Compensation Benchmarking: How to Set Pay That Attracts and Retains Talent

Compensation benchmarking is the process of comparing what you pay your employees against what the market pays for equivalent roles. Without it, you are either overpaying (eroding margins without improving retention) or underpaying (losing people to competitors who know what the market is paying). In most organizations, it’s some of both — and they don’t know which roles are in which category.

This guide covers how to run a benchmarking process, which data sources to trust, how to build usable pay bands from benchmark data, and how to handle the inevitably uncomfortable gap between where you are and where the market is.

Why Compensation Benchmarking Matters More Now

Two developments have made compensation benchmarking more urgent and more complex than it was five years ago:

Pay transparency legislation. Colorado (2021), New York City (2022), California (2023), and a growing list of states and cities now require employers to include salary ranges in job postings. Once you post a range, you are publicly committing to a market position. Without benchmarking, that range is a guess.

Candidate access to compensation data. Candidates now come to interviews having researched compensation on Levels.fyi, Glassdoor, LinkedIn Salary, Blind, and Payscale. The information asymmetry that once allowed employers to negotiate from a position of salary ignorance is largely gone for professional roles. Candidates who believe an offer is below market reject it — or accept it and start searching again immediately.

Step 1: Define the Job, Not the Person

The most common benchmarking mistake is benchmarking the person in the role rather than the role itself. You benchmark jobs — the scope, responsibilities, and skills required — and then separately evaluate whether the person in that job is at, above, or below the job benchmark.

Job matching: matching your internal job title and responsibilities to the standard job families and levels used by compensation survey providers. A “Senior Software Engineer” at your company may be a “Software Engineer III” in Radford’s data — or a “Software Engineer II.” The match matters because a one-level mismatch in engineering compensation surveys can be a 15-25% difference in market rate.

Level definitions: standardized level definitions (usually based on scope of work, independence, and expertise) help you consistently match jobs to survey data across departments that use different level nomenclature. A “Senior” label in engineering often means something different than “Senior” in operations.

Step 2: Choose Your Data Sources

No single compensation survey covers all roles with equal quality. The major sources:

Radford Global Compensation Database (Aon): The gold standard for technology, life sciences, and startup compensation. Particularly strong for software engineering, data science, product management, and sales roles at venture-backed and public tech companies. Requires survey participation to access full data; a significant annual cost for smaller companies.

Mercer Total Remuneration Survey: Broad cross-industry coverage. Strong for HR, finance, legal, marketing, and operations roles. Often used when Radford doesn’t have sufficient data for non-tech functions.

WillisTowersWatson (WTW) Compensation Surveys: Commonly used in financial services, healthcare, and manufacturing. Strong for executive and manager-level benchmarking.

Levels.fyi: Real-time, self-reported compensation data from technology professionals. More current than annual survey data (which can be 6-12 months stale by the time of publication) and highly specific for individual companies and levels. Not statistically rigorous but useful as a market reality check for engineering and product roles.

LinkedIn Salary: Broad, accessible market data for many roles. Less granular than Radford or Mercer; useful as a directional check, not as primary benchmarking data.

Glassdoor and Payscale: Self-reported, potentially biased, and less reliable for precise benchmarking. Useful as a candidate perspective check — this is what your candidates are seeing before they interview with you.

Geographic differentials: Market rates vary significantly by location. A software engineering role in San Francisco commands 40-60% higher total compensation than the same role in Austin or Atlanta in current market data. Benchmarking must account for geography at the metro level, not just nationally.

Step 3: Pull and Analyze the Data

Once you have access to your data sources, the benchmarking process:

Match jobs to survey data. For each job you’re benchmarking, identify the closest match in each survey. Record the survey source, the matched job family and level, and the market data percentiles (P25, P50, P75, P90) for base salary and total cash compensation.

Gather multiple data points per role. For important roles, use 2-3 survey sources and compare. If the data is consistent across sources, you have high confidence. If there’s significant variance, investigate whether the job matching differs or the data sources are capturing different market segments.

Update for aging. Annual survey data ages. A Radford survey published in Q1 2026 with data collected in Q3-Q4 2025 is already 6-15 months old by the time you use it. Apply an aging factor (typically 3-6% annually for technology roles; 2-4% for other functions) to bring data to the current date.

Choose your market positioning target. Where does your organization want to sit? P50 (market midpoint), P75 (competitive lead), or somewhere else? This is a business decision, not a data decision — but it must be explicit and documented.

Step 4: Build Pay Ranges

A pay range (or pay band) defines the minimum, midpoint, and maximum salary for a job level. The range should be wide enough to accommodate employees at different tenure and performance levels within the level, without being so wide that it provides no guidance.

Standard range width: typically 50-80% range spread (the maximum is 150-180% of the minimum). Wider ranges allow more compensation differentiation within a level; narrower ranges create more compression between levels.

Setting the midpoint: the midpoint of the pay range should align with your market positioning target. If you target P50, the range midpoint equals the P50 market rate.

Minimum and maximum: 75-80% of the midpoint for the minimum; 120-125% of the midpoint for the maximum is a common structure. An employee at the minimum is new to the level; an employee at the maximum has typically maximized what this level can support and is a candidate for promotion.

Overlap between ranges: ranges typically overlap 15-25% between adjacent levels. This allows an employee near the top of one level to earn more than an employee at the bottom of the next level — avoiding the “promotion pay cut” scenario where a promotion doesn’t immediately increase pay above the previous level’s maximum.

Step 5: Compare Current Pay to Market

With pay ranges built from market data, compare each employee’s current salary to the market midpoint for their role and level. The result:

Compa-ratio: (current salary ÷ range midpoint) × 100. A compa-ratio of 95 means an employee is paid 95% of the midpoint. 80-120 is a typical acceptable range; below 80 indicates potential flight risk; above 120 indicates potential overpay.

Red circles (above range maximum): Employees whose salary is above the range maximum. Common causes: legacy pay that wasn’t adjusted when ranges tightened; promotions that didn’t move to the appropriate level; individual deals made outside process. Strategy: freeze increases until the range catches up (through range increases) or re-level the role.

Green circles (below range minimum): Employees whose salary is below the range minimum — often called “salary compression” in the specific case where newer employees are hired at rates near or above longer-tenured employees in the same role. Compression is a significant flight and equity risk. Strategy: bring all employees to the range minimum as quickly as budget allows; prioritize this over other compensation actions.

Handling the Gaps

Benchmarking almost always reveals that some roles are underpaid relative to market. The harder part is deciding what to do about it within budget constraints.

Prioritize by risk. The highest priority for market adjustments are roles with active talent risk: employees who have flagged compensation concerns, roles with recent turnover, and roles where you’re losing candidates in the offer stage due to compensation. These have quantifiable cost — the replacement cost of a departing employee.

Set an adjustment budget separately from the merit budget. Market adjustment increases are driven by competitive necessity, not performance merit. Conflating them distorts both processes. Many organizations set aside 1-2% of payroll as a market adjustment budget separate from the annual merit pool.

Be transparent about the gap and the timeline. Telling an underpaid employee “we know, and here is the plan to address it over the next 12 months” is better than silence or deflection. Employees who understand the gap and see a credible plan are more likely to stay than those who discover the gap through a competitor’s offer.

Conduct a pay equity audit in parallel. Compensation benchmarking often reveals patterns that are also pay equity problems — where employees in the same role at the same performance level are paid differently based on gender, race, or other protected characteristics. Running the benchmarking analysis through a pay equity lens is both a legal risk management step and an ethical one.

Compensation Benchmarking Tools

For organizations with dedicated compensation teams: Mercer Comptryx, Radford Insight Engine, WTW Compensation Software — purpose-built analytics tools for survey data analysis and range building.

For mid-market HR teams: Payfactors (now Payscale), CompAnalyst (from Salary.com), Levels.fyi for tech roles. These provide more accessible interfaces at lower cost than enterprise survey platforms.

For ad hoc benchmarking: LinkedIn Salary, Glassdoor, and Levels.fyi are free and useful for quick reality checks on specific roles, even if they lack the statistical rigor needed for formal range-setting.

Frequently Asked Questions

How often should we benchmark compensation? Annual benchmarking is the minimum for most organizations. For high-demand roles in competitive markets (software engineering, data science, certain sales roles), semi-annual review helps ensure ranges stay current as market rates move. Triggered reviews make sense when you observe unusual turnover in specific roles or lose multiple candidates at the offer stage.

Should we share pay ranges with employees? Increasingly, the answer is yes — both because of legal requirements in many US states and because employees who understand the range and their position within it are less likely to perceive compensation as unfair or opaque. Organizations that share ranges need to be prepared to explain why employees are at their specific point within the range.

How do we handle geographic pay differences for remote employees? Three approaches: (1) Geo-based: pay based on employee location (common in larger companies with established geographic pay structures); (2) Role-based: pay the same range regardless of location (simpler administratively, more expensive in lower-cost markets); (3) Hybrid: pay at a national median for remote roles, regardless of employee location. Each approach has trade-offs in equity, cost, and administrative complexity.

What is the difference between base salary benchmarking and total compensation benchmarking? Base salary benchmarking compares base pay only. Total cash benchmarking includes base plus target variable (bonus/commission). Total compensation benchmarking includes equity and benefits value. The right benchmark depends on how you want to compete — if your equity package is strong, total compensation benchmarking shows that more clearly than base salary alone.