How the numbers hold up
A benchmark is only worth as much as it can be defended. Here is exactly how a lumi figure is built — the definition behind it, how the data is validated, how the maths works, and how deep your peer group is — so it survives a Remuneration Committee, not just a first glance.
How the benchmark is built
One standard definition
Every metric has a single, published definition and unit, so every member answers the same question the same way. No two organisations are quietly measuring different things.
Validated on entry
Answers are checked as they are entered — units are enforced, and out-of-range or internally inconsistent responses are flagged before they ever reach the pool.
Medians, not averages
Every comparison is a median or percentile across valid peer answers. One unusually generous — or unusually lean — organisation can’t drag your position around.
The n≥3 floor
When fewer than three employers answered a question, lumi hides the whole card — on screen and in exports — so no single employer's data can be worked out. If a peer cut is too thin to be safe, you see that, not a fragile number.
Percentiles, cited
Percentiles are computed by linear interpolation across the valid peer set, and every figure carries its percentile, peer group and n (the number of organisations behind it). Nothing is unattributed.
Where the data comes from
The comparison pool is the reward data of UK organisations that shared it with lumi directly, quality-assured metric by metric against our anchor register of graded published UK sources. As members join, their answers join the same pool. Every metric is quality-assured against our anchor register — graded published UK sources with explicit bases — so a figure outside the known pattern is investigated before it reaches a comparison. Nothing is scraped and nothing is bought.
A share is computed once from the counts, never rounded twice. Inside the platform it is shown to the nearest whole percent, half rounding up, and a percentile is the midrank rounded the same way; below 30 organisations a share is shown whole. The research papers publish the same shares to one decimal place, which is why a distribution there can add to 100.1%.
How deep is your peer group?
The answer to the question every RemCo asks. You choose the lens; lumi shows you the depth behind it — and suppresses anything too thin.
Choose your lens
- All peers — the full UK pool, the deepest cut.
- Your sector — one of 14 UK sectors.
- Your size band — from 50 to 10,000+ employees.
- Organisations most like you, or a custom peer group you build.
And how many sit behind it
- 275 organisations contribute. Specialist metrics run thinner than the rest, and every figure carries its exact n. The current count is on the register.
- 48–65 in each employee-size band.
- 9–37 in each sector, where the great majority of metrics still clear the floor.
- Every peer cut carries a confidence label: 20 or more employers reads as High confidence; 5–19 reads as Directional — a steer, not a verdict; 3–4 reads as Indicative — too few to read alone.
- Cross a sector and a size band and the pool thins — where fewer than three employers remain, lumi shows nothing rather than something fragile.
Illustrative, for a fictional organisation. The last row falls below the n≥3 floor, so lumi hides the card for that cut (shown as “—”) rather than publish a figure fewer than three employers answered. On the demo, we show the real depth for your cut.
Kept current
- The benchmark recomputes as members contribute and update their data — a pool that changes, not a once-a-year snapshot.
- When your position moves against your peers, a signal fires and lumi emails the right people — ranked by market gap, with the evidence and n attached.
- Every figure states the peer group and n at the moment it was produced, so an export is always self-describing.
What lumi deliberately does not do
- It does not benchmark base salary — it benchmarks the rest of reward, and is designed to sit alongside a pay survey, not replace it.
- It holds no individual employee records — it benchmarks reward practices and policies, so there is nothing personal in the pool.
- It does not fill thin cuts with estimates — where the data is not there, or rests on fewer than three employers, you are told rather than shown a guess.
Public headline figures
A small set of national headline figures is published at lumihr.co.uk/metrics, one page per metric, so a figure can be cited with its base and fieldwork window. Each set is built from a frozen copy of the pool, never from the running platform, and every figure is checked against that copy before its page is written.
- Pool composition. Figures describe organisations in the lumi HR community, not a random sample of UK employers. Only the national headline is public; sector, size and peer-group views stay inside membership.
- Collection window. Every page states the fieldwork window of the data behind it, taken from the snapshot the figure was computed from.
- Not applicable and don’t know. Whether either answer leaves the base is set per question, not across the board. For a question that only makes sense for some organisations — income protection waiting periods, where the employer has no income protection — those answers are declared exclusions and leave the base, so the percentage describes only the organisations that could be assessed. For every other question they stay in it. Each figure says which rule applied to it.
- Public floor. No public figure rests on fewer than 50 organisations — a higher bar than the n≥3 floor inside the platform — and every figure carries its own base.
- Excluded organisations. lumi’s demonstration organisations, staff accounts and any test account are removed before a public figure is computed.
What a public page is not
- Not a base salary figure. Pay levels are never published — only how pay is managed.
- Not a time series. Each page states one figure for one fieldwork window, with no comparison to an earlier window.
- Not a peer cut. Membership shows where your organisation sits against sector and size peers on every metric; the public page does not.
See it computed on your own data
Thirty minutes — we’ll show you the real depth of your peer group and how each figure is built.