
Nobody has measured what re-keying costs an Australian business. Here are the inputs that are published.
No Australian government source measures the cost of manual data entry. What is published is enough to do the sum: the ATO’s per-invoice processing cost, ABS and award wage rates from 2026, controlled error-rate studies, and how few businesses have moved off paper and PDF. Multiply your own document count by those, not by a vendor’s average.
Every figure on this page, in one place. None is ours.
- Paper invoice, processing cost
- $30.87
- PDF invoice, processing cost
- $27.67
- Peppol eInvoice, processing cost
- $9.18
- Full-time adult ordinary earnings
- $2,083.70 a week, May 2026
- Clerks Award, Level 1 first year
- $26.97 an hour from 1 July 2026
- Super guarantee
- 12%
- Businesses registered for eInvoicing
- about 410,000, January 2025
- Actively trading businesses
- 2,814,778, June 2026
Sources and dates are in the body text beside each figure. The Deloitte estimates date from 2016; the ATO says so and we repeat it.
Start with the absence, because it shapes everything else
There is no Australian government statistic for what manual data entry costs a business. Not hours, not dollars, not error rates. We looked at the ABS, the ATO, Treasury, the Productivity Commission, the Ombudsman and business.gov.au. The nearest thing is the per-invoice cost the ATO publishes to make the case for eInvoicing, and that rests on consulting work from 2016.
That matters because the figure gets quoted constantly, and every version traces back to somebody selling the fix. The "1% error rate" for manual entry has no primary source we could find. The per-invoice dollar figures in circulation come from US surveys of large accounts-payable departments. None of it is checkable against an Australian tradie’s office.
So this page does something narrower. It publishes the inputs that do exist — what an invoice costs to handle, what an hour costs in wages, how often typed data is wrong, how much of Australia still works from paper — and gives you the arithmetic. Admin hours by trade found the same gap for hours; this is the dollars side.
The one Australian government figure: what an invoice costs to process
The ATO puts the cost of handling a paper invoice at $30.87, a PDF invoice at $27.67 and a Peppol eInvoice at $9.18. Of that cost, 40% sits with the business sending the invoice and 60% with the business receiving and paying it. ATO, Peppol eInvoicing value assessment — cost calculations, page updated 17 October 2025. The ATO is explicit that the averages come from Deloitte Access Economics and that "these DAE estimates date back to 2016", and says they align with 2024 APEC and European Commission studies. Treat them as decade-old estimates that two later studies did not contradict.
More useful than the dollars are the minutes underneath them, because you can check those against your own desk. On the sales side the ATO’s model allows 6 minutes to create an invoice, 3 to send it and 1 to reconcile it. On the purchase side: 7 minutes to receive it, 2 to validate, 7 to review and 5 to approve — 21 minutes per supplier bill before anything goes wrong. ATO, Peppol eInvoicing value assessment — cost calculations, page updated 17 October 2025.
Then things go wrong. The same model puts sales-invoice exceptions at 4 to 16 minutes each, arising on 10% to 48% of invoices depending on the step, and purchase-invoice exceptions at 5 to 20 minutes on 3.6% to 48% of invoices. ATO, Peppol eInvoicing value assessment — cost calculations, page updated 17 October 2025. The width of those ranges is the honest part: a supplier who sends clean PDFs costs a fraction of one who photographs a docket.
- KNDR·0110 minutesSales invoice: 6 to create, 3 to send, 1 to reconcile. Before exceptions.
- KNDR·0221 minutesPurchase invoice: 7 to receive, 2 to validate, 7 to review, 5 to approve. Before exceptions.
- KNDR·034–20 minutes morePer exception, on anywhere from 3.6% to 48% of invoices. The range is the finding.
What an hour of re-keying costs in wages, on 2026 rates
Full-time adult ordinary time earnings were $2,083.70 a week in May 2026, up 3.7% on the year — about $54.83 an hour over a 38-hour week. ABS, Average Weekly Earnings, Australia, May 2026, released 13 August 2026. That is the economy-wide average and it is higher than most people pay for admin, so it is a ceiling rather than a default.
The floor is the award. From 1 July 2026 modern award rates rose 4.75% and the national minimum wage is $26.44 an hour, or $1,004.90 a week. Fair Work Commission, Annual Wage Review 2026 decision, operative 1 July 2026. Under the Clerks — Private Sector Award, a Level 1 employee in their first year is on $26.97 an hour and $1,024.70 a week; Level 2 is $30.00 and Level 3 is $31.11, with casuals loaded 25%. Clerks — Private Sector Award wage table, 1 July 2026, reproduced from the Fair Work pay guide, June 2026.
Neither figure is what an hour costs you. Add the super guarantee at 12% (ATO, Super guarantee, rate from 1 July 2025), then leave, public holidays, workers compensation and the desk. A Level 1 clerk at $26.97 is $30.21 an hour with super alone; the ABS average is $61.41 on the same basis. Leave and on-costs sit on top of both, and the arithmetic section below uses whatever you enter rather than either.
Stating both ends matters because the cost of re-keying is often not the clerk’s rate at all. It is the owner on a Sunday night, and that hour is worth what it would have earned quoting.
How often typed data is wrong, from the only controlled studies we could find
The controlled evidence on human data-entry error comes from psychology labs, where researchers needed to know how much to trust their own data, not from accounts departments. In a 2011 study, 195 people were randomly assigned to enter data by single entry, by single entry with visual checking, or by double entry. Visual checking produced 2,958% more errors than double entry and was not significantly better than no checking at all. Barchard and Pace, Computers in Human Behavior 27(5), September 2011. Two thirds of the visual checkers ended up with a wrong statistical result downstream.
A follow-up with 412 participants in 2020 compared four checking methods. People using double entry left 0.64 errors on average and 83% produced a perfect dataset; visual checking left 1.66 errors and 35% were perfect; reading aloud to a partner left 3.27 errors and 21% were perfect. Overall accuracy ranged from 99.91% for the best method to 99.48% for the worst, and people with prior data-entry experience made 41% fewer errors. Barchard et al., Behavior Research Methods 52, published 2020.
Two things follow. "99.5% accurate" sounds fine until you multiply it across the eight fields on a supplier bill, which gives a clean document about 96% of the time — one bill in twenty-five wrong somewhere. And the fix that worked in the lab, typing everything twice, is the one nobody in a trade business will do, which is why the exceptions in the ATO model exist.
The caveat: these are students entering numbers from a sheet, not a bookkeeper under time pressure. We would expect the real rate to be worse and have no Australian measurement of how much. The extraction accuracy page puts the machine equivalent beside these — and finds the vendors publish less than the psychologists did.
How much of Australia still works from paper and PDF
About 410,000 Australian businesses were registered to receive eInvoices on the Peppol network at January 2025, and 129 government entities. Accounting Times, quoting the ATO, 23 January 2025. Against 2,814,778 actively trading businesses at 30 June 2026 (ABS, Counts of Australian Businesses, 30 June 2026, released 18 August 2026), that is roughly one business in seven with the plumbing to receive a bill as structured data. A separate report put it at 16% of small and medium businesses, with about 1.5 million eInvoices sent in total. The Mandarin, 8 March 2025.
Two years earlier Xero told Treasury that Australian adoption was "around 1 percent", against 21% in Singapore and 15 to 40% across the EU, and estimated eInvoicing pays "up to seven days faster" — a vendor estimate, quoted as one. Xero submission to the Treasury review of the Payment Times Reporting Act, 1 March 2023.
The broader ABS picture is the same. 12% of businesses reported using AI in 2024–25; 32% received orders online, barely changed from 2021–22; 10% actively collected or analysed data; 35% reported barriers to innovation, led by lack of funds at 15% and skills at 11%. ABS, Characteristics of Australian Business, 2024–25, released 25 June 2026. For roughly six businesses in seven, every supplier bill still arrives as a picture of a bill, and somebody types it.
How to work out your own figure in an afternoon
Since nobody publishes it, count it. Crude and yours beats precise and somebody else’s.
Pick one ordinary week. Count the documents typed from one place into another: supplier bills into the accounting file, quotes retyped as invoices, timesheets into payroll, job sheets into the job system. Skip the ones that flowed automatically. For each type note the minutes — start from the ATO’s 10 and 21 above and correct against your stopwatch — and how many needed a call, a resend or a fix.
Then price it. Documents a week × minutes ÷ 60 × a loaded hourly rate × 50 weeks. Use your own rate; the ABS and award figures above bracket it. The leak calculator does this with ranges rather than a point, and its defaults are the ABS earnings figure and the ATO super rate, not a Kindra benchmark. Add the days: an invoice that exists on Tuesday rather than Friday is paid earlier, and Australian small businesses already wait an average of 22.9 days to be paid. Xero Small Business Insights, Australia, June quarter 2026, published 30 July 2026. The DSO page puts a cost on those days.
Finally, sort the documents into three piles: only you can handle it, anyone could, and it exists only because two systems do not talk. The third pile is the automation candidate, usually supplier bills and the quote-to-invoice step. Accounts payable and quote to invoice describe what that looks like built.
- KNDR·01Count, one ordinary weekDocuments typed from one place into another, by type. Not the ones that flowed.
- KNDR·02Time it, then correct the ATO minutes10 minutes a sales invoice, 21 a purchase invoice is the published starting point. Yours will differ.
- KNDR·03Price it, loadedWages plus 12% super plus on-costs, times 50 weeks. Or the owner’s quoting hour, if that is who types.
- KNDR·04Sort into three pilesOnly you. Anyone. Nobody, because two systems do not talk. Automate the third pile first.
What we have not measured, and what would change this page
Nothing above is a Kindra measurement. This page was specced as one — first-party data on what re-keying costs by business size and trade, licensed for anyone to reuse with attribution. We have not collected it, so the page publishes what is published instead, with the dates on.
What would change it is not a survey. It is a count of documents and minutes taken from the systems themselves — timestamps on when a bill arrived and when it was posted, before and after the join is built, on real Australian businesses, for long enough that the numbers stop moving. When that exists it gets published here with the sample size, the method and the cases where it failed, and the borrowed figures above come out.
If our own numbers turn out smaller than the ones vendors quote, they still get published. That is the point of measuring instead of asserting. Until then, the arithmetic section above runs on your figures, which are worth more than anybody’s average.
The questions people actually ask.
01Why is there no Kindra figure on a page filed under research?
Because we have not measured one, and a number you cannot show the working for is a marketing claim with a chart. What we could do honestly was find everything that is published, date it, put the sources that disagree next to each other and mark where the data stops. That turned out to be more useful than the page we set out to write.
02Everyone quotes a cost per invoice. Why not just use that?
We did, and we said where it came from. The ATO’s $30.87 and $9.18 are the only Australian government figures of their kind, and the ATO flags that they rest on 2016 Deloitte estimates. The other numbers in circulation come from surveys of large US accounts-payable departments, are usually undated, and describe a business with an AP team. We would rather you ran your own count.
03If automation replaces typing, doesn’t it just replace one error rate with another?
Yes, and nobody should tell you otherwise. Machine extraction has a field-level error rate too, and the vendors are less forthcoming about theirs than the psychologists were about human ones. The difference is where the errors land: a well-built system routes uncertain fields to a person and posts the rest, so the effort goes on exceptions rather than the whole pile. The extraction accuracy page covers how to test that on your own bills.

Stop bending.
Start shipping.
Bring the one flow that keeps costing you a Tuesday afternoon. We map it live and tell you what we'd build, what we wouldn't, and what it costs.