Last updated: November 16, 2026
By Mike Carter, Managing Director, KORE1
A data warehouse engineer job description should name the platform and the modeling method, the grain of the central fact tables, who pays the cloud bill, and a base range of roughly $130,000 to $200,000 by level. Most postings ask for “strong dimensional modeling” and stop there. That tells a candidate nothing about the model they would inherit.
Three reports, one Monday morning, three answers to the same question. How many claims are open?
Here’s a composite, details changed. A property and casualty insurer outside Hartford, Connecticut, about 2,100 employees, kept its reporting in a SQL Server 2016 warehouse fed by SSIS packages, with a Snowflake pilot running beside it. Claims said 41,200 open claims. Actuarial said 57,800. Finance, counting payments, came back with a third figure. Nobody in the meeting could say which number was wrong. Nobody could say what one row in the main claims table stood for, either.
The posting for its first dedicated warehouse engineer read like most of them. Seven years of experience. Snowflake, dbt, Airflow. “Strong dimensional modeling.” That was the whole spec. No pay range, no platform dates, and not a word about the old SQL Server box passing its end-of-support date in July, so the finalist we liked, who had spent four years at a Boston health insurer, ended up asking one question on the first call. “What does a row in your claims fact table represent?”
Silence. Then the hiring manager said, “A claim.” The data lead, on the same line, said, “Sort of. Per coverage.”
She took another offer. Fair enough, since the first job in that seat was to write down a sentence the company had never written, and the posting hadn’t warned her.
Where I sit, briefly. We get paid when a company hires a warehouse engineer through our data warehouse engineer staffing desk, so a posting that fills itself costs us business. I’d still rather you wrote a good one. A bad one sends a pile of wrong résumés to my recruiters first, and they read every one. For the search itself, our guide to hiring a data warehouse engineer covers the interview loop, and the data warehouse engineer salary guide prices the seat by level, so between the two you can run the search without guessing at either half. This page is the posting. If the real job is pipelines, the data engineer job description template fits better, and if it’s mostly dbt models for analysts, try the analytics engineer job description template.

What the Seat Owns, in Posting Terms
A data warehouse engineer designs, builds, and runs the analytical store a company reports from. That covers the dimensional model, the load logic, query speed, access rules, and the cloud bill. The seat sits between the team that delivers raw data and the analysts who answer to finance.
Four neighbors share pieces. Pipelines belong to the data engineer, the dbt layer to an analytics engineer, the shape of the whole estate to an architect, which our data architect job description template lays out, and one platform’s internals to a specialist, covered in the Snowflake engineer and Databricks engineer templates. A draft that pulls duties from all four is a team budget wearing one job title. The middle is theirs. The model, and the meter.
Name the Modeling Method, Because Candidates Have Loyalties
People who build warehouses for a living tend to hold a view. Some came up on Kimball’s star schemas, some on a normalized enterprise warehouse, some on Data Vault, and a growing number on wide tables in a medallion layout, the bronze, silver, and gold layers Databricks describes in its documentation. A candidate reads your posting looking for their method. If it isn’t there, they assume you haven’t picked. Fair assumption.
| What you run | What a candidate expects to find | A line you could print |
|---|---|---|
| Dimensional (Kimball) | Fact tables at a declared grain, conformed dimensions shared between them, history kept on slowly changing dimensions | “Star schemas on Snowflake, 14 conformed dimensions, Type 2 history on customer and policy.” |
| Normalized enterprise warehouse | A central store in third normal form feeding departmental marts, usually built before the cloud | “Normalized core in SQL Server with six marts, moving to Snowflake over two years.” |
| Data Vault | Hubs, links, and satellites in a raw layer, with something friendlier on top for people to query | “Data Vault 2.0 raw layer, star schemas in the presentation layer.” |
| Wide tables on a medallion layout | Bronze, silver, and gold tables, few joins, models written in dbt | “Medallion layout on Databricks, gold tables built in dbt.” |
| A mix nobody wrote down | A sharp question in the first interview | “Stars in the finance marts, wide tables everywhere else, and we’ll mark which is which in week one.” |
Not a failure, that last row. I haven’t counted, but it’s the one I run into most, and the posting that admits it draws better finalists than the one pretending to be the first row.
Print the Grain of Your Biggest Fact Table
Ralph Kimball put a price on this mistake in a 2003 design tip. He had debugged “literally thousands of dimensional designs” from his students, and the most frequent error “by far” was not declaring the grain of the fact table at the start. That tip is 23 years old. The definition fits in one line. “Declaring the grain means saying exactly what a fact table record represents.” One of his own examples is an individual transaction against an insurance policy. The four-step design process on the same site puts that declaration second, right after choosing the business process and before anything else gets built.
Back to Hartford. In the composite, the claims team counted claims. Simple. Actuarial’s table held one row per claim per coverage line, so a claim with collision, liability, and glass on it showed up three times, which is just what happens when nobody writes the grain down and each team builds the table it needs. At about 1.4 coverage lines per claim, 41,200 claims turn into 57,800 rows. Two honest teams, two honest numbers. Neither was wrong.
Printing the grain does two jobs at once. A finalist learns the model has an owner’s attention, and the hiring manager gets a sentence to test the team against before the job goes live, which costs one meeting. If three people give three versions, you’ve found the first quarter’s work. Better to find it now.
Some other lines worth printing beside it, none of which you need all of.
- One row per order line, per policy transaction, or per subscription per day. Pick the real one.
- How many conformed dimensions exist, and which one nobody owns.
- Whether history is kept, and since what year. “Type 2 on customer since 2019” is something a candidate can plan around.
- The share of models with a test. “About a third” is a perfectly good answer.
- Where the metric definitions live, whether in a semantic layer, a BI tool, or one analyst’s head.
Two printed lines beat none. A candidate who asks about the other three has told you something about themselves.
Platform Dates a Warehouse Posting Should Carry
A warehouse seat usually inherits something old. Two support dates passed this year, and plenty of warehouses still run on both products. Both dates are public.
On July 14, 2026, Microsoft said it plainly. SQL Server 2016 had reached end of support, and the company “no longer provides regular security updates or technical support” for it, according to its SQL Server blog. The three suggested paths were Azure SQL, an upgrade to SQL Server 2025, and Extended Security Updates for teams still deciding. Informatica’s download page lists end of support for PowerCenter 10.5.x as March 31, 2026. Same problem, different clock, for any load built on PowerCenter.
Any candidate who reads either date can see the seat has a second job. So say so on the page. The engineer who has retired an estate like that wants it spelled out, and the one who hasn’t should know the first year includes it. Our guide to staffing a data warehouse migration covers who sits on that work phase by phase, and what to do with the old server in the meantime is in our SQL Server 2016 end-of-life guide.

Data Warehouse Engineer Job Description Template
Fill the brackets, delete my parenthetical notes before the posting goes live, and drop any section your seat doesn’t need. I wrote the notes for the person at the keyboard, not for candidates.
Job Title
[Data Warehouse Engineer] (Say “Senior” only if this person will set modeling standards for others. Mostly pipeline code? That’s a data engineer. Mostly dbt models for analysts? An analytics engineer. One platform’s internals? Name the platform.)
The Warehouse You’ll Inherit
- Runs on [Snowflake / BigQuery / Redshift / Databricks SQL / SQL Server], loaded from [N] source systems
- Modeled as [dimensional / normalized core with marts / Data Vault / wide tables on a medallion layout / a mix, described below]
- [N] fact tables, the largest at the grain of “one row per [business event]”
- [N] conformed dimensions, with history kept on [customer, product, policy] since [year]
- Monthly spend of about $[band], held by [title]
- [Any platform date that applies, like end of support for SQL Server 2016 or Informatica PowerCenter 10.5]
What You’ll Own
[The model behind finance’s month-end close and the claims dashboards, plus the loads that feed it and the 6 a.m. deadline they have to hit. Fast queries for about [N] analysts. A voice in what the platform costs each month.] (Name real deliverables. “Support analytics” says nothing. “Own the 6 a.m. finance load” says what a Monday feels like.)
Decisions You Make
- Table and column design for new subject areas, reviewed by [title]
- Which slow queries get rewritten and which get a bigger warehouse
- [Whether to retire tables nobody has read in [N] days, alone or with finance sign-off]
- Who gets access to what, inside the rules [the governance team] writes
What Sits Elsewhere
[Source extracts belong to data engineering, and dashboards to whoever asked for them. Governance writes the access policy. You carry it out.] (Candidates ask where the job ends, so tell them before they have to.)
What Success Looks Like by Month Twelve
- By day 90, a written grain for every fact table, agreed by the teams that use it
- [One reliability result, such as the 6 a.m. load finishing on time 95% of days]
- [One cost result, such as spend held flat while query volume grows]
- [A migration milestone, say the first subject area off [SQL Server 2016 / PowerCenter] and onto [platform]]
Background We’re Looking For
- Has designed a dimensional model, or a close cousin of one, and can explain why they chose its grain
- Writes SQL well past joins and CTEs, with window functions and the habit of reading a query plan
- Has run dbt or a similar tool in production, with tests and code review
- Has owned a warehouse bill, or killed a query that was costing real money
- [Years of experience, scaled honestly to the level, usually 3 to 5 for mid and 5 to 8 for senior]
Helpful, Not Required
- Moved a warehouse from one platform to another
- Written masking or row-level access policies
- A certification from the vendor you run, as a tiebreaker and nothing more
- [Time spent with your industry’s data, such as claims, loans, or orders]
Pay, Benefits, and Location
Annual base of $[low] to $[high], which is the range we set in good faith for this level. [Bonus target N%, plus equity if there is any.] Benefits include [health insurance, a retirement plan, and paid leave, one line each]. You’d work from [city], [remote / hybrid N days a week]. (Connecticut wants the benefits description as well as the range, and printing both in every posting is easier than tracking which ones need it.)
Pay Sources, and Connecticut’s Rule From October 1
The sources disagree. The gap is useful. The salary guide goes through each of them in detail, so this is the short version for setting the number in a posting.
| Source | What it measures | Figure | The catch |
|---|---|---|---|
| ZipRecruiter, Nov. 14, 2026 | Data Warehouse Engineer | $126,499 average, $113,500 to $136,500 middle range | Every experience level in one number |
| Glassdoor, captured Apr. 27, 2026 | Data Warehouse Engineer | $121,901 average, $93,820 to $159,779 middle range, 385 salaries | Self-reported, and I read an archived copy because the live page blocks scripts |
| Bureau of Labor Statistics, May 2025 | Database architects, nationwide | $139,500 median, $109,370 to $169,290 middle range | No line for the title itself, only its parent |
| Bureau of Labor Statistics, May 2025 | Database architects, Connecticut | $144,460 median, $133,850 to $164,280 middle range, about 1,140 employed | Small headcount, so one large employer moves it |
| Bureau of Labor Statistics, May 2025 | Database architects, Hartford metro area | $137,980 median, about 460 employed | Sits below the state figure |
Our bands run above those averages, which is the point. Averages count everybody and a band prices one person, and since these are the same bands the hiring guide and the salary guide use, nothing on this page contradicts either of them.
| Level | Base range to post | What the posting should show |
|---|---|---|
| Mid (3 to 5 years) | $130,000 to $160,000 | Builds and maintains models inside a design someone else set |
| Senior (5 to 8 years) | $160,000 to $200,000 | Owns the model end to end and sets the cost guardrails |
| Lead or staff (8+ years) | $200,000 to $260,000 | Platform strategy across vendors, and the executives who fund it |
Principal seats are a separate posting, because total compensation at the biggest product companies runs $280,000 to $420,000. All of the figures above are base salary for the United States before any metro adjustment, and if you need one city, the salary benchmark assistant will get you there faster than any table.

Now Connecticut. Its date passed about six weeks ago, and the rule is easy to miss. Public Act 26-12 started life as H.B. 5003 and rewrote the state’s pay statute, section 31-40z, effective October 1, 2026. The text of the Act runs 124 pages. The part that matters hides in Section 2, and six points from it apply here.
- Postings. An employer can’t fail to disclose, in an internal or public job advertisement, “the wages or wage range for such position and a general description of the benefits to be offered with such position.”
- Size doesn’t matter. The Act’s definition of employer reaches anyone “using the services of one or more employees for pay.”
- The range has to be one the employer “sets in good faith for a position.” The Act says it may refer to a pay scale, a previously determined range, the actual range for people in comparable jobs, or the budgeted amount.
- The word “benefits” stretches further than people expect. It names health insurance, retirement, fringe benefits, and paid leave, then adds “any other compensation other than wages.” I’d put a bonus target in there. Equity is a question for counsel, not me.
- Remote roles count when the employee has to “report directly to a supervisor, office or other worksite located within the state.” A warehouse engineer working from Texas for a manager in Hartford is covered.
- An applicant or employee can sue within two years, and an employer found in violation may owe compensatory damages, attorney’s fees, and costs, plus any other relief the court finds just.
Nothing in the text carves out an ad a recruiter writes. The duty sits on the employer, so I’d treat those ads as covered, which means the range has to exist before the first draft of the posting does. Settle the number first. A range scribbled in a hurry after the first candidate names a figure isn’t one the employer “sets in good faith.”
Questions We Get on Warehouse Postings
Data warehouse engineer or data engineer, which goes in the title?
Data warehouse engineer, if the seat owns the model and the bill. Data engineer is the bigger search term, and it draws a bigger pile of people who build pipelines.
If most of the week is loading and scheduling, use the pipeline title. Be honest about it. A wrong title costs you a month of interviews with people who are excellent at a different job.
Do we have to say Kimball, or is “dimensional modeling” enough?
Say Kimball if you run Kimball. Candidates read method names as promises, and a posting that names none sounds like a team that hasn’t decided.
Mixes are fine. “Stars in the finance marts, wide tables everywhere else” is a sentence a finalist can plan around.
We run Snowflake with a Databricks pilot. Can we list both?
List the platform the seat runs on day one and call the other a pilot.
“Snowflake or Databricks” in the requirements reads as two jobs under one title, which is the point our hiring guide makes about picking a lean before you post. A pilot is a lean you haven’t finished. Say so instead of hedging.
How many years should a senior warehouse seat ask for?
Five to eight years, and name what the person will have owned in that time. A number alone filters nobody. Someone with six years of ticket work and someone with six years of design work both clear it.
We’re moving our own analytics engineer into the seat. Does Connecticut still want a range?
Yes, an internal job advertisement has to carry the wage range and a general description of benefits, and an employee whose position changes is owed the range for the new one.
That’s two separate obligations. Put the range in the internal ad and again in the offer letter, and don’t wait for them to ask.
Does a SQL Server 2016 warehouse change who we should hire?
A warehouse still on SQL Server 2016 is past Microsoft’s end of support, so the first year includes a move, and a move is often better staffed as contract work.
The migration lead who has retired an estate like that is a different person from the engineer who runs the new platform afterward, and one posting written for both is how a company ends up paying a migration lead’s rate for a maintenance seat. Different skills, different pay. Some companies hire both, one on contract and one for the long haul, and the posting for each should say which job it is.
One Last Pass Before It Goes Live
Give the draft to the analyst who files the most tickets against the warehouse and ask one question. Can you tell, from this page alone, what one row in the biggest table means? That’s the whole test. If the answer is no, a finalist who builds warehouses for a living can’t tell either. Once the posting is out, our data warehouse engineer interview questions put the same grain question to a finalist on paper.
Since 2005, KORE1 has placed technology talent across 30+ U.S. metros, with recruiters who average more than 15 years in the work. Of the people we place, 92% are still in the seat a year later, and our technology searches average 17 days to fill. A permanent owner of the model is usually a direct-hire search. If you’d like a second read, send us the draft and we’ll mark the line a finalist would push on first.

