§ Decide

Should I hire a data engineer or use a data platform firm?

If you have people reading data and nobody building it, a data platform firm gets you a working platform sooner, and you own what it builds. Hire once the platform needs a full-time owner, or if you already employ a data engineer.

By Kendall Barber-Mayo · Updated

The short answer

The case we are answering is a company with readers and no builder. Analysts, finance and operations leads open reports every day, and nobody's job is to connect the sources, run the warehouse and test the numbers. For that company, a firm builds the platform first, documents it, and hands it over. You hire when the platform has enough work to keep one person busy full time.

Not sure what your stack needs before you post the job? The Data Platform Blueprint surveys every system you run and ends in a sized, phased plan, the plan a new hire or another firm would build from.

The gate: consumers versus builders

The question is not how big you are or whether the work is a project or ongoing. It is whether you have people who consume data and nobody who builds it.

Analysts, data scientists, IT managers, operations leads and finance people never rule you out. They are the reason a platform is worth building. A data engineer or analytics engineer on staff does rule you out, because you have already committed to building the platform layer in-house. If that is you, do not buy from us first: hire, or promote, and use the sections below to plan it.

Fits a firmDoes not fit a firm
11 to 200 peopleYou already employ a data engineer or analytics engineer
Analysts, data scientists, IT or ops leads read the numbersData is part of the product you sell
Nobody is paid to build pipelines and modelsNew sources arrive every few weeks, indefinitely
The numbers disagree between systems and spreadsheetsYou need an engineer available every day, or staff-only access

What a first data engineer does in year one

A first hire does five things, in roughly this order:

  1. Connects your sources (ingestion), including the ones with no API.
  2. Stands up and runs the warehouse.
  3. Builds and tests the models that turn raw tables into agreed metrics.
  4. Keeps the pipelines alive, which means being the person who gets paged when a load fails.
  5. Answers ad hoc requests from the people who read the numbers.

Items 1 to 3 are a build. Items 4 and 5 never end. A Wakefield Research survey of 300 data and analytics leaders for Fivetran, published in 2022, found the average data engineer spends 44% of their time maintaining pipelines, as those leaders report it (Fivetran, 2022). That survey was commissioned by a vendor that sells pipeline automation, so read it as a signal, not a measurement.

What the hire costs

The salary sources disagree because they measure different things: base pay versus total compensation, candidates placed by a staffing agency versus self-reported pay, and a sample that skews to large tech companies.

SourceMeasureFigure
PayScale, 2026Median base pay, 3,271 profiles$101k (10th to 90th percentile: $72k to $142k)
Levels.fyi, retrieved 2026-10-28Median total compensation, 1,175 submissions$157.5k (25th to 75th percentile: $117k to $226k); skews to large tech
Robert Half, 2027 guideProjected salaries for placed candidates, by experience$130,500 (new to the role) to $185,500 (extensive experience), with $160,500 for moderate experience; Robert Half's projections for 2027, retrieved 2026-10-28

There is no U.S. Bureau of Labor Statistics occupation called data engineer. The nearest are database administrators, with a median wage of $104,620, and database architects, at $139,500 (May 2025, BLS Occupational Outlook Handbook). Data scientists, a related role, had a median of $120,230 (BLS, 2025). These are neighbors, not the same job.

The $100k to $150k we quote is a base salary for a typical small-company hire. It is not a ceiling: the placement and total-compensation figures above run higher.

Salary is not the whole cost. For private-industry professional and related occupations, the closest BLS group to a data engineer, employer benefits were 30.8% of total compensation in June 2026: $22.71 of benefits on $50.98 of wages per hour, about 45% on top of wages (BLS, Employer Costs for Employee Compensation, Table 4). That is an occupation-group average, not a data-engineer figure, and the arithmetic is ours. Applied to a $100k to $150k base, it puts the fully loaded cost near $145k to $217k a year, before recruiting. Treat the top of that range as a ceiling: the group's average wage is about $106k a year, and some benefits, health insurance for one, are flat dollar amounts, so the load is probably a smaller percentage at higher salaries.

What it costs to find and keep them

The short version: a search takes weeks, the direct cost is small next to salary, and the person will probably leave within a few years. These are all-role figures, not data-engineer figures.

Time to fill: SHRM's 2026 benchmarking of more than 4,600 organizations puts the median at 39 calendar days for a nonexecutive role (SHRM, 2026). That covers all roles. A recruiting firm that places data engineers puts the typical internal search at 6 to 10 weeks (KORE1, 2026), though a recruiter has a commercial interest in how hard hiring looks.

Direct cost: SHRM's 2025 benchmarking (2,371 members, fielded January to March 2025) puts nonexecutive cost per hire at a median of $1,200 and an average of $5,475; executive roles average $35,879 (SHRM, 2025 press release; the median is from the SHRM, 2025 data brief). The gap suggests a few expensive hires pull the average up. A 2022 SHRM article quotes one consultant's estimate that the total cost of a hire runs three to four times salary (SHRM, 2022); treat that as a rule of thumb, not a measurement.

Turnover: median tenure with the current employer was 4.6 years in computer and mathematical occupations and 4.1 years for all workers in January 2026 (BLS, Employee Tenure). Plan for the person to leave. Replacement estimates are wide: Gallup puts it at one-half to two times annual salary (Gallup, 2019), while a 2012 review of case studies found a median of about one-fifth of salary, drawn mostly from jobs paying under $75k, with higher-paid roles costing more to replace (Center for American Progress, 2012). One hire is also one person holding all the context.

Side by side

Hire a data engineerUse a data platform firm
Time to first trusted reportA search (SHRM's median is 39 days for any nonexecutive role, and specialized engineers often take longer), then onboarding and ramp-up before the first build starts. No platform until the person is productive.The Blueprint comes first, and how long it takes depends on how many systems you run. The build then takes 14 to 21 days from scoping. That is our own measurement: one platform took 17 days from first commit to documented handover, from git history, and we quote 14 to 21 (What is Wide Open Tech?).
Year-one cost shapeSalary plus benefits (about 31% of total compensation, or 45% on top of wages, for professional and related occupations; BLS, June 2026), recruiting, equipment and tools, and the cost of the vacancy. Roughly $145k to $217k loaded for a $100k to $150k base. Recurring whether or not there is new work.Fixed scope, fixed fee, quoted after scoping (see the cost question below for why we do not publish a number). A one-time build, then an optional maintenance and feature retainer.
Bus factorOne person holds the context. Median tenure in computer and math jobs is about 4.6 years (BLS, January 2026), so plan for a handover at some point.The work is in your repo and documented. Handover is a deliverable. The protection is that repo and handover, not the firm's headcount, since a small firm also leans on a few people.
Who owns it afterYou, provided the code and data sit in accounts you own, which is yours to require of the hire and to check.You: the cloud account, the code and the data are yours.

When to hire instead

If any of these describes you, do not buy from us first.

  • New sources keep arriving every few weeks. When the work is continuous rather than a build followed by running it, a full-time person earns their seat.
  • Data is part of what you sell. If your roadmap includes machine learning or product features built on data, that is a different job from reporting, and it is outside what we pitch.
  • You are around 200 people or more. The workload usually keeps a specialist busy by then. This is our judgment, not a sourced threshold.
  • You already employ a data or analytics engineer. You are not our buyer. Hire the next one, or promote.
  • You need someone every day. Ad hoc engineering requests all week, or compliance that requires staff-only access, point to an employee.

The hybrid: the firm builds, your first hire inherits

A sequence that fits companies in this range: a firm builds and documents the platform, and you hire when it needs a full-time owner. That person reads the docs and the tests on day one instead of reverse-engineering someone's pipelines.

Whichever route you take, insist on the same things:

  • Code and data in accounts you own.
  • Open formats, so you can leave.
  • Tests that fail loudly when a number looks wrong.
  • A written handover.
  • A named owner on your side for each metric definition.

Software licences are small next to labor. dbt's Starter plan lists at $100 per developer seat per month, for up to five seats (dbt Labs, pricing), and Power BI Pro at $14 per user per month, billed yearly (Microsoft, pricing); both are list prices as of October 2026. The choice is mostly about who does the work, not which tools.

Questions

How much does a data engineer cost?

Sources disagree because they measure different things. Median base pay is about $101k (PayScale, 2026). Robert Half's 2027 projections for placed candidates run from $130,500 (new to the role) to $185,500 (extensive experience), with $160,500 for moderate experience. Levels.fyi's median total compensation is about $157.5k (retrieved 2026-10-28). Benefits are about 31% of total compensation for professional and related occupations, which is about 45% on top of wages (BLS, June 2026), so a $100k to $150k base lands near $145k to $217k fully loaded, before recruiting.

How long does it take to hire a data engineer?

SHRM's 2026 median time to fill a nonexecutive role is 39 calendar days across all roles. One recruiter puts a data engineer at 6 to 10 weeks through an internal search (KORE1, 2026, a source with a commercial interest). That is before the person starts, learns your systems and ships the first report.

Can an analyst I already have do this?

An analyst who reads and shapes data is a consumer, and fits our gate. If they are building the pipelines and the warehouse, they have become your data engineer, and the answer changes: you are in the hire path.

What happens if the person who built it leaves?

Median tenure is 4.6 years in computer and mathematical occupations and 4.1 years for all workers (BLS, January 2026), so it will happen at some point. The protection is the same for a hire or a firm: code in your accounts, tests, a written handover and a named owner for metric definitions.

When should I hire instead of using a firm?

When new sources arrive continuously, when data is part of what you sell or your roadmap includes machine learning, when you are around 200 people or more and the work keeps a specialist busy, or when you already employ a data or analytics engineer.

What does the firm route cost?

Fixed scope, fixed fee, quoted after scoping. We do not publish a number because the scope sets it: how many systems feed the platform, how many of them have no API, and how many metrics must match today's numbers. The Blueprint measures those first. The Data Platform Blueprint is the first step: it sizes and phases the platform your stack needs, so you can compare against a hire with real scope in hand.

Find out what your stack needs before you post the job

The Data Platform Blueprint is a small, fixed-scope engagement. The map, the question matrix and the plan are yours whether we build, your own hire builds, or it waits. Kendall has built data platforms for six companies. Pulling numbers from POS systems into Excel took a week or more; now a report takes a few minutes. What is Wide Open Tech? has more on how we work.

Sources

Next step · Data Platform Blueprint

Map every system you run.

Tell us what the business runs on and we will reply within 24 hours to set up the survey.