How to Hire an ETL Developer: 2026 Guide

Last updated: August 8, 2026

By Tom Kenaley, Co-Founder and President, KORE1

An ETL developer builds and maintains the pipelines that move data between systems, and in 2026 most U.S. hires land between $103,000 and $185,000 depending on whether the job is running a modern stack or escaping a legacy one. That second clause is doing almost all the work in the sentence. Two people with the same title, the same years of experience, and a $45,000 gap between them, because one of them can get you off Informatica and the other has never seen it. Same title. Different job.

Which one you need is a question about your estate, not about your req.

One date reshaped this market. Standard support for Informatica PowerCenter 10.5 ended on March 31, 2026, and the vendor’s own modernization guidance now points customers toward the cloud platform or the door. Extended support buys a year. Sustaining support after that keeps the lights on without new critical fixes. Meanwhile Salesforce closed its roughly $8 billion acquisition of Informatica on November 18, 2025, which is a fine outcome for shareholders and a source of genuine uncertainty for the VP of Data at a mid-market insurance carrier trying to decide whether to renew.

A lot of companies opened an ETL req in the last nine months. Most wrote it wrong.

One thing to know about who is writing this. KORE1 has been placing data talent since 2005, we run an ETL developer staffing desk, and nothing bills on our side until somebody signs an offer. Further down I tell a decent share of you to hire a contractor for a quarter instead of running a permanent search, which is the smaller invoice by a wide margin. The paragraph after that says promote the SQL analyst you already employ, which is no invoice at all. I keep those sections in because the alternative is watching a client pay $160,000 a year for someone to rebuild something a specialist could have finished by Labor Day. Fair trade.

What an ETL Developer Actually Owns

An ETL developer owns the movement of data between systems: extracting it from source applications, transforming it into a usable shape, loading it into a warehouse, and keeping all three steps running on schedule. They write the pipeline code, the tests that catch bad data, and the recovery logic that runs when a source system changes without warning.

That clause is the job. Everything before it goes smoothly.

Read fifty ETL job descriptions and roughly forty-five of them describe the happy path. Build pipelines from source systems to the data warehouse. Write SQL. Work with stakeholders to understand requirements. Nothing about failure. Nothing in there addresses what happens when Salesforce adds a field on a Tuesday, the load silently drops nine thousand rows, and the finance team builds a forecast on top of the gap for six days before anybody notices. Handling that is the difference between a mid-level engineer and a senior one, and it never appears in the posting.

Four Jobs, One Title

The phrase “ETL developer” now covers at least four distinct roles that share almost no day-to-day overlap. Companies write one req that asks for all four, get a candidate pool of people who match none of them, then conclude the market is tight.

The market is fine. The req is the problem. Every time.

Pick one. Write the req against that one. You can hire the second lane in six months once the first fire is out, and the candidate you attract with a focused posting is meaningfully stronger than the generalist you attract with a wish list.

The Keyword That Filters Out Your Best Candidate

This one costs companies real money and almost nobody sees it happening.

You have a PowerCenter estate. You need off it. So the recruiter screens for Informatica, the ATS ranks on Informatica, and the shortlist comes back full of people who have spent nine years inside PowerCenter and nowhere else. Predictable result. Those candidates know your current system beautifully. Ask them to design the Snowflake and dbt target architecture and roughly a third of them will rebuild PowerCenter in SQL, mapping by mapping, because that is the shape their brain makes when it thinks about data movement.

The engineer you want has done both. Five years in a legacy tool, then three or four years on a cloud stack, and a migration on the resume where they can tell you what they got wrong the first time. That profile ranks low on a keyword screen because Informatica appears once, in a job that ended in 2021.

A manufacturer outside Columbus spent five months on this exact loop last year. Two Informatica specialists hired, both competent, and eleven months later the target warehouse held 340 dbt models that were transliterated PowerCenter mappings, nested six deep, with the business logic still buried in the same undocumented places. They did not migrate. Expensive lateral move. They relocated the problem and paid a cloud bill for the privilege. The fix took a third engineer at $178,000 and another seven months.

Screen on the migration, not the tool.

What the Money Looks Like in 2026

Compensation for this title is unusually noisy, and the reason is structural rather than statistical. “ETL developer” is a legacy title that skews toward legacy work, so the trackers pick up a lot of maintenance roles at established enterprises. The people doing the same work on a modern stack usually have “data engineer” on their badge and get counted somewhere else entirely.

The federal numbers are worth a minute because they cut against the panic narrative. The Bureau of Labor Statistics put database administrators at a $104,620 median in May 2024 and database architects at $135,980, with the combined category projected to grow 4% through 2034 and about 7,800 openings a year. Four percent is ordinary. It is not a collapsing field, and it is not an exploding one, which means anyone quoting you a 30% premium on scarcity grounds should be asked to show their work.

Aggregator numbers to compare against are on Glassdoor, ZipRecruiter, and Salary.com. Read all three or none of them. Pick a lane.

Our own placements sit outside all four of those numbers. A mid-level engineer with three years on a cloud stack closes at $118,000 to $142,000 in most metros. A senior who owns orchestration and carries the pager runs $145,000 to $185,000. Migration leads with a real cutover behind them clear $195,000 in California, New York, and Boston, and we have seen $210,000 twice this year on searches with a hard regulatory deadline attached. Contract is where the legacy premium shows up most clearly. Senior pipeline engineers bill $85 to $135 an hour, while a straight SSIS maintenance contractor averages closer to $55 an hour nationally, which tells you exactly what the market thinks maintenance is worth versus escape. Our salary benchmark tool will price a band against your own metro, and for the adjacent titles our data engineer salary guide and senior data engineer bands track the wider market.

The Reliability Gap Nobody Budgeted For

There is a genuinely useful number in the 2026 dbt Labs State of Analytics Engineering report, which surveyed 363 practitioners and leaders. Seventy-two percent of teams now prioritize AI-assisted coding. Twenty-four percent prioritize AI-assisted pipeline management, meaning testing, observability, and quality control.

Sit with that ratio.

Three quarters of the industry has accelerated how fast transformations get written. One quarter has invested in knowing whether the output is correct. Nobody funded the checking. The same report found 71% of respondents worried about incorrect or hallucinated outputs reaching stakeholders, 57% reporting higher warehouse and compute spend, and only 36% seeing team budgets grow. More code, more compute, same headcount, and trust in data jumped to an 83% priority from 66% the year before.

That gap is the business case for this hire, and it is a better one than “we need more pipelines.” Your analysts can already produce a transformation in an afternoon with an AI assistant. What nobody on the team currently owns is whether the number at the end of it is right, and whether it will still be right after the source system pushes a release on Thursday.

Worth naming the stack while we are here, since generic postings attract generic candidates. Airflow still carries the most volume in orchestration, with Dagster and Prefect growing among teams that got tired of Airflow’s edges. Fivetran and Airbyte handle managed ingestion, Debezium and Kafka Connect handle change data capture, dbt owns transformation, and PostgreSQL sits under more of this than people assume. The 2025 Stack Overflow Developer Survey put PostgreSQL at 55.6% usage, a fifteen-point gap over MySQL. Name your actual tools in the posting. Be specific. Candidates filter on them the same way you do.

Running the Search Without Losing the Quarter

1. Write down the estate before you write the req

One page. What systems hold the data, what tool moves it today, what breaks most often, and what the deadline is. If there is a support cliff on the calendar, name the date in the job description. Right in the posting. Candidates who have done migrations read that date and understand the job immediately, which is a filter working in your favor for once.

2. Pick one lane and say so out loud

Reqs that ask for Informatica, Airflow, dbt, Kafka, Spark, and Snowflake experience get either liars or nobody. Name the two that matter and mark the rest as nice to have. Our fill rates on single-lane reqs run substantially better than on wish lists, and the difference is not sourcing effort. It is that strong engineers self-select out of postings that look confused about their own problem. They can tell.

3. Price against the failure mode

Maintenance of a stable estate is a $110,000 to $135,000 job and should be titled and paid like one. A cutover with a regulatory date behind it is a $170,000-plus job, and pretending otherwise costs you the two months you did not have. Set the band against what happens if the work fails, not against what the title averages. Deadlines cost more.

4. Ask the backfill question

Best question in the loop. Ask it early. “You discover a transformation has been wrong for eleven days. Walk me through what you do.” A senior engineer immediately asks what downstream consumers already read the bad data, then talks about reprocessing without double-counting, then mentions telling the business before fixing it. A mid-level engineer starts rewriting the transformation. Neither answer is wrong for their level. Only one of them can be trusted with your revenue tables.

5. Replace the take-home with something broken

Skip the four-hour assignment. Hand over a real failing DAG or a mapping with an actual bug, redacted lightly, and screen-share for thirty minutes while they work it. You will learn more about how someone debugs under mild pressure than any polished submission will show you, and the completion rate on a thirty-minute live session beats a take-home by a wide margin because good candidates with three offers do not do homework. Nobody has time.

6. Close fast, because the deadline will not move

Our average time-to-hire across IT roles is 17 days, and our 12-month retention on direct-hire placements sits at 92%, but neither number helps a company that takes eleven days to schedule a second round. Migration candidates are usually interviewing for two or three of these at once right now. Move faster. Two rounds, decision inside a week, offer within 48 hours of the final. For a dated project, contract staffing gets a body on the problem in days rather than weeks, and you can convert later.

Three Situations Where You Should Not Run This Search

Three of these cost us the search. Listing them anyway.

If the work has a hard end date, hire a contractor. A migration is a project, not a headcount. We placed a senior Informatica-to-Snowflake specialist at a regional bank in Tampa on a fourteen-week contract at $128 an hour, which came to about $71,000 all in. The permanent alternative on the table was $175,000 a year plus fee, for a person whose entire mandate would have evaporated in April. The bank kept two of their own analysts alongside the contractor and those two now run the stack. That is the outcome you want. Our invoice was smaller and I am fine with that.

Promote when the domain knowledge outweighs the tooling. The strongest pipeline hire I have seen in the last two years was a SQL-heavy reporting analyst at a Milwaukee logistics company who already knew which of their four order tables was authoritative and why. Five weeks to pick up dbt. Context beats tooling. An external senior would have needed two quarters to acquire what that analyst already had, and would have spent the first one confidently building on the wrong table.

Do not hire at all if the actual complaint is modeling rather than movement. When the pipelines run fine and the argument is about what “active customer” means, that is a different req entirely, and probably an analytics engineer. We wrote up the distinction in data engineer versus analytics engineer because the two reqs get swapped constantly and the swap wastes a full quarter every time.

What Comes Up on the First Call

Can one hire cover the legacy estate and the new stack?

Sometimes, and that person costs about 25% more than either specialist alone. Engineers fluent in both PowerCenter-era tooling and a modern cloud stack are genuinely rare, which is why they price at a premium.

When the migration is under roughly 200 mappings, one strong dual-fluency engineer beats two specialists, mostly because the handoff between them is where migrations lose their timeline. Above that, split it. Name an owner. Give the migration lead the legacy side and hire a cloud-native engineer to own the target architecture, and make one of them accountable for parity testing rather than assuming it emerges from cooperation.

Our team says AI writes the transformations now. Is the role going away?

No. AI drafts transformation code well and owns none of the consequences, which is why 72% of data teams prioritize AI-assisted coding while only 24% prioritize AI-assisted pipeline management.

What the assistant cannot do is decide how to reprocess nine days of double-counted revenue, or notice that a source system’s definition of a field quietly changed, or be the person accountable at 6 a.m. when the executive dashboard is empty. The role shifted. It did not shrink. Less time in a drag-and-drop canvas, more time writing tested code that other engineers review, which honestly raises the bar on who you should be hiring.

Contract or direct hire for a migration?

Contract, in most cases. A migration has an end date, and hiring permanent headcount for finite work leaves you with a salary and no mandate once the cutover clears.

Direct hire makes sense when the migration is the beginning of a permanent data platform team rather than a one-time escape. Read your roadmap honestly. If there is a second and third phase already funded, hire permanently through direct hire. If the plan ends at cutover, contract and convert only if the person turns out to be someone you want anyway.

How do we interview for this without a take-home?

Give them something already broken. A thirty-minute screen-share debugging a real failing pipeline predicts on-the-job performance better than any assignment we have watched clients run.

Add one architecture conversation and one question about a decision they regret. Three signals total, two rounds, done. Our data engineer interview questions guide has the specific prompts, and most of them transfer directly to an ETL search with the vocabulary adjusted.

Every strong resume says “data engineer.” Are we posting the wrong title?

Probably, if you are on a modern stack. Engineers who work in dbt, Airflow, and Snowflake overwhelmingly identify as data engineers, and an “ETL developer” posting reads to them as legacy maintenance.

Post it as data engineer, or as ETL and data pipeline engineer if you want both pools. Test both. Keep the ETL title only when the estate is genuinely legacy and you want the people who are proud of that experience. Our data engineer staffing desk sources against both titles on the same search for exactly this reason, and the response rate difference between the two postings is not subtle.

What does a bad hire here actually cost?

Between $180,000 and $300,000 once you count salary, recruiting, and the rework, and considerably more if a migration deadline slips as a result.

The Columbus manufacturer above is the honest illustration. Two salaries and eleven months of work produced a warehouse that had to be substantially rebuilt, and the second attempt cost another $178,000 and seven months. Nobody put the compute spend on that estimate, either. It was not small. Bad pipelines are expensive to run as well as to fix.

Start With the Estate

Nearly every ETL search that goes badly went badly before anyone opened a resume. The req described four jobs, the screen ranked the wrong keyword, and the band was priced against a title instead of a deadline. Fix those three things and the search is unremarkable, which is the goal. Nothing exotic about it.

Write down what you have. Decide whether you are escaping it or running it. Then hire for that.

If you want a second read on which version of this hire your situation calls for, talk to our data recruiting team. Our recruiters on this desk average more than fifteen years placing pipeline talent across 30-plus U.S. metros, and we will tell you when the answer is a contractor or the analyst already sitting in your building.