How to Conduct a Compensation and Benefits Survey: An Operational Checklist

A compensation and benefits survey can be one of the most useful tools in your or your group’s compensation strategy, or it can be a spectacular waste of time and money. The difference comes down to how you run the process. Not just whether you have data, but whether you actually know what to do with it. This guide walks you through an operational checklist for conducting a compensation survey that moves decisions forward.

To be clear, part of this guide is to help you run the survey and the other part is to help you or your company, group, or constituency use the data.

Why some salary surveys miss the mark

Traditional compensation and benefits surveys do not match the speed of your constituency or group. They use a give-to-get model. Companies submit employee pay data maybe every other year. This process is time-consuming and often leans heavily toward big traditional companies in industries like manufacturing, oil and gas, pharmaceuticals, and finance.

By the time global survey providers collect and process data, the results often reflect pay levels from months ago, if not longer. That is not a real-time market snapshot. It is more like a Polaroid from last summer. That said, these surveys can still be valuable. The key is to understand their strengths while navigating their limitations. The goal is not to avoid them. It is to know how to run one effectively and fill in the gaps. But first, you need to grasp what you are actually getting.

What purchased compensation data provides

Scope varies by provider, but a typical purchased compensation dataset includes:

• Base salary range across percentiles.
• Variable pay like bonuses and commissions.
• Total compensation covering benefits, equity, and bonuses.

You get all of that broken down by job role, job level, industry, region, and sometimes company size or revenue range.

Big players like Mercer maintain their own job catalogues and level frameworks, so buying compensation data means mapping your (or your participants’) roles into theirs. That mapping step alone trips up more programs than you would expect.

The data typically arrives as spreadsheets, though more consultancies now offer software platforms for viewing and slicing it. Dedicated wage and benefits survey software makes it easier to segment purchased data by region, role, and level without drowning in raw exports, and to line it up against a first-party cohort you field yourself.

But raw numbers are only as useful as the context around them.

Step 1: Define your objectives before touching any data

Why run this survey? It is obvious, right? But most teams skip this step. Are you or your customers trying to benchmark salaries against the market? Set pay for new roles? Conduct an annual compensation review? Or plan a location strategy for remote teams? Or just figure out how much to pay local forklift drivers? Each goal dictates the data you need, where to get it, and how detailed to be. Without clear objectives, you will drown in data and come up empty.

Start by writing down three specific questions you want answered. Skip vague goals like “understand market.” Think real questions. “Are we paying our senior engineers at the 50th percentile for our region and company size?” That is a question you can answer with data. If you cannot articulate your question, you are not ready for the survey.

Step 2: Select the right data sources

So what does this actually look like in practice?

Not every compensation survey provider is going to work for your company. Here is what to think through before you pick one:

Most traditional providers update once or twice a year. For fast-moving companies, especially in tech, that might not be enough. You may want to layer in some real-time benchmarking tools that pull fresher data from a wider pool of companies.

That said, do not write off the traditional wage surveys. They still carry the broadest, most validated datasets out there, particularly for established industries and larger organizations.

The best approach is really just to use multiple sources. Layer them. No single provider is going to give you the full picture. Our guide to the four types of compensation survey covers which kind fits which question.

Step 3: Recruit participants and align on the job catalogue

A first-party survey is only as good as the cohort you get to sit inside it. The recruitment work is not logistics; it is where the study is either representative or not. Publish a participant list, or a cohort description if names are suppressed, and be clear about who you have and who is missing.

The single decision that shapes every downstream number is the job catalogue. Standardized frameworks (SOC codes, or an association’s own scheme) are the anchor; a “Senior Product Manager” at a 50-person shop is not the same role as one at a Fortune 500. Give participants a short mapping guide that asks them to match each of their internal titles to a catalogue role by:

• Scope of responsibility, not just the title
• Seniority level in the catalogue’s framework
• Function and specialization

If the mapping is off across enough participants, every number that follows is garbage. Get hiring managers and department leads involved on each participant’s side. Job descriptions and what people actually do are often two different stories.

Step 4: Collect and validate participant submissions

Collection is where a survey goes quiet for weeks and then produces the shape of the final dataset. Two things matter: making it easy enough that participants finish, and validating what comes in before it lands in the aggregate.

Give each participant a clear submission window, a template they can pre-fill in-house, and a single point of contact for questions. What they are sending is typically:

• Current base salaries by role and level
• Variable pay structures (bonuses, commissions)
• Equity grants and vesting schedules, where relevant
• Benefits packages

Then validate. Scan for hourly values entered as annual, decimal-place errors, headcounts that jump between cycles, and any role where one participant’s number would drag the median of a small cell on its own. Kick back the outliers to the participant with a specific question, not a generic “please review.” Sloppy inputs slow the whole cohort and skew every figure they touch.

If you are collecting benefits alongside wages, our benefits survey questions page covers the categories worth capturing and how to word them so answers stay comparable across employers.

Step 5: Aggregate and read the responses

Once submissions are in and validated, the responses aggregate into percentile tables by occupation and level. The reading discipline is the same one every convener has to enforce on themselves and communicate to participants:

Read the response count first. A percentile computed from four submissions is noise wearing a statistic’s clothes. Watch the 25th-to-75th spread: a tight band means the role is well-defined across the cohort; a wide band on a small sample usually means the title covers materially different work at different employers, and that is a job-matching finding, not a pay finding.

For a fuller walkthrough of how to read percentiles honestly, see reading your survey results.

Step 6: Apply Safe Harbor suppression to the aggregated data

Before any figure goes out, Safe Harbor rules decide what gets shown and what stays hidden. Two of the rules do most of the work.

• Suppression threshold. Any cell with fewer distinct participants than the threshold is hidden. Set it to fit the cohort: five is a common floor, higher for smaller groups where a lower number would let a reader reverse-engineer who submitted what.
• Three-month lag. Data is at least three months old before it is aggregated and published. That is not a bug in the process; it is the condition that lets participants submit real numbers.

Suppression is not something the convener has to plan every filter against. It is applied to the aggregated data automatically, so any cut a reader takes protects the underlying participants without the convener hand-picking which figures survive. Say all of this on the cover of the report. Participants are more willing to answer honestly the next cycle if they see the protection working.

Step 7: Prepare deliverables for participants

A single aggregate report is the raw material, not the deliverable. Package the results into what each audience actually needs:

• A full cohort report for the sponsoring body: methodology, cohort composition, percentiles by role, year-over-year movement, and suppression notes.
• A per-participant view for each employer that submitted: their position against the cohort by role, with a note on which cells were built from thin samples and should be read as directional.
• A short executive summary that is safe to circulate outside the cohort (aggregate only, no participant identifiers), so participants can share the headline findings with their own boards.

Our salary benchmarking template gives the grid structure most participant views end up sitting inside.

Step 8: Factor location and cohort splits into the reporting

Regional and sector splits matter as much as the overall percentiles. A software engineer in San Francisco sits in a different market than one in Nashville, and a mid-cap manufacturer’s cohort inside a broader study is not the same as the study’s headline number. Report those splits as filter axes rather than appendices, and set expectations up front:

• Which regions and sectors carry enough responses to publish, and which are suppressed.
• Where the cohort is skewed (a single industry dominating, or one metro over-represented).
• How the cross-region and cross-sector figures should be read given that skew.

A study that says “the market” without defining which cohort is which teaches its readers a habit they will use against the next report.

Step 9: Publish the report and plan the next cycle

Publishing is not the end; it is where the next cycle’s participation is either earned or lost. What works consistently:

• Deliver participant reports on the promised date. Slipping erodes the willingness to submit again.
• Host a short walkthrough for the sponsoring body and a separate one for participants, with time for questions on methodology and suppression.
• Send each participant a short feedback pass: what was useful, what was missing, what took too long to submit. That feedback becomes the instrument changes for next cycle.
• Schedule the next cycle before the current one is fully closed out. A predictable annual cadence is what builds a trend line participants and their stakeholders can rely on.

Keep the instrument stable between cycles. Changing a question changes the trend line; if a revision is needed, note it in the next report so nobody reads a wording change as a market movement.

The limits of traditional surveys, and what to do about them

Salary surveys take forever: collecting data, dealing with questions, compiling everything, and so much more. Often, bigger companies mostly participate. They offer a snapshot of the past. In fast-moving sectors like tech, that delay might mean your “competitive” salary is actually outdated by the time you make the offer.

They are not worthless. Purchased compensation data (Mercer, Willis Towers Watson, PayScale, Payfactors, and the sector-specific catalogues) is the largest, most organized dataset available, and useful for big established employers and regions with clear roles in traditional industries. Nothing is real-time. Under Safe Harbor, any figure a survey releases has to be at least three months old before it is aggregated and published, which is not a bug in the process but the condition that lets employers submit real numbers in the first place. The strongest programs pair purchased data with a first-party survey the convener runs on their own cohort, so their members and stakeholders see numbers that describe them specifically, alongside the broader syndicated context.

Quick-reference checklist

Here is the step-by-step:

• Define clear objectives and specific questions the cohort should answer
• Select the instrument and platform that fit your cohort’s scope
• Recruit participants and align on the job catalogue
• Collect and validate participant submissions
• Aggregate and read the responses honestly
• Apply Safe Harbor suppression to the aggregated data
• Prepare deliverables for the sponsoring body and each participant
• Factor location and cohort splits into the reporting
• Publish the report and plan the next cycle

Where to go from here

Running a defensible compensation survey process is not glamorous, but it is what earns your cohort’s trust for the next cycle and the one after that. The payoff is regional wage transparency your members and stakeholders can act on, tangible value for the participants who submitted, and the participant retention across cycles that turns a one-off study into a trend line.

Start with Step 1. Define what the cohort needs to know, then work through the checklist. Resist the urge to jump straight to the data.

If you are looking at the sponsor side of the same process (instrument design, fielding, publishing), the companion piece is how to run a wage and benefits survey.

If you need a platform to run the study on, take a look at Sensible Surveys. The strongest programs pair purchased data with a first-party survey the convener runs on their own cohort, so what the sponsoring body sees describes them specifically, alongside the broader syndicated context.