Salary Benchmarking: How to Run the Process Step by Step
Salary benchmarking end to end: where compensation survey data comes from, how to match a job to a benchmark, how to age it, and how often to rerun it.
Salary Benchmarking
The process rather than the artifact: where compensation survey data actually comes from, how to match one of your jobs to a survey benchmark without being fooled by the title, how to age a figure to the date you will pay it, how to blend disagreeing sources into one number, and how often the whole thing has to be rerun before it stops being true
The first salary I ever set for a role I had not done myself was a number I picked because it felt defensible. It was not benchmarked against anything. I then spent two years defending it to the person I had hired, to the next person I hired into the same role, and eventually to myself.
Salary benchmarking is what replaces that. It is not a subscription and it is not a dashboard. It is a sequence: pull market data from more than one place, match your actual job to the right benchmark, adjust every figure for how old it is and where it came from, blend them into one number, and convert that number into a range you can pay without flinching.
What follows is the whole process, including the parts that go wrong. I build the people and records tooling for businesses without an HR department at FirstHR, and FirstHR is an onboarding and HR platform rather than a payroll provider or a compensation data vendor. This is general information rather than legal or compensation advice.
What Salary Benchmarking Is
Salary benchmarking is the repeatable process of establishing what the external market pays for a job, so that your own pay decisions have an evidence base behind them. It produces a market rate; it does not by itself decide what you pay.
That distinction does most of the work in this article. A salary band is an artifact. So is a compensation range. Artifacts go stale quietly, without any signal that they have, because a spreadsheet does not tell you when it stopped being true.
The process is what keeps them true. Small businesses tend to build a set of ranges once, feel good about it, and then never rerun the exercise that created them. Three years later the ranges are still there, still being quoted to candidates, and no longer describing any market that exists.
Where the Data Comes From
Four categories of source are realistically available to a small employer, and each is wrong in a different direction. Using two of them is the minimum that lets you notice when one is wrong.
The federal option is the workhorse. The Occupational Employment and Wage Statistics program publishes mean and percentile wages by occupation and area, free, from a survey that combines six semiannual panels covering roughly 1.1 million establishments over three years. According to the Bureau of Labor Statistics Occupational Employment and Wage Statistics survey (May 2025), employment across all occupations nationally was 155,495,730 and the median hourly wage was $24.51.
One quirk worth knowing before you build a senior range on it. OEWS top codes its high-wage estimates: where a wage reaches or exceeds $115.00 per hour, or $239,200 per year, the published table shows a marker rather than a value. For executive roles that is exactly the number you wanted, so senior benchmarking needs a source that does not stop there.
Posted ranges deserve more respect than they get. In jurisdictions covered by pay transparency laws, every employer hiring your role in your metro is publishing a band, dated to this week. Collecting a dozen of those postings is an afternoon of work and produces a picture no annual compensation survey can match on currency.
Which Jobs to Benchmark
Benchmark the jobs that market data can actually price, then slot everything else against them internally. Trying to find external data for every distinct role in a small company is how benchmarking projects die.
A benchmark job is one that exists in broadly the same form at many employers: bookkeeper, customer support representative, warehouse supervisor, account executive. These appear in survey data with enough sample behind them to mean something. The job classification work you have already done, if you have done any, tells you which of your roles qualify.
Everything else is a slotted job. You price it by internal comparison: this role sits between the two benchmark jobs either side of it in scope, so its midpoint sits between theirs. That is not a compromise, it is standard practice, and it is more honest than forcing a hybrid role onto a survey benchmark that does not describe it.
Matching a Job to a Benchmark
Match on the job description and the level definition, never on the title. Titles are the least standardized element in all of compensation data, and matching on them is the most common way a benchmarking exercise produces a confidently wrong number.
Before you open a single survey, write down what the role actually does: the main duties and roughly what share of the time each takes, who it reports to, what it owns, what it can decide without asking, and what budget or revenue sits behind it. That document is what you match against.
| Check | What you are comparing | When the match fails |
|---|---|---|
| Duties | The survey job description against your written duty list | The overlap is partial and the missing parts are the expensive parts |
| Level | The survey level definition, not your internal seniority label | Your senior is their mid, or the survey has no level breakout at all |
| Reporting line | Who the role reports to and how many layers sit above it | Your role reports to the founder, the benchmark reports to a department head |
| Decision rights | What the role can commit to or approve unsupervised | The benchmark assumes an approval layer your role does not have |
| Scale | Budget, revenue, headcount supervised, transaction volume | The benchmark describes the same work at ten times the scale |
| Specialization | Generalist breadth versus specialist depth | Your role does three jobs and the benchmark describes one of them |
When the match fails, you have three honest options. Find a different benchmark. Match to two benchmarks and interpolate between them, documenting the weighting. Or accept that the role is a hybrid the market does not price, and slot it internally instead.
What you should not do is pick the closest available title and move on. A bad match does not produce a slightly wrong number; it produces a number from a different job, which is a different kind of error entirely.
The Scope Versus Title Trap
The scope trap is the specific failure where two roles share a title and share almost none of the actual work. It runs in both directions, and small businesses are exposed to both.
Upward, a small company gives a broad title to a role with modest scope, because titles are free and salary is not. Your office manager may hold a title that in survey data describes somebody running a facilities function. Benchmarking against that title produces a range you cannot afford and did not need.
Downward, and more damagingly, a single person at a small company does three jobs under one title. The bookkeeper who also runs payroll, manages the vendor relationships, and closes the month is not matched by a bookkeeper benchmark. Price her against it and you will lose her, then rehire the function as two people at a combined cost you never modeled.
This trap is also where wage compression starts. Benchmark the broad title, pay to that benchmark, and the person doing three jobs ends up at the same midpoint as the person doing one.
Effective Dates and Aging
Every compensation survey figure describes a moment in the past, and the number you need describes a moment in the future. Aging the data is the step that connects them, and skipping it is the most common technical error in small business benchmarking.
Two dates matter and they are not the same. The publication date is when the file appeared. The effective date, or wage reference period, is when the money in that file was actually being paid. Federal wage data illustrates the gap plainly: the May 2025 OEWS estimates were published on May 15, 2026, so the reference period was already a year old on the day the file went live.
The Bureau itself does this internally. Because OEWS combines six panels collected over three years, wages from the earlier panels are adjusted forward to the current reference period using aging factors derived from the Employment Cost Index. That is the same mechanism you are applying, run one step further.
Use a published wage movement rate rather than a guess. The Employment Cost Index reported wages and salaries for private industry workers up 3.1 percent over the twelve months ended June 2026, with benefit costs up 3.8 percent. Occupation-specific movement varies around that, sometimes sharply, which is a reason to check a hot job family more often rather than a reason to invent your own rate.
Blending Disagreeing Sources
When two sources disagree, the disagreement is information before it is a problem. Resolve it by working out why they differ, then weight them; do not simply average the two and hope.
Multiple sources are the norm among employers who do this seriously. A market pricing practices survey published by SHRM in 2011 found that over eighty five percent of companies that use compensation surveys use multiple surveys, and over seventy percent avoid surveys built on data collected from individual employees. The reason to avoid self-reported data holds regardless of the vintage of that finding: with employee-submitted figures, neither the job match nor the salary number can be verified.
| Source | Weight it heavily when | Discount it when |
|---|---|---|
| Federal wage data | The occupation code cleanly describes the role and your metro has a real sample | The role is senior, hybrid, or newer than the occupation taxonomy |
| Association compensation survey | Your industry and revenue band are broken out separately | The cut you need has a handful of respondents behind it |
| Posted ranges in your market | You have collected enough postings for the same role and level | The postings are wide, stale, or from employers of a different scale |
| Paid compensation survey | Level definitions exist and your job matched one of them properly | You matched the title because no description was close |
| Anything self-reported by employees | Never as a primary input | Always; treat it as a sanity check at most |
A practical rule: weight by match quality first and sample proximity second. A mediocre sample of a job that genuinely resembles yours beats a large sample of a job that does not. Where two aged figures land within a few percent of each other, take the weighted mean and move on. Where they land thirty percent apart, stop and recheck the match, because that gap is almost never the market.
Geography, Industry, and Size
Three cuts change the number materially, and you should decide deliberately which one you are pricing against rather than accepting whichever cut the source happens to default to.
Geography is the largest single mover for most roles. Price against the labor market you actually recruit from, which for an in-person role is the metro and for a remote role is whatever geographic policy you have written down. If you have not written one down, that is the real gap, and no amount of data will paper over it.
Industry matters most where the same job carries different risk or a different pull on revenue. A controller in construction and a controller in professional services are doing recognizably similar work at recognizably different pay. Where an association survey breaks your sub-sector out, that cut is usually worth more than a bigger national sample.
Organization size is the cut small employers most often skip and most often need. Survey medians are pulled upward by large employers with deeper pay structures. Where a size cut exists, use it, and where it does not, expect to sit below the headline median for reasons that are structural rather than stingy.
Turning Output Into a Payable Range
The blended, aged benchmark becomes your midpoint, and the range is built around it rather than from it. This is the step where benchmarking stops being an analysis and starts being something you can quote to a candidate.
Where you place yourself against the market is a policy choice, and it belongs in a written compensation philosophy. Benchmarking without one produces a market number and no way to decide what to do with it.
Once ranges exist, compa-ratio is the metric to manage against day to day, because it normalizes across every role. And remember that the benchmark is a wage figure rather than a total compensation figure, so budget on the loaded number.
How Often to Rerun It
Refresh the whole structure once a year, and spot check individual roles every time one goes to market. Annual is the right interval because a year of wage movement is large enough to matter and small enough that reworking quarterly generates churn rather than insight.
Timing matters as much as frequency. Run the refresh in the quarter before your compensation budget is set, so the data informs the budget. Run it after the budget is locked and you have created an exercise whose conclusion is already known, which is a slower way of not benchmarking at all.
The annual pass also feeds your merit increase cycle. Market movement and individual performance are separate reasons to change somebody's pay, and conflating them is how a company ends up giving a market adjustment and calling it a reward.
Documenting the Method
Record the method alongside the numbers, because a range you cannot explain is a range you will abandon under pressure. The documentation takes minutes and is what turns a spreadsheet into a defensible pay practice.
Six fields per benchmark job are enough: the sources used, the vintage or reference period of each, the geographic and industry cut applied, the aging factor and the date you aged to, the weighting between sources, and one line on why the job match holds. Anybody picking the file up a year later can then rerun it.
There is a legal dimension too. The federal equal pay provision permits a pay differential made under a seniority system, a merit system, a system measuring earnings by quantity or quality of production, or a differential based on any other factor other than sex (29 U.S.C. 206(d)). A documented benchmarking method is how a market-based differential gets evidenced rather than asserted.
That documentation is also the input to any pay equity review you run later. Without recorded midpoints and match rationales, an equity analysis has nothing to test individual salaries against.
Common Mistakes
Five failures account for most bad benchmarking at small companies, and the first is the one that produces the biggest error.
Matching on title is first. It feels like the fast path and it silently substitutes a different job for yours. Every other step is then executed perfectly on the wrong input.
Skipping the aging step is second. A figure from a reference period two years back, used unadjusted, is not conservative. It is simply wrong by whatever wage movement has happened since, and it will read as a lowball to every candidate who sees it.
Using one source is third. One source gives you a number and no way to detect that it is wrong. Two sources that agree give you confidence, and two that disagree give you a lead worth following.
Benchmarking base pay while comparing against total packages is fourth. If the survey reports base salary and your competitor is winning on base pay plus variable pay, you are answering a question nobody asked.
And treating it as a project rather than a cycle is last. The output has a shelf life measured in months. A brilliant benchmarking exercise done once and never repeated is a slightly better version of the number I made up when I hired my first employee.
Frequently Asked Questions
What is salary benchmarking?
Salary benchmarking is the process of pricing a job against external market data so that pay decisions rest on evidence rather than on instinct. You define what the job actually does, find comparable jobs in a compensation survey or other market source, adjust the resulting figures for date and location, blend them into a single midpoint, and turn that midpoint into a range. The output is a defensible number and a written trail explaining how you got there. It is a repeatable process rather than a one-off exercise, and the process is what makes the number credible when somebody challenges it.
What is a compensation survey and do small businesses need one?
A compensation survey collects pay data from participating employers against defined benchmark job descriptions, then reports it back by level, industry, geography, and organization size. Paid surveys give the cleanest job matching available and are the only source that reliably separates levels inside a single title. A small business rarely needs one for a first cycle. Free federal wage data from the Bureau of Labor Statistics, a trade association survey, and posted pay ranges in your own market will get you to a defensible midpoint for most roles. Buy a paid survey when a specific role is expensive enough that being wrong by ten percent costs more than the subscription.
How do you match a job to a survey benchmark?
Match on the job description and level definition, never on the title. Titles are the least standardized thing in compensation data: the same words describe wildly different scope at different companies, and the same scope carries different words. Read the benchmark description in full, compare it against a written summary of what your role actually does, and check reporting relationship, budget responsibility, and decision rights. If the duties overlap substantially the match holds. If they do not, either find a different benchmark, match to two benchmarks and interpolate, or accept that the role is a hybrid that market data will not price cleanly.
How current is BLS wage data?
Less current than the file name suggests, which is the single most useful thing to know about it. Occupational Employment and Wage Statistics estimates combine six semiannual survey panels collected over three years, roughly 1.1 million establishments in total, and the May 2025 estimates were published on May 15, 2026. The Bureau adjusts wages from earlier panels forward to the reference period using aging factors derived from the Employment Cost Index, so the published figures are internally consistent. They still describe a reference period that is in the past by the time you read them, and you have to age them forward yourself.
How often should you redo salary benchmarking?
A full refresh once a year, plus a spot check on any role before it goes to market. Annual is the right cadence for the structure because wage movement across a year is large enough to matter and small enough that quarterly reworking produces churn without insight. For scale, the Bureau of Labor Statistics Employment Cost Index put wages and salaries for private industry workers up 3.1 percent over the twelve months ended June 2026. Time the refresh for the quarter before your budget is set, so the data informs the budget rather than the budget deciding what the data is allowed to say. Individual job families sometimes need an off-cycle pass, usually signaled by declined offers at your maximum or by resignations that cite pay directly.
Should you use one source or several?
Several. A market pricing practices survey published by SHRM in 2011 found that over eighty five percent of companies that use compensation surveys use multiple surveys, and over seventy percent avoid surveys built on data collected from individual employees. That second habit still makes sense today, because with employee-submitted figures neither the job match nor the salary number can be verified. A single source gives you a number with no way to judge whether it is wrong. Two or three sources that broadly agree give you confidence, and two sources that disagree sharply tell you something useful: usually that your job match is wrong, or that the geographic cut on one of them does not describe your labor market.
Does benchmarking mean you have to pay the market median?
No. Benchmarking tells you where the market sits; where you choose to sit relative to the market is a separate decision that belongs in your compensation philosophy. Some employers deliberately target the twenty fifth percentile on base pay and make it up in equity, flexibility, or benefits. Others target the seventy fifth percentile for a handful of roles they cannot afford to lose and the median everywhere else. Federal wage data supports either choice, because Occupational Employment and Wage Statistics publishes the 10th, 25th, 50th, 75th, and 90th percentile wage for each occupation alongside the mean, so you can price to a position rather than only to the middle. What benchmarking removes is the ability to be below market by accident and to discover it only when somebody resigns.
What is the difference between benchmarking and a salary band?
Benchmarking is the process; a salary band is one of the things the process produces. Running a benchmarking cycle gives you a market midpoint for a job. Wrapping a spread around that midpoint produces a minimum and a maximum, which is a band or range. Comparing an individual salary against the midpoint produces a compa-ratio. The distinction matters because the artifacts go stale silently while the process is what keeps them true, and small businesses tend to build the band once and then never rerun the process that created it. A band nobody has rebuilt in three years is not a pay structure, it is a record of what one market looked like on one date.