Methodology
How we calculate what we show you.
Salary advice is easy to fake and hard to check. So here is exactly where our numbers come from, how thin the data has to get before we stop showing it, and the claims we will not make even when they would sell better.
Where the numbers come from
Our measured figures are computed from real UK job postings — advertised roles with advertised pay. We don’t survey members and we don’t buy a panel. Where the postings run out we will say so, and may show a researched or estimated figure instead — labelled as such, never blended into a measurement. The three kinds are set out below.
For a given role and city we take every posting whose title matches, annualise each advertised range to a like-for-like yearly figure, and report the 25th, 50th and 75th percentiles of those midpoints. Matching is deterministic keyword matching on the title, so we can always tell you how many postings a band is built from — and we do.
Postings age out. A band reflects what the market was advertising recently, not what it paid five years ago.
Five postings, or nothing
Below five matching postings we return no figure at all. Not a hedged one, not a wider range — nothing, with an explanation.
Two reasons, and both matter. A “median salary” drawn from two or three adverts can identify the specific jobs behind it. And a percentile over a handful of postings isn’t a measurement, it’s noise with a decimal point — the kind of number that feels authoritative in a salary conversation and falls apart the moment anyone checks it.
What our confidence labels mean
Every band we show carries its sample size and one of these labels. They’re deliberately coarse — a finer scale would imply a precision the underlying data doesn’t have.
| Label | Built from | What it means |
|---|---|---|
| Low | 5–19 postings | Directional. Enough to be worth seeing, not enough to argue from. |
| Medium | 20–99 postings | Usable. The shape of the market for that role and city is real. |
| High | 100+ postings | Solid. A figure we'd be comfortable taking into a negotiation. |
Three kinds of number, never mixed
Not every figure we show is the same kind of thing, so we say which is which every time. The label is part of the number.
Measured
Percentiles computed from real UK postings we hold, with the count shown. This is the only kind that carries a high confidence label, and the only kind built on data we gathered ourselves.
Supported by online research
Where our corpus is too thin, we can fill the gap from published sources — salary reports, surveys, job adverts. When we do, the sources are listed with the figure so you can read them yourself. A number our own checks couldn’t trace back to a specific source is discarded rather than shown, because an uncited figure is indistinguishable from a guess. These never reach the confidence a measured sample earns.
Estimated
A model’s read of the going rate, with no specific source behind it. Always labelled as an estimate, always pinned to low confidence, and never presented as a percentile. It exists so a thin corpus produces something useful rather than a dash — not so it can stand in for evidence.
What we refuse to do
These aren’t aspirations. Each one is enforced in the code that generates your documents and your numbers, because a policy that lives only in a document is a policy that gets forgotten under deadline.
We don't invent your experience
Tailoring reorders and reframes what's already in your CV. It never adds an employer, a project, a qualification, or a technology you didn't list. Generated output is checked against your own stated skills and history before you ever see it, and rejected outright if it drifts.
We don't pass off an estimate as a measurement
Where our corpus is too thin to price a role, we may show a figure drawn from published sources instead — always labelled, always with those sources listed so you can check them, and never mixed in with a measured percentile. A figure nobody could attribute isn't shown at all.
We don't price a market we haven't measured
Our corpus is UK job postings. Ask for a salary in a currency we hold no postings in and you get nothing rather than a converted guess wearing the wrong label.
We don't claim we caused your outcome
If members who did X went on to land offers more often, that is an association. Plenty else differs between those people. We report it as association and we don't dress it up as cause.
We don't show a number we can't stand behind
Below five matching postings there is no median, no percentile, no band — the space stays empty and says why. A metric that flatters the product is worse than no metric.
What we cover
Every role.What changes from one to the next isn’t whether we’ll benchmark it — it’s which of the three kinds of figure above we can give you, and we say which every time.
Our own posting corpus is UK job adverts, and it’s deepest in engineering, data, product and design — so those roles are most often measured. Elsewhere you’re more likely to see a researched figure with its sources listed. That’s a difference in provenance, not in whether we’ll help.
We build for people in the middle of a career. Early on, the honest advice is usually to stay somewhere long enough for it to count — a tool for moving is selling against that. In people leadership, moves happen through relationships and headhunters before a role is ever advertised. In between is where knowing the number actually changes what happens, and that’s who we design for. Anyone can use it.
Seniority and leadership are two different things, and we keep them apart. A Staff Engineer, a Principal Designer, a Consultant anything — senior, and hired through the same public postings and advertised bands as everyone else, so a benchmark changes the outcome just as much. They’re squarely who this is for. It’s the move into running an organisation that stops working this way, not the move up.
Found a number that looks wrong?
Tell us. A figure we can’t defend is a bug, and we treat it as one.
Get in touch →