How Britain pays, 2020 to 2025

What we found

The finding we would not publish on its own

Bacs payroll volume is the line our readers care about most, and it fell in 2025: from 410,159 to 377,933 thousand transactions, a drop of 7.9%. Occupational pensions fell in the same year. The obvious headline writes itself: employers are moving payroll off Bacs.

The data does not support that conclusion, and the reason goes back two years further. In 2025, Pay.UK’s catch-all line — “unspecified / unclassified credits” — rose by 26,990 thousand transactions, enough to cover 41% of the three largest declines combined. But the more telling year is 2023, when that same line fell by 103,609 thousand transactions — a larger move than any named category makes in any year of this series. Payroll rose 51,256 that year, after rising 10,019 and 4,410 in the two before it. Volume that had been sitting in the unnamed line got names.

That makes payroll’s 2023 peak of 410,263 the figure to distrust, not the 2025 fall. Some share of the 2023 jump was reclassification, so some share of the later decline is that reclassification unwinding — and the published figures give no way to say how much. Measured against 2022, Bacs payroll volume is up 5.3% over three years. Measured against 2024, it is down 7.9%. Both are true. We report both and draw no conclusion about whether payroll is leaving Bacs.

This is the reason the study exists. Read the PDFs one at a time and none of this is visible; read them as a series without the check and you publish a migration that may not have happened.

What actually grew

The eight largest verified rises, from categories with at least five million transactions in 2020. Smaller categories are excluded because percentage changes on a low base say more about the base than the behaviour.

What shrank

The benefits lines tell a coherent story rather than a surprising one. HMRC Child Tax Credits fall from 52,263 to 178 thousand transactions while Universal Credit rises by 62,428 — the payment data tracking the move to Universal Credit, which DWP completed for tax credit claimants in 2025. The rise is larger than the tax-credit fall and smaller than the sum of all the benefit declines, so it is a match in shape rather than an exact transfer. We use it mainly as a check that the series moves the way a known change predicts it should.

The series we threw out, and why

A reclassification is indistinguishable from growth if you only compare the first year with the last. These 6 categories each make one move far larger than anything else in their own history. Some settle at the new level; regular savings does not, and dividends keep drifting afterwards — which is part of why we publish the excluded series in full rather than asking you to take the exclusion on trust. An excluded figure a reader cannot see is a figure they cannot check.

“Facilities management” is the clearest case: four years between 1.1 and 1.6 million transactions, then 125 million. Reported as a six-year change that is growth of over eleven thousand per cent. Reported honestly, it is a category boundary that moved.

Method

Pay.UK publishes the Bacs annual processing statistics as one PDF per year. The figures are transaction counts in thousands, split between Bacs Direct Credit and Direct Debit and broken down by payment purpose. There is no machine-readable release and no combined series.

1. Every payment-purpose figure was extracted from the 6 annual PDFs by script, not by hand.
2. Category labels were normalised, because spellings and punctuation differ between editions.
3. Only the 76 categories present in every year were kept. A category appearing in five years out of six cannot be part of a six-year comparison.
4. Step test: a series is treated as reclassified when its largest single-year change exceeds 50% and is more than ten times the median of its other year-on-year changes. A genuine trend spreads across years; a reclassification sits in one step.
5. Displacement check: every year’s largest declines are compared against the change in “unspecified / unclassified credits”. Where that line grows as named lines fall, we say so.

The limits are worth stating plainly.

• Counts, not values. A payroll run and a magazine subscription weigh the same.
• Bacs only. A payment that moved to Faster Payments disappears from view rather than showing up as a transfer, so this data can never prove where volume went.
• Coverage. The 66 trendable series are about 84% of the Bacs total Pay.UK reports. Conclusions are about most of Bacs, not all of it.
• The step test is one-sided. It catches a category that moves once and stays moved. It is weaker against one booked in and then back out again, because the second step raises the median it is measured against — which is why a second rule looks for that round trip specifically. It also misfires the other way: dividends were excluded on a 55% rise in 2021, which is plausibly the real rebound from the 2020 suspensions rather than any bookkeeping change.
• Restatements are invisible. Each PDF carries one year, so there is no way to check whether the 2025 edition and the previous one agree about the overlapping year. A silent revision would enter this series looking like a real change.
• The 5-million floor is a choice. Percentage changes on a small base say more about the base than the behaviour, but the threshold is ours, not Pay.UK’s.

The excluded series are published on this page rather than hidden, so the line can be redrawn by anyone who disagrees with where we put it.

Common questions

What does this study add to what Pay.UK already publishes?

Pay.UK publishes the figures — one PDF per year, each a snapshot. What it does not publish is the time series. To see how a single payment purpose moved between 2020 and 2025 you would have to open six PDFs and reconcile the category names by hand, because they change between editions. We did that, and then applied two tests to separate genuine movement from bookkeeping.

Why exclude some categories from the trend analysis?

Because a reclassification looks exactly like growth. “Facilities management” sat between 1.1 and 1.6 million transactions for four years and then reported 125 million. That is not a business trend, it is a category boundary moving. Two rules catch this: a series is excluded when its largest single-year change exceeds 50% and is more than ten times the median of its other years, and also when it more than trebles at some point but ends within a quarter of where it started — the signature of volume booked in and then back out again, which the first rule cannot see. Excluded series are listed on the page rather than quietly dropped.

Did Bacs payroll volumes really fall?

The reported figure fell, and we would caution against the obvious reading. The bigger problem is two years earlier: in 2023 Pay.UK’s catch-all line lost more volume than any named category moves in any year of this series, while payroll jumped. That makes the 2023 peak the figure to distrust, not the later fall. Measured from 2022, Bacs payroll volume is up over three years; measured from 2024, it is down. Both are true, and the published figures do not let anyone say which reading is the real one.

Are these values in pounds or transactions?

Transactions, in thousands, exactly as Pay.UK publishes them. The annual processing statistics count payment instructions, not value, so a large supplier payment and a small subscription collection each count once.

Can I use these figures?

The analysis is ours and you are welcome to cite it with attribution. The figures underneath it are Pay.UK’s, and their terms do not permit redistribution — so we do not publish the reconstructed series as a downloadable dataset, even though that would make the work easier to check. What we do instead: the six source PDFs are free on Pay.UK’s site and our extraction script is in our repository, so anyone can reproduce the series themselves.

Source: Pay.UK, Bacs annual processing statistics, editions 2020 to 2025, retrieved 13 September 2026. Figures are transaction volumes in thousands, as published. The analysis, the two tests and any errors in them are ours, not Pay.UK’s.

Written by the financeb2b editorial team · Last updated 12 September 2026 · Editorial policy