Where benchmarking data comes from,
and what it misses.

The four sources compared

Published salary surveys

The traditional source, and still the most defensible for established roles. Participating employers submit data under a consistent methodology, which is what makes year-on-year comparison meaningful.

The limitation is time. A survey published in spring commonly reflects data gathered the previous autumn, so it describes a market six to twelve months old. In a stable market that matters little. In a market repricing quickly it produces offers that were competitive a year ago.

• Check the collection date rather than the publication date.
• Check who participates. A survey dominated by large employers describes large employers, whatever its title claims.
• Check the sample size for your specific role and market rather than for the survey as a whole.

Aggregator and self-reported data

Platforms that collect salaries from employees and candidates directly. Current, broad and cheap to reach, which makes them genuinely useful at volume.

Two biases run through them. The first is self-selection: people report when they have a reason to, and satisfaction with pay is rarely the reason. The second is composition: base salary is easy to report and bonus, equity and pension are frequently omitted, so the figures understate total compensation on exactly the roles where the variable half matters most.

They work well for roles with thousands of holders. They mislead on roles with dozens, where a handful of unrepresentative entries move the median a long way.

Recruiter and search firm data

The most current source available, because it records what offers actually closed at rather than what employers reported afterwards. It also captures the number candidates declined, which sits beyond the reach of any survey.

Its bias is the desk it came from. A firm that recruits mostly for venture-backed technology companies has data describing those companies, and applying it to a family-owned manufacturer produces a confident answer to somebody else’s question.

This applies to our own data as much as anyone else’s. Marquee sees what closes across the mandates we run, which is a real sample and a partial one. Where our figures and a broad survey disagree, the honest position is that we are more current and the survey is more representative.

Practitioners currently in the role

The smallest sample and the only one that explains the number rather than merely reporting it.

A survey tells you the upper quartile for a head of manufacturing operations. A person holding that job tells you the top of the range goes to people who have commissioned a new line, that the bonus attached to it pays out about half the time, and that two of the four companies you would benchmark against have frozen hiring. None of that appears in a dataset.

• Most valuable on senior and scarce roles, where every other source has a sample too small to trust.
• Most valuable when a role is new to the organisation and lacks any internal reference point.
• Gathered one conversation at a time, which is what makes it expensive and what makes it worth paying for.

This is the source the Nextyn network exists to provide, and it is why Marquee reads survey data against practitioner evidence rather than in place of it.

Choosing the comparator set

Before any source is read, one decision shapes the answer more than the data does: which companies you are benchmarking against. Most disputes about a benchmark are really disputes about this.

1. Who do you lose candidates to? The most honest comparator set is the list of employers that have taken people from you, or beaten you to a hire. It is frequently different from the list of competitors in your market.
2. Who do you hire from? The organisations your last ten appointments came from describe the market you actually recruit in, whatever the industry classification says.
3. Match on size and stage, then on sector. A five-hundred-person business competes for people with other five-hundred-person businesses across sectors, more than it does with the listed leader in its own.

Write the comparator set down and agree it before the numbers arrive. Choosing it afterwards, once the figures are visible, is how a benchmark becomes an argument for a conclusion somebody had already reached.

Reading the sources together

A defensible benchmark triangulates rather than picking a winner. Four habits do most of the work.

1. Weight by role scarcity. Common roles: lean on surveys and aggregators. Scarce roles: lean on live market evidence, because the datasets have too few holders to be stable.
2. Compare the spread rather than the midpoint. Where two sources agree on the median and disagree on the upper quartile, the disagreement is the finding.
3. Convert everything to total compensation. Sources report different things. Comparing a base-only figure with a total-package figure produces an error larger than any real market movement.
4. Record what you used. A benchmark whose sources are written down survives a challenge. One presented as a single number invites the question of where it came from at exactly the wrong moment.

How often the data needs refreshing

Benchmark data ages at different speeds depending on what it describes, and treating it all as annual is why organisations discover a gap at offer stage rather than at review time.

• Established roles in stable functions. Annual is adequate. The market moves within the band a survey already describes.
• Scarce or contested roles. Every six months, and sooner where you have lost anyone. Pay in a squeezed segment can move a full quartile inside a year.
• Roles you are hiring for the first time. At the point of opening them, because there is nothing internal to anchor against and a stale figure becomes the anchor by default.

Two events override the calendar entirely. Losing two people from one level in quick succession is the market telling you something before anybody says it. So is a candidate declining at offer stage with the reason unstated, which is usually the reason.

What every source misses

Three things sit outside all four datasets, and they decide more offers than the benchmark does.

• What this person will accept. Shaped by their notice period, unvested equity and how content they are where they sit.
• What a competitor will offer. Visible only once it has happened, and by then the candidate has decided.
• What the role is worth to you. A function stalled for six months costs more than the difference between the median and the upper quartile, and that sits outside every dataset.

A benchmark is a starting position for a decision. Treating it as the decision is how organisations arrive at offers that are defensible, competitive on paper, and declined.