Almost every small firm runs software that only almost fits. The CRM that needs a spreadsheet bolted to the side. The job system nobody exports from without retyping. The subscriptions that overlap just enough to be annoying, but not enough to cancel.

The usual numbers used to describe that problem are mostly junk once you trace them. The industry’s favourite feature-waste stats come from software-maker analytics, a 2002 conference talk about four internal applications, decade-old enterprise desktop-licence research, or worked examples from vendors selling the cure.

The honest picture is narrower, but more useful. Off-the-shelf software plainly wins for standard jobs. The “drowning in unused subscriptions” story is mostly an enterprise story. But the real small-business cost still exists: it is the fit tax — the paid time your team spends carrying information across the gaps between systems that almost work.

There is no reliable external benchmark for that. Not for a ten-to-fifty-person UK firm. The only number worth having is the one you measure yourself.

The claim landscape: the numbers that don’t survive

Search for the cost of off-the-shelf software and the first page is written almost entirely by companies selling the alternative — custom-software agencies, or SaaS-management vendors selling the cure for subscription sprawl.

That is not a complaint. It is a measurement. And it matters, because when a whole genre is written by sellers, the numbers tend to drift in the seller’s direction.

Here is where the load-bearing claims actually come from.

“80% of software features are rarely or never used.” This is the fit stat, the one that supposedly proves you are paying for software you do not need. Its modern source is real and named: Pendo’s 2019 Feature Adoption Report, which analysed feature usage across 615 of Pendo’s own customer subscriptions and found that 80% of features were “rarely or never” used — though the honest breakdown is that 24% were never touched and the rest were simply used rarely.

The trouble is what the number measures. Pendo sells product-analytics software to the people who build applications; the report is a pitch to software makers about which of their own features get clicked, and it prices the finding as $29.5 billion of wasted cloud-vendor R&D. It says nothing whatsoever about a buyer wasting money. Relabelled from “the average product ships features few people use” into “you waste 80% of the software you pay for”, it becomes a different claim its own data cannot support.

“64% of features are rarely or never used.” The older, more-abused cousin, usually cited to the Standish Group. Trace it and, as agile consultant Mike Cohn documented, it comes from a keynote the Standish chairman gave at a conference in 2002, “based on a study of four internal applications. Yes, four applications. And, yes, all internal-use applications. No commercial products.” It was never published as a study; there is no methodology to read. Cohn’s plea is blunt: “if you’re citing this data and using it to imply that every product out there contains 64 percent ‘rarely or never used features,’ please stop.” A number about four bespoke in-house apps is now routinely quoted to prove off-the-shelf software is wasteful.

“$34 billion is wasted on unused software.” Still quoted in 2026 as if it were a SaaS figure. It comes from 1E’s Software Usage and Waste Report — dated 2016, measuring on-premise desktop software such as WinZip, Visio and Project that had been installed but “not run within the past 90 days”, across 149 enterprises averaging around 30,000 seats each. Three things break it for a small firm: it is desktop licensing, not SaaS; it is enterprise-scale, not ten-person-scale; and it is a decade old, published by a vendor whose old web address now redirects to TeamViewer. A decade-old desktop-licence study from giant estates is not evidence about your subscriptions.

“25% — or is it 30% — of SaaS spend is wasted.” There is a real Gartner figure here: a 2021 Gartner note (abstract) forecasts that 25% of SaaS software “will be underutilized or overdeployed” — an analyst prediction, about enterprises, with the detail behind Gartner’s paywall. What you meet in the wild is the escalated version: “30% of SaaS spend is toxic”, “$45 billion wasted industry-wide”, attributed to Gartner with a bare name and no document. Those bigger numbers do not appear in the Gartner note. They are the attribution graduating, quietly, into a fact of its own — which is exactly the move that should make you distrust the whole genre.

“UK small firms waste up to £10,000 a year on unused software.” The most quotable UK number in the field, and it is not a survey finding at all. It originates as a worked example from Fasthosts, a web-hosting company, imagining a fifteen-person marketing agency and calculating that “removing just two tools could result in savings of up to £10,000” — then syndicated until it reads like measurement. It is arithmetic about a firm that does not exist.

The pattern is always the same. A real number, measured on the wrong population — usually a large enterprise — for the wrong purpose — usually selling software — loses its date, its size, and its hedge, and walks around as a fact about your firm.

What the evidence actually says

Strip the genre back to what survives, from sources that are not selling you anything, and the picture is smaller and more honest.

Off-the-shelf already won the standard jobs. The government’s Longitudinal Small Business Survey 2024 — 8,396 SME employers — finds that 69% used technologies or web-based software to sell to customers or manage the business. Among those digitally active firms, accountancy software was used by 80%. That is not a market with a fit problem in its core admin; it is a market that has, sensibly, bought the standard tool for the standard job.

The same survey shows where adoption thins out: customer-relationship management runs at 26% of micro firms, 42% of small and 54% of medium; HR management software runs at 14% of micro firms, 44% of small and 62% of medium. The further a tool sits from a universal task, the less universal adoption becomes.

The official worry is under-investment, not waste. This is the finding that cuts hardest against the “SMEs drown in software” story. When the Bank of England and DBT surveyed 2,885 UK SMEs about their investment, 76% felt they had invested about the right amount, 22% felt they had invested too little, and just 2% thought they had invested too much. A market where one firm in fifty thinks it over-bought is not a market with a shelfware epidemic.

The real problem has a name, and it is fit. The OECD’s 2026 review of UK SME technology adoption puts it in flat official language: many SMEs face “difficulty in the identification of solutions that match their operational needs”. It also names the cost the sticker price hides — “not only the upfront investment costs but also implementation, maintenance and subscription costs”. That is the fit problem, described by an institution with nothing to sell.

And more software is not a reliable fix. The Enterprise Research Centre found, even among advanced digital tools, that “more isn’t always better” — some combinations lower productivity through “overlapping functions, integration challenges, or the complexity of managing multiple systems”.

And the sprawl is real — for someone else. The alarming waste numbers are not fabricated; they are just measured on giants. When Zylo reports about half of provisioned licences sitting unused, or BetterCloud counts an average company running over a hundred applications, they are describing corporate estates of thousands or tens of thousands of seats, not a ten-to-fifty-person UK firm. Tellingly, even BetterCloud — a company that sells the cure for sprawl — reports the number of apps per company falling for the first time in a decade, from 130 in 2022 to 112 in 2023, as firms consolidate. The runaway-sprawl narrative is softening even where it was strongest, and it was never really about you.

The real cost is the fit tax

Notice what the honest evidence does not give you: a pound figure for what off-the-shelf software costs a ten-to-fifty-person UK firm.

I looked hard. I could not find one — not at the ONS, not at DBT, not in any vendor report segmented small enough to matter. Every waste percentage is enterprise-scale and sold by someone; every spend figure, as the chart above shows, swings wildly with size, country and sector. Even Cledara’s own numbers put per-employee spend anywhere from $1,741 to $8,000.

That gap is not just a research failure. It is the point.

The cost of almost-fitting software is genuinely hard for national statistics to see, because it does not live neatly in the subscription line. It lives in three places the accounts never separate.

• The workaround. The spreadsheet bolted onto the CRM. The figures retyped from one system into another because they do not connect. The manual export, every Friday, that turns two tools into one report. This is labour, and it is usually the largest hidden cost — and the one, tellingly, that not a single survey in this field even tries to price.

• The overlap. The three tools that each do part of the job because none does all of it. Cledara found that firms underestimate their own tool count by around 40% — “for every 10 tools a company thinks they’re using, there are actually 14 in play”. You cannot manage, or cancel, what you have lost count of.

• The unused seat. The licences bought for a team that shrank, the annual plan nobody reviews. Real, but — for a small firm — usually the smallest of the three, and the easiest to fix without building anything.

A small firm runs around 36 applications, on Okta’s 2024 identity-network data for firms of 50 or fewer staff worldwide — itself a number to hold lightly, since it comes from firms organised enough to buy single sign-on. But 36 tools is 36 possible seams, and every seam is a place where information may need to be carried across by hand.

That is the fit tax: the paid human effort required to make almost-right software behave as if it fitted properly.

Reckon your own fit tax

Because no honest external number can price this for you, here is the method that can. It takes an afternoon, and it is the same shape as any decent cost estimate: count, time, and price.

1. List the seams. Every place the same information is typed into two systems, exported from one to feed another, or held together by a spreadsheet. Be specific: “bank lines into the job sheet”, “job sheet into the invoice”, “enquiry email into the CRM”.

2. Time one crossing of each, with a clock. Not a guess — every failed number in this article began as somebody’s guess. Then count how often it happens in a normal week.

3. Price the labour. Minutes ÷ 60 × weekly occurrences × 46 working weeks × the loaded hourly wage of whoever does it. Loaded wage means their gross hourly pay plus employment costs. For a rough UK SME estimate, gross pay plus 15% employer National Insurance is a better starting point than pretending staff time is free.

4. Add the overlap. List every subscription. Mark the ones whose job substantially overlaps another’s, and the ones you had lost count of. That annual total is the ceiling on what consolidation could return.

For example:

Four boring seams: £10,580 a year.

The exact number is not the point. The discipline is. A measured £6,000 problem should not be sold a £40,000 build. A measured £30,000 problem should not be waved away because the subscription itself only costs £99 a month.

The question is not “could this be automated?” Most things can. The question is:

Is this crossing costing meaningfully more than fixing it would?

That is the honest bar.

What it means for you

For most firms, most of the time, off-the-shelf software is the right answer. This piece is not an argument against it.

The standard tool for accounts, payroll, email, card payments, bookkeeping or tax filing is cheaper, better-maintained and more secure than anything worth building from scratch. That is exactly why off-the-shelf dominates the standard jobs. If your loops are small, your processes ordinary and your tools genuinely talk to each other, the fit tax is a rounding error.

Leave it alone.

The tax bites in a narrower case: when your process is genuinely non-standard — the thing you actually do differently from your competitors — and the off-the-shelf tools were built for the average firm, not yours.

The signals are familiar:

• you export to a spreadsheet to make two systems agree;
• you keep a tool for one feature and pay for the other ninety;
• a person spends part of every day being the pipe between two systems;
• management information exists, but only after someone has assembled it by hand;
• the business has changed, but the software stack still reflects how things worked three years ago.

That is the fit tax, and it compounds quietly on the payroll where no statistic can see it.

None of this requires a grand replacement. In fact, the grand replacement is usually the dangerous option. Custom software has earned its bad reputation precisely when it was sold as a big-bang rescue.

The honest move is smaller: take the single most expensive crossing on your list — the one costing four hours a week — and close just that gap, with something built to fit the way you actually work. Measure it. Then decide whether the next one is worth touching.

Built one fitted module at a time, against a number you timed yourself, custom software stops being a leap of faith. It becomes a normal, reversible business decision.

That is the work I do: building the custom processes and workflows — increasingly with AI at their core — that close the specific gaps off-the-shelf software leaves, one measured module at a time.

If you would rather someone else held the stopwatch first: book a free process audit. We will map your worst software seams, time the manual work, and tell you whether custom software is worth touching.

And if the honest answer is that off-the-shelf already fits, I will tell you that too.

Sources and method

Every claim above is hyperlinked where it is made; the notes below restore the labels — sponsor, sample, date — and add an archived copy of each source, checked 6 July 2026 and re-verified 14 July 2026. Numbers named in the debunk as manufactured or misapplied are named but, deliberately, not linked: no authority is passed to bad figures.

• Pendo, 2019 Feature Adoption Report, 5 Feb 2019. Product-analytics vendor; feature-usage across 615 Pendo customer subscriptions; “80% of features… rarely or never used” (of which 24% never). A software-maker’s metric, priced as cloud-vendor R&D. Primary PDF · archive.

• Mike Cohn, “Are 64% of Features Really Rarely or Never Used?”, Mountain Goat Software (updated 13 Nov 2016). Traces the “64%” to Jim Johnson of the Standish Group, a 2002 XP conference keynote based on four internal-use applications, never published. Live · archive.

• 1E, Software Usage and Waste Report, 2016. On-premise desktop software “not run within the past 90 days”; 149 enterprises, 4.6 million machines, ~30,000 seats average, US + UK; “$34 billion” combined waste; 38% total (30% unused + 8% rarely). Vendor sold licence-reclaim tooling; its old domain now redirects to TeamViewer. Mirror PDF · archive.

• Gartner, “Why Are You Wasting Your SaaS Expenditure?” (infographic, doc 4006574), 7 Oct 2021. Forecast that “25% of… software will be underutilized or overdeployed”; enterprise-framed; full document paywalled. The circulated “30% toxic / $45bn” figures do not appear in it. Abstract, archived.

• Fasthosts / SME-Today, “UK SMEs wasting up to £10k a year on unused SaaS tools”, Apr 2026. The £10k is an illustrative worked example (a hypothetical 15-person agency), not a survey; named here as a specimen of the genre, deliberately not linked.

• DBT, Longitudinal Small Business Survey 2024 (SME employers), 25 Sep 2025. n = 8,396 SME employers (1–249 staff). 69% use technologies or web-based software to sell to customers or manage the business; among those, accountancy software 80% (89% in 2023); CRM 26% micro / 42% small / 54% medium. Live · archive.

• OECD, SME Technology Adoption in the United Kingdom, Apr 2026 (CC BY 4.0). “Difficulty in the identification of solutions that match their operational needs”; “implementation, maintenance and subscription costs”; ERP adoption 6.7% (2018) → 10.9% (2022). Live PDF · archive.

• Enterprise Research Centre, Research Paper 119 (exec summary), Oct 2025. LSBS 2022–23, nationally representative; among six advanced technologies (AI, cloud, BI, CAD, IoT, VR/AR), “more isn’t always better” — some combinations associate with lower productivity. Cited here as an advanced-tech finding, not a claim about basic SaaS. Gov-funded academic centre. Live PDF · archive.

• Bank of England / DBT, Finance and Investment Decisions Survey 2023 (Quarterly Bulletin 2024). n = 2,885 UK SMEs; 22% “invested too little”, 76% “appropriate”, 2% “too much”; “little variance… across firm size”. Live · archive.

• Cledara, 2025 Software Spend Report (published 30 Sep 2024) and “Average SaaS Spend Per Employee 2026” (25 Mar 2026). SaaS-management vendor. Per-employee spend ~$8,000 (up to 20 staff) → $2,583 (50–100) → $1,741 (100–200); UK ~$4,180 per employee at a 50-person firm; firms underestimate tool count by ~40%. Sample described in public materials as tech companies under 200 staff across the US, UK and EU; carried here as a vendor-sponsored directional benchmark, not a UK SME population estimate. Report (archive) · per-employee (archive).

• Okta, SMBs at Work 2024, Aug 2024. Identity-platform telemetry across 18,000+ companies; firms of 50 or fewer employees deploy “around 36 apps”. Vendor sells single sign-on; sample skews to digitally-mature firms. Live · archive.

• BetterCloud, State of SaaSOps 2024, Jul 2024. SaaS-management vendor; practitioner survey (company size undisclosed, skews larger). Apps per company 130 (2022) → 112 (2023), “first decline in over a decade”; 106 in 2024 on BetterCloud’s later restatement; 53% consolidated redundant apps. Press release · archive.

• Zylo, SaaS Management Index 2024 & 2026. SaaS-management vendor; enterprise estates (30–40 million licences, firms up to 10,000+ staff). Reports about half of provisioned licences unused (2024) and 36% “against recommended utilization” (2026) — different measures, not a like-for-like trend; cited only as enterprise context. 2026 edition · archive.

That’s roughly where the United States has just landed. And it’s a bigger problem for a ten-person firm than for the AI labs the rules are aimed at.

Three Financial Times articles over the past week paint a remarkably consistent picture.

First, OpenAI’s chief executive, Sam Altman, used the FT’s opinion pages to argue for a US-led international body to oversee frontier AI. His proposal resembles an International Atomic Energy Agency for artificial intelligence: countries would agree to common safety standards, companies would be certified against them, and access to the most advanced models would depend on compliance. Whether you agree with him or not, the underlying message is clear — if AI is becoming critical infrastructure, it needs a predictable rulebook.

The FT’s Lex column accepted the diagnosis but doubted the prescription. The closest historical comparison, it argued, is the Basel Committee on Banking Supervision — which made banking safer, but also demonstrates two uncomfortable truths: international standards take years to emerge, and complex regulation almost always favours the largest incumbents, because big organisations can afford compliance departments and smaller competitors often can’t. Any global AI framework would also face a reality banking largely avoided: China. A genuinely global agreement would need Beijing’s participation; without it, the world risks splitting into competing AI blocs — different standards, different models, restricted access between them.

Then came the most revealing article of all. The Trump administration’s departing AI adviser insisted there would be “no FDA for AI” — no formal licensing regime, no central regulator, no bureaucracy deciding which models reach the market. Yet in the same interview he confirmed that the White House had recently used emergency powers to force Anthropic to withdraw its most capable model and hold up OpenAI’s next release on national security grounds, while introducing a framework that gives the government time to review frontier models before deployment.

Read together, those positions expose the real issue. America hasn’t chosen deregulation, and it hasn’t built a comprehensive regulatory system. It’s regulating frontier AI through discretionary intervention — and that distinction matters.

Markets cope surprisingly well with rules. Businesses complain about regulation, but they’re remarkably good at adapting to it: give a company a clear constraint — even an expensive one — and it will budget for it, redesign its processes and move on. What businesses can’t cope with is uncertainty. Uncertainty can’t be priced, can’t be scheduled, and can’t be built into next year’s operating plan. It becomes its own tax.

That isn’t a figure of speech; it’s one of the better-measured effects in economics. Faced with an unclear future, firms don’t panic — they wait. The Stanford economist Nicholas Bloom showed in 2009 that when uncertainty spikes, companies temporarily freeze investment and hiring: with the picture unresolved, the rational move is to sit on your hands until it clears. He and colleagues later built an Economic Policy Uncertainty index from newspaper coverage and found that, in the United States, a jump in policy uncertainty of the size seen between 2006 and 2012 foreshadowed roughly a 6% fall in business investment. The uncertainty didn’t have to be bad news. It just had to be unresolved.

Britain has a fresher example, and a cautionary one. In the three years after the 2016 referendum — before a single trading rule had actually changed — Brexit uncertainty alone reduced UK business investment by around 11%, according to work by Bloom and colleagues using the Bank of England’s Decision Maker Panel survey of firms. Not a recession, not a new regulation — just years of not knowing the rules, and companies quietly deciding to wait. Whatever your politics, the economics are the point: the cost wasn’t the eventual outcome, it was the fog on the way there. That same fog is now forming around AI, only faster.

For the frontier labs, that fog is a lobbying problem. For small and medium-sized businesses, it’s an operational one.

Most SMEs don’t think they’ve built their business on AI. They think they’ve bought some software. Increasingly, that’s no longer true. If your CRM drafts emails using GPT, your support desk summarises tickets using Claude, your accountants review documents with Gemini, or your developers rely on Claude Code, then part of your operating model already depends on decisions made by organisations — and increasingly governments — thousands of miles away. Most days, that’s perfectly acceptable. One day, it may not be.

And this isn’t only about politics. Models are retired, APIs change, pricing changes, performance rankings shift every few months, new open-weight models appear, and providers merge or discontinue products. Government intervention simply adds another source of volatility to a stack that’s already moving at extraordinary speed.

So the mistake many businesses make is assuming the important question is which AI model they should choose. It isn’t. The better question is: how easily can we stop using it?

That might sound counterintuitive, but it’s how resilient businesses have been built for decades. Manufacturers diversify supply chains. Finance teams manage counterparty risk. IT departments avoid unnecessary vendor lock-in — because suppliers change prices, discontinue products, or occasionally fail altogether. AI deserves exactly the same discipline.

The emerging best practice isn’t to build directly against one provider’s model. It’s to put an orchestration layer between your business processes and whichever models sit behind them. That layer decides which model performs which task; it lets providers be swapped with minimal disruption; it keeps you running if one service goes dark; it can move sensitive work onto privately hosted open-weight models where appropriate; and it cuts cost by routing simple jobs to inexpensive models while reserving premium ones for work that genuinely needs them. In short, it turns changing AI providers from a redevelopment project into a configuration change — which is worth doing even if governments never intervene again.

The events of the past week simply make the point harder to ignore. You don’t need to predict whether Washington, Brussels or Beijing will regulate AI next year, or whether OpenAI, Anthropic, Google or an open-weight model leads the market in 2028. You only need to accept that the answer will change.

The businesses that benefit most from AI over the next decade won’t be the ones that picked the “winning” model in 2026. They’ll be the ones that designed their systems so they never had to care. Technology changes, politics changes, markets change — your business architecture should assume all three.

If you’re not sure where your business is quietly depending on a single AI provider, that’s exactly the sort of risk a free process audit should uncover — before the next Tuesday headline becomes Wednesday morning’s operational problem.

Sources

• Investment-under-uncertainty evidence (primary): N. Bloom, “The Impact of Uncertainty Shocks,” Econometrica (2009); S. Baker, N. Bloom & S. Davis, “Measuring Economic Policy Uncertainty,” Quarterly Journal of Economics (2016) — US data, index built from newspaper coverage; N. Bloom et al., “The Impact of Brexit on UK Firms,” NBER (2019), using the Bank of England’s Decision Maker Panel survey of UK firms (~11% lower investment over the three years to 2019).
• Regulatory developments as reported and argued in the Financial Times: Sam Altman’s op-ed (1 July 2026); the Lex column “Altman’s AI safety proposal: let us win, or everybody loses” (2 July 2026); and the interview with the departing White House AI adviser (3 July 2026).

Late payment is sold to small businesses as a discipline problem: chase harder, be firmer, get better at credit control. But when the UK government finally costed it properly in 2025, the bill didn’t fall where the advice points. Late payment costs the UK economy almost £11 billion a year (central estimate; the 90% confidence interval runs £4.7bn–£17.6bn). The single largest line in what it costs an individual business is staff time chasing money — about £2.3 billion. Each affected firm spends about 86 hours a year on it. The delay’s financing cost is real, but it doesn’t out-rank the labour.

Which points somewhere uncomfortable for the “be more disciplined” advice: the chasing is the cost, and the chasing is exactly the thing software does without a person. This is a software gap wearing a discipline costume.

The numbers everyone repeats

Search “cost of late payments UK” and one figure greets you everywhere: poor payment culture kills 50,000 businesses a year and costs £2.5 billion in lost output. It’s on vendor blogs, advisory sites and trade press, always in the present tense.

It comes from a single Federation of Small Businesses report — “Time to Act,” published in November 2016. A figure modelled nearly a decade ago is still being reported as this year’s news. And in 2025 the government checked it: its own econometric estimate of closures attributable to late payment came out at around 14,000 a year (a wide and fragile estimate — more on that below), not 50,000 — a roughly 72% haircut on the number the internet still repeats.

The “how much are small firms owed?” figure is no steadier. Depending on who you read, UK small businesses are owed £23.4bn (Bacs, 2019, from a survey of ~355 firms), £26bn (the government’s all-business figure, 2025), £70.4bn (Hiscox, 2026, extrapolated from 1,000 owners), £112bn (Sage, 2025) or £141bn (Xero, 2018). That’s a sixfold spread — and it isn’t because anyone’s lying. It’s because they measure different things: all businesses or only small ones; invoices genuinely overdue or merely on long terms; a survey of a few hundred firms or the actual invoice data of millions. The number you get is mostly a decision about what you count. Worth remembering the next time a statistic arrives without its method attached.

None of this means late payment isn’t a real problem — it plainly is. It means the size of it has been asserted more often than measured, usually by people with something to sell or a campaign to run. Which is exactly why the 2025 government study matters: for the first time, there’s a number you can check.

What the evidence actually says

The reason we can be precise now is that in July 2025 the Department for Business and Trade and the Small Business Commissioner published the first proper study of this — researched by London Economics, with academic advisors from Warwick and Aston, built on a survey of 1,455 businesses plus econometric modelling. It is not a vendor selling a fix. It is the source worth anchoring to, and its findings are more interesting than the headlines.

Start with what late payment is not. At any moment, UK businesses are owed about £26 billion in late payments. The instinct is to call that a £26bn cost. The report is careful to say it isn’t: it’s “£26 billion of interest free finance to their customers… not a net cost for businesses,” because every pound a supplier is owed late is a pound a customer is holding onto — a transfer, with winners and losers, not money set on fire. Almost every write-up drops that caveat. It matters, because it moves the real cost away from the amount and onto what firms do about it.

And what they do about it is mostly chase. Across the economy, chasing late payers burns an estimated 133 million hours of staff time a year — 86 hours per affected business. When the report adds up what late payment actually costs businesses out of pocket — roughly £7 billion — the biggest single line isn’t interest or bad debt. It’s staff time chasing debtors, at about £2.3 billion. Legal costs, debt collection and invoice financing follow, each around £1.1–1.2bn.

Source: DBT/OSBC & London Economics, “Late Payments Research,” July 2025 (survey of 1,455 UK businesses; government-funded).

The heaviest burden falls on the smallest firms. A micro-business affected by late payment is owed 4.6% of its turnover on average — the equivalent figure for a large company is 0.2%. Small firms are least able to carry the gap and least equipped to close it: the same study found only 20% of micro-businesses use invoicing software at all, against 67% of large ones. That gap is the quiet scandal in the numbers. The tool that turns chasing from a person’s job into a background process exists, is cheap, and is used by two-thirds of big companies — and by one micro-business in five. The firms carrying the heaviest load are the ones least likely to own the thing that lifts it.

And it doesn’t stay in one business. A firm paid late is often a firm that then pays late; the shortfall runs down the chain. In a 2025 survey of its members by the Federation of Small Businesses — run with the payments firm GoCardless, so read it as a trade body’s own members rather than a neutral sample — 36% of firms hit by late payment said it left them unable to pay their own suppliers on time, and 28% had turned to short-term borrowing to cover the gap. Sixty-one per cent said late payment was holding back their growth. The single overdue invoice is rarely the whole story; it’s the first domino.

And some invoices never arrive at all. Late payment shades into bad debt — the money written off when a customer folds or simply refuses to pay — which is the same problem at its sharp end. The estimates here are shakier and tend to come from lenders with a product to sell: one invoice-finance provider’s 2026 survey put the average sum a small firm writes off at around £29,000, affecting roughly a third of them. Treat the figure as directional given who’s counting, but the direction isn’t in doubt — the tail of late payment is money that turns into nothing.

Now the honest part. If you’re tempted to conclude the financing cost of late payment doesn’t matter, don’t. A 2024 peer-reviewed study across eleven European economies found that firms suffering late payment are around two percentage points more likely to be credit-constrained — the delay pushes up the price of the finance they can get and cuts how much of it they can get at all. (It’s a European dataset, not a British one, so treat it as the direction of travel rather than a UK figure.) And in the government’s own out-of-pocket numbers, the finance-related lines — servicing debt plus factoring, about £2.4 billion together — roughly equal the chasing line. So the fair statement is not “financing is irrelevant”; it’s that the chasing labour is comparable in size to the financing cost — and unlike the financing cost, it’s the part you can simply delete.

The financing cost is real, and it has grown. Xero, which says its figures come from its customers’ invoice data rather than a survey, put the interest cost of late payment to UK small firms at about £1.6 billion in 2023 — more than double its 2021 estimate, as higher base rates made the delay dearer to carry. (Xero sells accounting software, so weigh it accordingly — though invoice timestamps are harder to spin than survey answers.) The honest position isn’t that financing doesn’t hurt. It’s that of the two big costs, one is a price paid to a bank and the other is hours your staff never get back — and only one of them is optional.

One more caveat we’ll wear openly, because this piece’s whole point is that a number should travel with its method. The eye-catching “14,000 businesses close a year — 38 a day” is the government’s central estimate, but its 90% confidence interval runs from about 2,000 to 27,000, the effect is statistically firm only in the 2018–19 windows (and not under the report’s own stricter test), and it’s drawn from a 300-firm sample using pre-pandemic data. It’s the best estimate we have that late payment closes businesses. It is not a body count, and anyone who prints “38 a day” as hard fact hasn’t read the annex.

What we build

Here’s where the “be more disciplined” advice quietly contradicts itself. The same study notes that the fixes it recommends — charging interest on overdue invoices, demanding a purchase order before delivery, chasing before the due date rather than after — are barely used (each by under 7% of firms), and adds the telling line that doing more of them “would come at the cost of additional staff resource.” In other words: the cure for the labour problem, as usually prescribed, is more labour.

That’s the gap we build into. Credit control shouldn’t be a discipline you impose on a busy person; it should be embedded in the software the business already runs. When we build it into a business, the chase ladder runs itself — a polite reminder on day one overdue, a firmer one a week later, an escalation after that — and it stops the instant the payment lands, so nobody ever chases someone who’s already paid. It watches the terms, it never forgets, it never has an awkward week, and it logs every touch so the audit trail writes itself. The 86 hours the research prices don’t get reduced; they stop landing on a person at all.

This isn’t hopeful arithmetic — every source that tries to measure chasing measures it in hours. QuickBooks (an accounting-software firm, and so a motivated counter) put it at about four hours a week in its 2024 survey of UK small firms; the government’s figure works out at 86 hours a year for an affected business. Vendor and government numbers on this rarely land on the same total, and these two don’t either — but they agree on the shape that matters: chasing is repetitive, rules-based work, counted in hours, week after week. Which is exactly the kind of work software is built to absorb.

The embedding is the point, not a detail. Because the credit control lives inside the system the business already runs on — the same place invoices are raised and payments arrive — there’s no second tool to log into, no export-and-reconcile, no task that quietly becomes someone’s least favourite part of the week. The chasing doesn’t move from one person to another. It stops being a person’s job.

That’s the difference between treating late payment as a character flaw and treating it as what the evidence says it is: a cost that mostly takes the form of manual work — and manual work is the thing software is for.

What it means for a business like yours

If late payment hurts, the reflex is to blame the chasing — that you’re not firm enough, not on top of it, not chasing early enough. The numbers say the chasing is the cost, and being better at it by hand just means paying that cost more diligently. The £2.3 billion the country spends chasing money it’s already owed is the sound of a solved problem being solved by hand.

So the useful question isn’t “how do we chase better?” It’s “why is a person chasing at all?” Count the hours your team spends this month on reminders, statements and “just following up on my last email” — then decide whether that’s work, or work that could run itself.

Here’s a rough way to price it. Take the government’s 86 hours a year and put your own cost on the hour — say £20 to £30 once employer costs are loaded on. That’s £1,700 to £2,600 a year for each person doing the chasing, spent on work whose entire output is the money you were always owed, slightly sooner. Then add the invoices that came so late they nearly didn’t, and the few that never came at all. Framed that way, the question stops being whether automating it pays for itself and becomes why it hasn’t happened already.

If you would rather not do the arithmetic by hand, the late-payment calculator does both sums — the statutory interest and compensation you could add to what you’re owed, and what the chasing itself costs — free, no email required.

That’s exactly what the free process audit does: it maps your manual processes, chasing included, and shows you which ones a computer should be doing. You leave with the map whether or not we ever build a thing.

Book a free process audit →

Sources

1. Department for Business and Trade & Office of the Small Business Commissioner — “Late Payments Research: Estimating the total economic cost of late payments and their impact on the UK economy” (London Economics, July 2025). Survey of 1,455 UK businesses (YouGov/IFF, Jan–Feb 2025) plus econometric modelling; government-funded. The spine of this piece: the £11bn, £26bn, £2.3bn chasing, 86 hours, 14,000 closures and software-adoption figures.
2. Federation of Small Businesses — “Time to Act” (November 2016). The origin of the widely-repeated “50,000 closures / £2.5bn” figure; a representative-body estimate now superseded by (1).
3. Pay.UK / Bacs late-payment research (2019) — the £23.4bn “owed” figure; survey of ~355 firms.
4. Sage / Cebr (May 2025) — £112bn owed; analysis of 1.2m invoices from Sage customers. Vendor-funded (accounting software).
5. Intuit QuickBooks / Opinium (2024) — corroborating labour figures (~4 hours/week chasing). Vendor-funded.
6. Xero Small Business Insights (2018 and 2024) — the £141bn “owed” extrapolation and a £1.6bn lost-interest estimate. Vendor-funded; based on real invoice data.
7. Hiscox Late Payments Report (2026) — the £70.4bn figure; Censuswide survey of 1,000, extrapolated. Insurer-funded.
8. Kaya, “The impact of late payments on SMEs’ access to finance,” Economic Modelling vol. 141 (2024) — peer-reviewed; the financing-cost counter-evidence (credit rationing across eleven European economies).
9. Federation of Small Businesses / GoCardless — Late Payments Report (2025; survey of ~2,298 FSB members). The knock-on figures — unable to pay own suppliers, short-term borrowing, growth held back. Representative-body survey, co-produced with a payments vendor.
10. Bibby Financial Services (2026) — the ~£29,000 average bad-debt write-off figure. Invoice-finance provider; treated as directional given the commercial interest.

Few owners believe productivity statistics any more — and not because they are complacent. Many of the most repeated numbers do not deserve to be believed.

Re-keying — typing the same information into a second system by hand — is a perfect example. Every small business owner recognises the problem. Bank receipts copied into a spreadsheet. Emailed instructions retyped into a job system. Job sheets turned into invoices. Timesheets copied into payroll. Nobody needs a statistic to know that processes go stale, software ages, and workarounds quietly become part of the payroll.

The trouble is that the statistics used to prove this obvious point are often worse than useless. They are usually real figures, but with the labels torn off: date, country, sample, sponsor, method, and caveat. Restore those labels and the claims become much less useful for a UK small business.

Nobody directly measures what re-keying costs UK SMEs. Not the ONS, not HMRC, not the Department for Business and Trade. We searched the main official sources extensively. They measure wages, admin burdens, tax compliance, digital adoption, software use, and business population. They do not measure the thing itself.

Into that gap, five numbers circulate endlessly: “2.5 hours a day searching for information”, “$3.1 trillion a year”, “$12.9 million a year”, “120 working days on admin”, and “7–18% productivity improvement per tool”. Almost every one traces back to a real document. What has been stripped away in circulation is the context that lets you decide whether it applies to your firm.

A decade of that has bought the opposite of urgency. Owners have learned — reasonably — to shrug at the whole subject.

So this piece does two things. First, it puts the labels back on the five statistics you are most likely to meet. Second, it shows you how to produce the one number that actually matters: your own, timed in your own office, using the same basic arithmetic HMRC uses when it values admin time — hours multiplied by an ONS median wage.

As a scale marker: five hours a week of re-keying at a book-keeper’s loaded wage is about £4,300 a year. That is not a productivity vibe. That is a payroll line.

Why re-keying exists in the first place

Re-keying is rarely caused by laziness. It is usually caused by history.

A firm buys payroll software one year, a CRM three years later, a job-management tool after that, and a finance system somewhere in between. The bank has its own export format. HMRC has its own gateway. A supplier insists on a portal. A client sends work by email because that is how they have always done it. Then someone builds a spreadsheet to bridge the gap, and the spreadsheet becomes a department.

Nobody chose the system as it exists today. It accreted.

That matters, because it changes the solution. If re-keying were just a discipline problem, the answer would be “train people better”. Sometimes that helps. But most re-keying is not a people problem. It is a systems problem. The same piece of information has to move from one place to another, and the business has not given it a pipe, so a person becomes the pipe.

That is why the cost is so easy to miss. The work is not dramatic. It is ten minutes here, twelve minutes there, a copied address, a retyped amount, a job status moved from one tab to another. In isolation, each loop looks too small to fix. In aggregate, it becomes a hidden subscription.

The question is not whether re-keying exists. The question is whether it is worth removing. And that cannot be answered by a marketing statistic.

The claim landscape: five famous numbers, relabelled

Type “cost of manual data entry” into a search engine and every result on the first page is written by a company selling the cure. That is not a complaint. It is a measurement; we ran the searches. Official sources simply do not appear, because official sources have no direct figure for re-keying.

Into that vacuum, five numbers circulate endlessly. Here is where each one actually comes from.

“Knowledge workers spend 2.5 hours a day searching for information.” This traces to an IDC white paper from July 2001 — sponsored, it says on the back page, by Inktomi, an enterprise-search vendor. The paper is honest about what the number is: “We use a general estimate that the typical knowledge worker spends about 2.5 hours per day… searching for information. This number also needs to be adjusted to reflect the circumstances of each specific enterprise.”

It was an assumption, plugged into a costing scenario, in the year Wikipedia launched. IDC’s own later work drifted steadily downwards — “between 15% and 30% of their time” by 2004, and by 2011 an IDC survey figure of 8.8 hours per week, roughly a third lower than the 12.5 hours a week the 2001 assumption implied. The 2.5-hours figure has now outlived the company that sponsored it, circulating as a “finding” it never was.

“Bad data costs the US $3.1 trillion a year.” The conduit is a 2016 Harvard Business Review piece by Thomas Redman — a consultant who sells data-quality work — who attributed the figure to IBM. IBM’s source was an infographic — since deleted; the address now redirects to a general IBM AI page — whose entire sourcing was one undifferentiated list of nine organisations, with no per-figure attribution.

The earliest traceable appearance of $3.1 trillion is a 2011 marketing press release, also without methodology. Redman himself, in a 2023 update, wrote that IBM was “pretty cagey about their methodology, but they stood by the number”, and that his own objective had been “to grab people’s attention with an outrageously big number”. Today, IBM’s only live publication on the cost of bad data uses a different figure entirely — Gartner’s. Even taken at face value, the $3.1 trillion claim is US-only, not UK SME evidence.

“Poor data quality costs organisations an average of $12.9 million a year.” This one has a real, named source: Gartner’s 2020 Magic Quadrant for Data Quality Solutions. The primary text says the figure comes from a survey of reference customers “identified by each vendor” — 154 organisations supplied by the 16 software firms being rated, asked to estimate their own costs.

That is useful context if you sell enterprise data-quality systems. It is much less useful if you run a small UK service business. Enterprise customers of data-quality vendors, self-estimating in 2020, are not a representative sample of all businesses. Gartner has published no update since. The number now circulates as if it were a measured average across all organisations — and in at least one live article as “$12.9 billion”, a thousand-fold mutation that hyperlinks, without apparent embarrassment, to the Gartner page that says million.

“Small businesses spend 120 working days a year on admin.” Real research: a 2017 study by Plum Consulting for Sage, with fieldwork by FTI Consulting across roughly 300 SMEs in each of eleven countries. But 120 days is the cross-country average; the UK-specific finding was 5.6% of staff time.

The mutations started at the launch. Sage’s own CEO foreword says “120 days” and then, four paragraphs later, “120 hours”. One UK finance-news outlet covered the launch the very next day; its page title today reads “Sage research reveals UK SMEs spending 120 hours a year on admin tasks” — wrong country scope and a days-to-hours shrink in a single line. The hours version is still live on a UK small-business site today.

“Technology adoption lifts productivity 7–18% per tool.” The origin is a genuinely good 2018 study by the Enterprise Research Centre — of micro-businesses: firms with one to nine employees, trading three years or more. Cloud computing was associated with 13.5% higher sales per employee, CRM with 18.4%, web accounting software with 11.8%.

By July 2025 that range had migrated into the ministerial foreword of a GOV.UK taskforce report as “firm-level productivity improvements of 7 to 18 per cent per technology” for SMEs in general — credited to the ERC by name, but with no citation or footnote anywhere in the document, and with 5.7 million firms inheriting a finding about the smallest sliver of them.

The pattern across all five is the same, and it is not fabrication. The date falls off first. Then the country. Then the population. Then the hedge — “up to”, “we estimate”, “micro-businesses” — sands away, and a modelling assumption from 2001 walks around in 2026 wearing the clothes of a fact.

The uncomfortable part is that the mutations keep starting inside the publishers’ own documents: Sage’s foreword; the ministerial foreword; a 2023 whitepaper that states the finding correctly on its opening pages — 53% of technology adoption efforts rated unsuccessful — then restates it, pages later in the same document, as “53% of SMEs fail”.

Nobody launders these numbers to us. They arrive pre-laundered.

What the evidence actually says

Strip the genre back to what survives verification and you get a smaller, stranger, more useful picture.

Typing errors are rarer than the horror stories say — and the everyday cost is hours, not blunders. The best evidence on manual entry accuracy is an NIH-funded meta-analysis published in 2025, pooling studies of trained clinical data staff from 1978–2008: single typing passes ran at 0.29% errors per field; double entry halved that to 0.14%. A 1992 clinical trial found double entry caught more errors but took 37% longer — for a difference that was not statistically significant.

So the pantomime-villain version of re-keying — “1–4% error rates!”, usually cited to studies that do not say that, or to nothing — fails. Two honest cautions survive, though. These are trained operators; a distracted office on a Friday is likely worse, and nobody has measured that cleanly. And rare errors carry tail risk: a 2011 experiment found eyeballing your typing, or “visual checking”, produced roughly thirty times the errors of double entry, and that a single wrong cell can flip an analysis.

A wage formula prices the hours. It cannot price the one bad cell that misprices a job.

UK small firms have software; what they do not have is software that talks to itself. The government’s current survey of SME technology — DBT/Ipsos, published July 2025 — finds accountancy software widespread: 47% in the online sample, 72% by telephone. But ERP, the category that exists to make systems share data, sits at 6% and 4%.

The British Chambers of Commerce, in a September 2025 report with Intuit’s sponsorship disclosed, found just 11% of firms use technology “to a great extent” to automate or streamline operations. The closest thing to a measured UK admin benchmark sits adjacent to re-keying rather than on it: the Federation of Small Businesses, surveying 1,436 owners in summer 2024, puts tax compliance at 44 hours and £4,500 a year for the average small firm — a figure that explicitly includes software subscriptions and accountants’ fees, not just time.

Here is the detail that justifies this whole article: the DBT report — the best official evidence we have — never quantifies a single hour or pound saved. It records that 40% of tech-using SMEs say technology saved time. How much? Nobody asked, or nobody could answer.

The hours numbers that do exist are beliefs, honestly labelled as such — until they circulate. ABBYY, with fieldwork by Opinium in November 2020 among 1,000 UK office workers, found people believe — their word — they lose about 1 hour 23 minutes a day to automatable tasks.

Starling Bank, using Opinium again in October 2019 with 1,009 micro-businesses, found the average micro firm puts 15 of its 79 weekly working hours into financial admin — a firm-level figure, promptly garbled by a trade outlet covering it, whose URL still reads “15 hours a day”.

These vendor surveys are not worthless. They are just not measuring the same thing. One is per person; one is per business. One is all automatable tasks; one is financial admin. Neither can tell you what re-keying costs your firm.

And the counter-evidence cuts against easy automation promises, including ours. A 2020 Be the Business/McKinsey survey found 53% of UK SMEs’ technology adoption attempts were rated unsuccessful by the businesses themselves. The Enterprise Research Centre’s October 2025 analysis of nationally representative data concludes technology benefits “are not automatic”: gains are tool-specific, and some combinations associate with lower productivity.

Even HMRC’s record is two-sided. Its 2025 final evaluation of Making Tax Digital for VAT found 41% of the smaller businesses mandated in 2022 felt benefits outweighed costs — but 23% felt the reverse. Its March 2026 impact note prices the income-tax expansion at a net ongoing cost of roughly £104 per person per year, using our division of its £101m across 970,000 people.

Typing into government-approved software is not the same as systems that cooperate. Any honest costing of re-keying has to admit that removing it also costs something, and does not always pay off.

The evidence hierarchy

The mess of statistics above is not an argument against measurement. It is an argument for better measurement.

There are roughly four levels of evidence in this subject.

At the bottom are marketing claims with no usable method. These are the “$3.1 trillion” style numbers: memorable, repeatable, and mostly impossible to apply.

Above those are vendor surveys. Some are perfectly legitimate, but they usually measure beliefs, intentions, or broad categories like “admin” and “automation”. They tell us something about business sentiment. They do not tell one business what to do next.

Above those are official datasets: ONS wages, HMRC admin-cost modelling, DBT adoption surveys, Business Population Estimates. These are more reliable, but they still measure around re-keying, not re-keying itself.

The highest level, for one business, is direct measurement: time the loop, count the occurrences, price the labour, then compare that number with the cost and risk of fixing it.

That is the only level where the decision becomes practical.

What we measured

The measured evidence I can put against all this comes from the most recent of the systems I have built — fifteen years of them, from child-protection casework software to an organisational social network, recruitment analytics, and credit-note automation.

The latest is the one with publishable numbers: two years rebuilding the systems of one UK firm, which stays unnamed here. In that time:

• five separate software subscriptions replaced by one integrated system;
• around 80 automated jobs running every night;
• more than 20 external systems connected;
• over 2,000 screens and endpoints built.

Every one of those is countable in the codebase, which is the standard this article has been applying to everyone else’s numbers.

I will not dress those up as a benchmark. Measured numbers from one deployment are one deployment’s worth of evidence, and the honest literature above says outcomes vary. But the context matters.

This is a firm operating in the “great extent” band that the BCC found contains 11% of the firms it surveyed, in the integration category — ERP-like, systems-sharing-data — that the government survey found at 6% and 4%. And McKinsey’s 2017 activity analysis — a modelled ceiling, US data, but the most careful of its kind — found collecting and processing data to be the most automatable of office activities: 64% and 69% technical potential, against 9% for managing people.

Eighty nightly jobs live precisely there. The re-keying those jobs replaced did not show up in any national statistic. It showed up in payroll.

What it means for your Tuesday

You now know why no article can tell you what re-keying costs you. The honest answer does not exist in any dataset.

But it exists in your office, and it takes about an afternoon to extract. The method is the one HMRC used when it valued digital record-keeping: hours of admin time, multiplied by an ONS median wage, with your own stopwatch supplying the hours.

1. List the loops. Anywhere the same information is typed twice: bank statement lines into the job spreadsheet, emailed instructions into the job system, completed job sheets into invoices, timesheets into payroll.

2. Time one occurrence of each — with a clock. Do not estimate. Every failed number in this article started life as somebody’s estimate.

3. Count a normal week’s occurrences. The diary, sent-items folder, bank feed, job system, and invoice list will usually tell you.

4. Multiply out: minutes ÷ 60 × weekly occurrences × 46 working weeks × the loaded hourly wage of whoever does it.

Loaded wage = the ONS April 2025 median for the role, plus 15% employer National Insurance. Medians: data-entry administrator £14.45/hr, book-keeper or payroll clerk £16.33, office manager £19.20.

A worked example, for a 12-person service firm:

• bank receipts into the job ledger: 25 minutes, four times a week;
• emailed instructions into the job system: 10 minutes, fifteen times a week;
• job sheets into invoices: 12 minutes, ten times a week.

That is 370 minutes — about six hours — a week. At a book-keeper’s loaded wage, it is a little over £5,300 a year. For one member of staff’s loops.

If you would rather not do the arithmetic by hand, the re-keying calculator does the same sum with the same verified wage data — free, no email required.

Two honesty clauses before you act on your number.

First, it prices the hours only — not the invoice that went out wrong, which the error studies say is rare per keystroke but real in consequence.

Second, a priced cost is not a captured saving. The 53%-of-attempts finding and the ERC’s “not automatic” verdict both say the payback depends on fixing one process properly and measuring it, not buying a platform and hoping.

That is why I build one module at a time, and measure each one.

The direction of travel

Two verified facts before the opinion, because the opinion leans on them.

HMRC is extending quarterly digital reporting for income tax down a published timetable — from April 2028 it reaches sole traders and landlords with incomes of £20,000 and above.

And in the British Chambers’ 2025 survey, the share of SMEs actively using AI rose from 25% to 35% in a single year.

Read those however you like; the direction only points one way. The systems around your business — your competitors’, your suppliers’, the tax authority’s — are being re-invented on a schedule, whether yours is or not.

My view is this: it should be stating the obvious that a process needs re-inventing every now and then. Nothing else in a business is expected to run untouched for a decade — not the van, not the wiring, not the price list.

A process left alone does not hold its position, because everything around it moves. It rots in place, and re-keying is what the rot looks like from the inside.

That does not mean every manual loop should be automated. Sometimes the fix costs more than the friction. Sometimes a spreadsheet is the right answer. Sometimes the process is too rare, too messy, or too dependent on judgement to justify building anything around it.

But if your loops resemble the worked example, you are paying something like £5,000 a year for typing that may no longer need a person in the middle — and the £5,000 is only the part payroll can see. The part it cannot see is the distance opening up between how your firm works and how the firms you compete with will.

A decade of laundered statistics made all of this easy to shrug at. When the numbers are junk, the whole subject reads as sales patter.

Your own timed number takes the shrug away.

Sitting on the old system is not always caution. Sometimes it is a subscription to the problem, and the price is no longer just the typing.

If you would rather someone else held the stopwatch: book a free process audit and we will time the loops together.

Sources and method

Every claim above is hyperlinked where it is made; the notes below restore the labels — sponsor, sample, fieldwork dates — and add an archived copy of each source, checked 2 July 2026. Where a live page has died, or where the point is a mutation the owner may later fix, the body links to the archived capture.

• IDC / Feldman & Sherman, “The High Cost of Not Finding Information”, July 2001. Vendor white paper, sponsored by Inktomi; the 2.5 hrs/day is stated in-text as “a general estimate” used in costing scenarios. Mirror (byte-identical to a 2020 Wayback capture).
• Martin White, “Time spent searching: a chronology of a myth”, LinkedIn Pulse, May 2020. Documents IDC’s own revisions (2004: 15–30% of time; 2011 survey figure: 8.8 hrs/week). Archive.
• Thomas C. Redman, “Bad Data Costs the U.S. $3 Trillion Per Year”, Harvard Business Review, 22 Sep 2016 (author is president of Data Quality Solutions, a data-quality consultancy), plus his June 2023 update. Mirror PDF · 2023 repost · the companion “50%” figure self-cites his 2013 HBR piece, where it is “up to 50%” from unnamed studies.
• IBM “Four V’s of Big Data” infographic, c.2013–16 (deleted). Sole sourcing: an undifferentiated nine-organisation list. Archived image.
• Hollis Tibbetts press release, “Dirty Data Costs the US Economy $3.1 Trillion Yearly”, 11 Sep 2011. Earliest traceable instance; no methodology. Archive.
• Gartner, Magic Quadrant for Data Quality Solutions, 27 Jul 2020 (G00389794). “Organizations estimate the average cost… $12.9 million” — survey of 154 reference customers identified by the 16 rated vendors. Licensed-reprint mirror. No newer Gartner figure exists as of July 2026.
• Fluxygen, “Impact of human error rates”, Dec 2023. Carries “$12.9 billion” while linking to Gartner’s million. Archive.
• Plum Consulting for Sage, “Sweating the Small Stuff”, Sept 2017. Fieldwork FTI Consulting, Jul–Aug 2017, ~300 SMEs in each of 11 countries; 120 man-days is the cross-country per-company average; the UK figure is 5.6% of staff time (implied loss £39.9bn). Report PDF · launch release.
• The two mutation carriers. globalbankingandfinance.com, 13 Sep 2017 — current page title: “Sage research reveals UK SMEs spending 120 hours a year on admin tasks” (archive) · smallbusiness.co.uk, “UK small businesses are still wasting time on admin” (Sep 2017, modified Jan 2024, still live) — carries “average of 120 working hours a year on administrative tasks” (archive).
• Enterprise Research Centre, State of Small Business Britain 2018. Micro-business Britain Survey: CATI, Jan–Apr 2018, firms with 1–9 employees established 3+ years; sales-per-employee associations (the report’s own text slides from “linked” to “leads to”). Live PDF · archive.
• DBT, SME Digital Adoption Taskforce final report, 31 Jul 2025. Ministerial foreword states “7 to 18 per cent per technology” for SMEs; the ERC is credited by name in the foreword, but there is no citation or footnote anywhere in the document. Live · PDF archive · HTML archive.
• McKinsey & Co and Be the Business, “The UK’s Technology Moment”, 2020 (Opinium, June 2020, N=1,476 SMEs; 53% = share of N=1,007 adoption efforts self-rated unsuccessful), plus the BtB/Amazon whitepaper, Sep 2023 (restates it as “53% of SMEs” against its own earlier, correct wording). 2020 archive · 2023 archive.
• Garza et al., “Error rates of data processing methods in clinical research…”, Int J Med Inform 195:105749, Mar 2025. NIH-funded systematic review/meta-analysis; 93 manuscripts, underlying studies 1978–2008, trained clinical staff; single entry 0.29% errors per field, double entry 0.14%. Free full text · archive.
• Reynolds-Haertle & McBride, Controlled Clinical Trials 13(6), 1992. 42,278 fields; 22 vs 15 errors per 10,000 (P=.09); double entry +37% time. PubMed · archive.
• Barchard & Pace, Computers in Human Behavior 27(5), 2011. 195 undergraduates; visual checking produced 2958% more errors (≈30×) than double entry; single errors can flip results. Abstract PDF, archived; corroborating APA 2008 poster at barchard.faculty.unlv.edu.
• DBT/Ipsos, “Understanding technology adoption among UK SMEs”, 31 Jul 2025. Online n=2,000 (Nov–Dec 2024) + telephone n=1,001 (Jan–Mar 2025), 1–249 employees excluding sole traders, weighted to BPE 2024; the modes diverge (4% vs 19% non-users), so figures are reported with their survey mode. Live PDF · archive.
• BCC Insights Unit with Intuit, “The Turning Point for SMEs”, Sep 2025. n=1,558 online, 23 Jun–18 Jul 2025; 93% SMEs; sponsorship disclosed in-report. 11% “great extent” automation (full scale 11/42/29/14). AI use 25% (2024) → 35% (2025), base n=1,630 weighted. Live PDF · archive.
• McKinsey Global Institute, “A Future That Works” (In Brief), Jan 2017. Modelled technical automation potential on US BLS activity data: collect data 64%, process data 69%, managing people 9%. Archive.
• ABBYY “COVID-19 Technology and Business Process Survey”, Dec 2020. Opinium, Nov 2020, 4,000 office workers (1,000 UK); self-reported belief, vendor-sponsored. Archived primary.
• Starling Bank, “Make Business Simple”, Jan 2020. Opinium, Oct 2019, n=1,009 UK micro-businesses; firm-level figures. Live report PDF · archive · the “15-hours-a-day” URL garble.
• HMRC, “Making Tax Digital: estimating the wider economic benefit”, 27 Feb 2025. Kantar Public/Verian survey, n=2,300; central estimate 33 hrs/business/yr (95% CI 26–40) valued at ASHE medians; below-threshold central 20 hrs/£382. Live · archive.
• ERC Research Paper 119 (exec summary), Oct 2025. LSBS 2022–23, nationally representative; “benefits are not automatic”; some technology bundles associate with lower productivity. Live PDF · archive.
• HMRC, MTD VAT final evaluation, 27 Feb 2025. 2022 cohort: 41% benefits-outweigh vs 23% costs-outweigh. Live · archive.
• HMRC, MTD Income Tax threshold TIIN, 24 Mar 2026. Mandation threshold reduced from £30,000 to £20,000 from April 2028; £380m transitional cost, £101m/yr continuing, 970,000 people; the ~£104/yr is our division. Live · archive.
• ONS, Annual Survey of Hours and Earnings 2025 (provisional), 23 Oct 2025. 1% PAYE sample, achieved n=174,000; occupation medians from dataset Tables 2.6a/14.6a (all-employees sheets), verified at cell level. Bulletin · Table 2 dataset · Table 14 dataset.
• GOV.UK, “Rates and thresholds for employers 2026 to 2027”. Employer Class 1 NI 15% above £5,000/yr; unchanged from 2025-26. Live · archive.
• DBT, Business Population Estimates 2025, 2 Oct 2025. 5,690,265 UK private sector businesses at the start of 2025 (Table C). Live · archive.
• FSB, “Taking a Toll”, 22 Apr 2025. Survey by Verve, n=1,436 (FSB members + wider self-employed, fieldwork 31 Jul–14 Aug 2024, self-selecting, membership-weighted); the £4,500 includes accountants’ fees and software subscriptions, not just time; the ~£25bn/242m-hour aggregates are gross-ups of the survey mean over the business population. Live report page · archive.