" Strong: "Finance tables by 7 a.m. for the daily close," and what they did when it slipped. Red flag: no expectations anyone could name.
- "Were you on call for data pipelines?" Strong: rotation details and a recent incident. Red flag: none, for a role that includes it.
Data quality, governance and cost
| Question | What a strong answer sounds like | Red flags |
|---|---|---|
| What tests ran on your data, and what did they catch? | Specific checks (uniqueness, not-null, accepted values, row count ranges) in dbt tests, Great Expectations or custom code, and a real catch. | No tests; "the analysts check it." |
| Tell me about a number that did not match another team's number. | A reconciliation story: different definitions, time zones, late records, and how they settled which was right. | Has never been asked to reconcile. |
| How was personal or sensitive data handled in your pipelines? | Masking, restricted schemas, role-based access, and deletion requests handled through the pipeline. | Everyone could query raw customer data. |
| What did your warehouse or cluster cost a month, and what did you do about it? | A rough figure and a measured saving: clustering or partitioning, smaller warehouses, incremental models instead of full rebuilds. | Never saw the bill. |
| How was pipeline code reviewed and deployed? | Git, pull requests, CI running tests, separate development and production environments. | Edited jobs directly in production. |
Worked example: checking a cost claim (invented numbers)
The resume says "cut Snowflake costs 60%." You ask for the numbers.
- Monthly spend before: $50,000. After their change to incremental models: $38,000. That is 12,000 ÷ 50,000 = 24 percent.
- Ask: "Where did the rest come from?" A common answer is a contract renegotiation or a team shutting down unused workloads.
A 24 percent saving from their own engineering is strong. Record the real number and who did the rest.
How data engineering resumes overstate the work
| Pattern | What it looks like | How to check |
|---|---|---|
| Stack list | Spark, Kafka, Flink, Airflow, dbt, Snowflake, Databricks, BigQuery | "Which three did you use every week last year?" |
| "Big data" | "Processed petabytes" | "How much did your own pipeline process per run?" |
| Streaming by association | "Real-time pipelines with Kafka" | "Did you run the consumers in production, and what latency did users need?" |
| Platform credit | "Built the data lake" | "What existed before you, and what exactly did you add?" |
| Analyst as engineer | Scheduled SQL queries described as pipelines | "How did data get into the warehouse in the first place?" |
| Migration presence | "Migrated from on-premise to cloud" | "Which pipelines did you move, and how did you validate them?" |
Certifications: issuers and validity
| Certification | Issuer | Notes, as of September 2026 |
|---|---|---|
| AWS Certified Data Engineer – Associate | Amazon Web Services | Covers ingestion and transformation, data stores, operations and governance on AWS; valid for three years, per AWS. |
| Professional Data Engineer | Google Cloud | Google Cloud certifications are valid for two years, with renewal options, per Google Cloud. |
| Databricks Certified Data Engineer Associate and Professional | Databricks | Valid for two years, renewed by retaking the current exam or a renewal exam, per the Databricks certification FAQ. |
Certifications are rarely required, but they are a reasonable signal for the client's platform. A current certification on the platform the client runs, plus a pipeline story, is the strongest combination. A stack of certifications across every platform with no production pipeline is common among career changers and is a reason to ask more questions, not to reject.
Logistics, knockout checklist and scorecard
| Question | What a strong answer sounds like | Red flags |
|---|---|---|
| The team is on a rotation for pipeline failures before 8 a.m. Can you commit to that? | A clear yes or limit. Describe the schedule; do not ask about family to judge it. | Surprised by any early-morning responsibility. |
| What salary or rate do you need, and on what basis? | A number and a basis. See salary expectation questions. | Only a salary-site figure. |
| Where else are you interviewing, and how far along? | Stages and dates. | "Nowhere," then an offer deadline appears. |
| Are you legally authorized to work in the US, and will you now or in the future need visa sponsorship? | A direct answer to both. | None. Ask everyone the same two questions. |
Knock out, or flag to the client before submitting, if:
- The candidate cannot describe one pipeline end to end, including a failure.
- The client's warehouse or orchestration tool is a must-have and the candidate has never used it in production.
- The role involves sensitive data and they have never worked under access controls.
- The role includes on-call and they cannot commit to the rotation.
| Area | 1 | 2 | 3 | 4 |
|---|---|---|---|---|
| Pipeline ownership | Used data others delivered | Changed existing jobs | Built and ran pipelines end to end | Designed patterns other engineers follow |
| Reliability | Found out from users | Basic failure alerts | Safe reruns, backfills, freshness checks | Owned delivery commitments and reduced incidents |
| Data quality and governance | No tests | Some tests | Tests that caught real problems, access controls | Set quality and access standards |
| Cost and scale | Unaware | Aware of volume and spend | Measured optimization | Savings with trade-offs explained |
| Logistics fit | Deal-breaker | Two open questions | One open question | All aligned |
Write the pipeline story into your notes in the candidate's words. "Built the Stripe-to-Snowflake pipeline; 'we merged on charge ID so reruns never duplicated'" tells a data engineering manager more than "strong ETL experience." Interview Signal attaches quotes like that to each score from the call.
Questions people ask
What is the difference between a data engineer and a data analyst?
A data analyst uses data to answer business questions and presents findings. A data engineer builds and runs the pipelines, tables and platforms that make the data available and trustworthy. Some analytics engineers sit in between, writing transformations in SQL tools such as dbt. Ask whether the candidate's week produced answers or data other people used.
How can a non-technical recruiter test pipeline experience?
Ask the candidate to describe one pipeline end to end: the source, how often it ran, what tool scheduled it, where the data landed, who used it, and the last time it failed. People who own pipelines know the failure story and what they changed after it.
Do data engineering certifications expire?
Most do. As of September 2026, AWS says its Data Engineer Associate certification is valid for three years, Google Cloud certifications are valid for two years, and Databricks certifications are valid for two years. Check the expiry date on the issuer's badge.
Should a data engineer screen include a coding test?
Leave coding and SQL tests to the hiring team. The recruiter screen should confirm ownership, scale, tools and logistics, so the technical interview is spent on candidates who have built what the role needs.