Curated research

Manual data entry in finance: what the research actually says

Everything citable about what manual finance work costs — hours, euros, errors, close days and people. Each figure below carries its publisher, its year and a link to the document it came from, so you can check it rather than take our word for it.

Last updated:

Sources cited: 18

How much time do finance teams spend on manual data entry?

This is the hardest number in the whole field to source well. Plenty of vendors quote a tidy hours-per-week figure with no study behind it. Here is what can actually be traced to a published survey — and where the honest gaps are.

What we take from this

Notice what is missing. There is no well-sourced, finance-specific figure for hours per week per accountant on rekeying, and we are not going to invent one. The numbers that circulate — eight hours a week, ten hours a week — trace back to vendor landing pages and LinkedIn posts citing surveys nobody can produce.

The defensible way to size your own problem is APQC's productivity median. If your AP function processes materially fewer than 12,000 invoices per person per year, the gap is time going somewhere, and rekeying is the usual place. That comparison uses your own numbers rather than a borrowed average, which is why our calculator is built around it.

What does it cost to process an invoice, manually and automated?

This is the best-measured part of the whole subject. Ardent Partners has benchmarked accounts payable annually for two decades and defines Best-in-Class as the 20% of enterprises with the lowest processing costs and shortest cycle times.

Cost to process one supplier invoice

Ardent Partners, The State of ePayables 2025. Best-in-Class is the 20% of enterprises with the lowest processing costs and shortest cycle times.

Best-in-Class
$2.65
Average
$9.84
All others
$12.42

What we take from this

The single most useful thing in this section is that the gap is not a technology gap. Best-in-Class organisations run the same ERPs as everyone else. What separates them is the share of invoices that never reach a human: 51% against 29%. Cost per invoice is downstream of that one number.

The benchmarks are published in US dollars and we quote them unconverted, because invoice processing cost is overwhelmingly labour and a currency conversion would imply a precision the benchmark does not have. Use them as a shape, not a price list, and put your own numbers through the calculator.

How often does manual entry go wrong, and what do the errors cost?

Two different questions get muddled here: how often a person mistypes a field, and how often an invoice ends up needing human rework. The second is far better measured than the first, and matters more.

What we take from this

Be sceptical of the keying-error statistics in circulation. The often-quoted "1% of manually entered data is wrong" and Gartner's "$12.9 million a year in poor data quality" are repeated everywhere and sourced almost nowhere — we could not put either in front of a primary document, so neither is stated as fact above.

The exception rate is the number to manage. An 18.4% exception rate on 50,000 invoices is 9,200 invoices a year that stop and wait for a person. That is where the cost, the cycle time and the supplier phone calls all come from, and it is measurable inside your own ledger this week.

The fraud figures matter for a reason people miss: a billing scheme works precisely because a human keys the invoice and a human approves it. Controls that check every invoice against a purchase order and a receipt, on every invoice rather than on a sample, are the same controls that make processing cheaper.

How long does the month-end close take, and why is it not getting shorter?

The close is where manual work becomes visible to everyone outside finance, because it is the one deadline the rest of the business notices. The benchmarks have barely moved in six years.

Closing the monthly books within six business days

Ventana Research / ISG. The 2019 figure is Office of Finance Benchmark Research; the 2023 figure is Dynamic Insights: The Smart Financial Close. Ventana describes the change as statistically insignificant.

2019
60%
2023
59%

What we take from this

The interesting finding is not the median. It is that automation demonstrably shortens the close — 88% against 40% in Ventana's own numbers — and yet only 11% of organisations use workflow extensively and only 31% automate most reconciliations. The technique works and almost nobody applies it.

That gap is the honest reason closes are not getting faster. It is not that the tools do not work. It is that reconciliation and journal preparation are exactly the tasks that get postponed in favour of the deadline in front of you, every month, forever.

How much of finance still runs on spreadsheets, and how wrong are they?

The shadow ERP. Organisations buy the system, and the real work carries on in a workbook beside it. The most striking evidence comes from teams that own dedicated planning software and use spreadsheets anyway.

How FP&A teams actually work

2025 AFP FP&A Benchmarking Survey Report: Technology and Data. The bottom bar is the share using dedicated planning software; the top three are spreadsheet use by the same population.

Spreadsheets, at least quarterly
100%
Spreadsheets for planning
96%
Spreadsheets for reporting, daily or weekly
93%
EPM software for planning, at least quarterly
71%

What we take from this

The AFP finding is the one to remember, because it kills the usual explanation. These are not teams who could not afford a system. Seven in ten own dedicated planning software and every one of them is still in a spreadsheet at least quarterly. The workbook is not a gap in the tooling; it is where the awkward last mile lives — the reclassification, the accrual nobody automated, the report the system will not produce in the shape the board wants.

Panko's audits are the honest version of the spreadsheet-error statistic. You will see "94% of spreadsheets contain errors" quoted constantly; the properly weighted figure across studies with published methods is 84% of 163 spreadsheets, and it is more damning than the round number, because the better-designed audits found more.

Public Health England is worth citing precisely because it was not incompetence. It was a well-run organisation, a routine file format, and a limit that failed silently. That is the actual risk profile of a spreadsheet in a finance process: not that someone is careless, but that nothing tells you when it breaks.

What is manual work doing to finance teams and the talent pipeline?

The cost that does not appear on any benchmark: the people. Two things are happening at once — the work is exhausting the people doing it, and fewer people are arriving to replace them.

US accounting degrees awarded, annual change

AICPA Trends reports. Bachelor's and master's degrees combined; 2023–24 is the most recent year reported, at 55,152 degrees.

2021–22
−7.4%
2022–23
−9.6%
2023–24
−6.6%

What we take from this

Put the two halves together and the arithmetic is unforgiving. The US needs roughly 124,000 accountants a year and is producing about 55,000 accounting graduates, of whom only some enter the profession. Whatever manual work is not automated in the next few years will not simply be done more slowly. In many teams it will not be done at all.

This is also the strongest argument against the framing that automation is about headcount reduction. In a profession that cannot fill the roles it already has, taking the rekeying away is not how you employ fewer accountants. It is how you keep the ones you have.

How this page was put together

Every figure was read out of the publisher's own document — the report, the benchmark record, the press release, the paper — and not out of an article quoting it. Each line carries the publisher, the edition and a link, so you can check any of it in a click.

Statistics are tagged by where they come from. Independent research, analyst benchmarking, academic work and official statistics are all stronger evidence than a vendor-commissioned survey. Vendor research is included where the sample is large and the method is disclosed, but it is labelled so you can weigh it accordingly.

Where a widely-repeated statistic could not be traced to a primary source, it was left out and the gap noted rather than papered over. That applies to the popular claims about keying error rates, the cost of poor data quality, and hours per week per accountant on manual entry. If you have the underlying study for any of those, we would genuinely like to see it.

Reviewed annually. Ardent Partners, APQC and the AICPA all publish on a yearly or biennial cycle, so the figures here are re-checked against the newest editions each August.

Questions people ask

How much does it cost to process an invoice manually?

Ardent Partners' State of ePayables 2025 puts the average all-inclusive cost at $9.84 per invoice. Organisations Ardent classes as Best-in-Class — the 20% with the lowest costs and shortest cycle times — process the same invoice for $2.65, while everyone else averages $12.42. The gap tracks almost entirely with how many invoices are processed straight through without a person touching them: 51% for the leaders against 29% for everyone else.

How many hours a week do finance teams spend on manual data entry?

There is no well-sourced, finance-specific figure for this, and the hours-per-week numbers in circulation generally trace back to vendor marketing rather than research. The best-evidenced adjacent figure is Intuit's 2024 Business Solutions Survey, in which businesses reported spending 25 hours a week on manual data entry and reconciling data between applications — but that is whole-business time, not one accountant's. A more defensible way to size your own problem is APQC's benchmark of 12,000 invoices processed per accounts payable person per year: if your team is well below that, the difference is time going somewhere.

What is a normal invoice exception rate?

Ardent Partners puts the 2025 average at 18.4% — nearly one invoice in five stops and needs a human to resolve it. Best-in-Class organisations run at 11.1% against 20.9% for everyone else. Ardent identifies exceptions as the single biggest reason the other accounts payable benchmarks are not better, because an exception drives cost, cycle time and the supplier phone calls that consume 21.9% of AP staff time.

How long should a month-end close take?

APQC's cross-industry median for completing the monthly financial close is 8.0 days, from a sample of 3,123 organisations. The narrower measure — trial balance to consolidated financial statements — has a median of 6.0 days. Ventana Research found 59% of organisations closing within six business days in 2023, statistically unchanged from 60% in 2019.

What percentage of spreadsheets contain errors?

Across nine spreadsheet audit studies with published methodologies, covering 163 real operational spreadsheets, errors were found in 84% by weighted average, and only one of the nine studies found errors in fewer than 86%. That figure comes from Raymond Panko's long-running review of the audit literature at the University of Hawaii. The widely-quoted "94%" is a looser summary of the same body of work; the 84% figure is the one with the methodology attached.

Do finance teams still use spreadsheets if they have an ERP or planning system?

Yes, almost universally. The 2025 AFP FP&A Benchmarking Survey found 96% of FP&A professionals use spreadsheets for planning and 93% use them for reporting daily or weekly, with every respondent using them at least quarterly — while 71% of the same group also use dedicated enterprise performance management software. Owning the system does not replace the workbook.

Is there really an accountant shortage?

In the US, yes, on the numbers. The AICPA's 2025 Trends report counted 55,152 accounting degrees awarded in 2023–24, down 6.6% after falls of 9.6% and 7.4% in the preceding two years, while the Bureau of Labor Statistics projects about 124,200 annual openings for accountants and auditors through 2034. Undergraduate accounting enrolment has now risen for three consecutive years, which should ease the graduate numbers later this decade, but not before the shortfall is felt.

Where do these statistics come from, and how often are they updated?

Every figure on this page was read from the publisher's own document and carries the publisher's name, the edition year and a link to the original. Where a widely-repeated statistic could not be traced to a primary source — including the popular claims about keying error rates and the cost of poor data quality — it is either left out or flagged as unverified rather than quoted. The page is reviewed annually against the newest editions of the underlying research.

Want to know which of these numbers is yours?

Thirty minutes with a Chartered Accountant who builds this software for a living. We will work out your cost per invoice, your exception rate and your close duration from your own systems, and tell you honestly whether automation is worth it.