What the ratio actually counts
The Office for National Statistics publishes a figure that gets quoted more than it gets read: 2.5 unemployed people per vacancy, for March to May 2026. It is arithmetic on two published stocks. The numerator is people classified as unemployed - out of work, available to start, and actively looking. The denominator is vacancies open in the period, 712,000 of them.
What that ratio does not count is almost everything a hiring market feels like. It does not count applications, and applications are what both sides of the market actually meet. One unemployed person can send fifty in a week; the ratio counts them once. It does not count the employed people applying for other jobs, who in most markets are a large share of any applicant pool and are, by definition, not unemployed. And it does not count the economically inactive - at 20.9% of working-age adults, a fifth of the population is outside the numerator entirely, including people who would take a job if one arrived.
So a recruiter buried under a posting's worth of applications and a national ratio of 2.5 are not in contradiction, and neither one is wrong. They are answers to different questions. The ratio measures how thinly the stock of advertised work is spread across the stock of jobless people. It was never a measure of competition per application, and it cannot be turned into one by dividing differently.
| Measure | Latest | What it actually answers |
|---|---|---|
| Unemployed people per vacancy | 2.5 | How many jobless people share each open post |
| Vacancies open | 712,000 | How many posts are advertised at once |
| Unemployment rate, 16+ | 4.9% | What share of the labour force is jobless |
| Economic inactivity, 16 to 64 | 20.9% | Who is outside the count altogether |
It stopped rising. Read that carefully.
The ONS notes that the ratio has remained at 2.5 since July to September 2025, after increasing quarter on quarter from July to September 2024. A number that climbed and then stopped climbing invites one word - stabilised - and that word is doing more work than the data supports.
Two things sit against it. First, the level: the same release puts the ratio up from 2.3 in the same period a year ago. A plateau is only good news relative to the climb that preceded it; measured against a year earlier, more jobless people are sharing each post than were sharing it then. Second, and less obvious: a quotient can hold still while both of its terms move against you. Vacancies fell by 7,000 on the quarter and 18,000 on the year. If the ratio did not move while the denominator shrank, the numerator was not improving either.
That is the reading this edition exists for. Flat is not the same as better. The ratio stopping is compatible with a market that is still contracting on both sides at similar speed - and the other published series point that way, with payrolled employees down 90,000 over the year and unemployment at 4.9%, up from 4.7%.
The fall is smaller than the ruler
Here is the part that rarely survives into a headline. The ONS states the confidence interval on its vacancy estimate as approximately plus or minus 32,000. The quarterly fall being reported is 7,000. The annual fall is 18,000. Both are comfortably inside the measurement error of the instrument that produced them.
That does not make them false, and it is not a reason to ignore the series. It is a reason to stop treating each quarterly print as an event. A 7,000 move on a 712,000 base, measured with a ruler marked in 32,000, is not a signal you can act on by itself - it is one reading in a trend, and it is the trend that carries the weight. The one gap here that clears the error bar comfortably is the distance to where this series used to sit: vacancies are 77,000 (9.7%) below their January to March 2020 level, before the pandemic.
The ONS is candid about this in its own language, and the candour deserves repeating rather than editing out. Labour Force Survey estimates are published as official statistics in development, and the agency states plainly that caution should be taken when drawing conclusions from short-term changes. A desk that quotes the 2.5 and drops that sentence has taken the number and left its warranty behind.
| Reference period | Vacancies |
|---|---|
| January to March 2020, before the pandemic | 789,000 |
| April to June 2025 | 730,000 |
| January to March 2026 | 719,000 |
| April to June 2026 | 712,000 |
Where the vacancies actually went
The aggregate hides the only part of this that is directly useful, which is that the contraction is not evenly spread. Over the year, the largest falls were in human health and social work, down 14,000, and accommodation and food service, down 9,000. On the quarter, professional, scientific and technical activities and human health and social work each shed 4,000.
Health and social work appearing at the top of both lists is worth sitting with, because it is the sector that spent the post-pandemic years as the reliable answer to "where is the hiring". It is no longer that answer, at least not in advertised volume. Whether that is budget, retention, or posts being filled and not replaced, a vacancy count cannot say - it counts what is advertised, not what is needed.
For anyone reading this while applying: the ratio is a national average and you are not applying nationally. The number that governs your week is the one for your sector and your area, and the two sectors above moved far enough to change the arithmetic for the people in them. For anyone reading it while hiring: postings are almost 10% below their pre-pandemic level while the number of jobless people sharing each one has risen from 2.3 to 2.5 in a year. That is a market in which screening quality, not sourcing volume, is the binding constraint - we are not quoting an applications-per-post figure here, because we could not source one.
We publish how our own matching is tested, including where it is weakest, at our bias audit. If a number here is wrong, that page is where the correction should start.