Data entry error statistics: how often people key data wrong, and what it costs

The best-measured human error rate in data entry is about 1%. In a controlled study, people keying data once got 0.95% of values wrong, and typists in a study of 136 million keystrokes left 1.17% of characters wrong. Double entry cuts the rate to about 0.03%, while copying from medical records runs far higher, at 6.57% across 93 clinical studies.

Every figure below links to the study or survey behind it. We also list the popular numbers we couldn't trace to any source. Last checked September 2026.

How often people key data wrong#

The cleanest evidence comes from controlled studies where researchers know the right answer for every value. In Barchard and Pace's 2011 study, 195 people each keyed 1,260 values.

• 29.58 times more errors

  Visual checking left 2,958% more errors than double entry and wasn't significantly better than keying once. Barchard & Pace, 2011

• 5.5% keyed everything right

  Only 5.5% of people keying once got all 1,260 values right, against 17.1% who checked by eye and 77.4% who used double entry. Barchard & Pace, 2011

• 33% longer for double entry

  It took 49.73 minutes on average, against 37.43 for visual checking and 30.03 for keying once. Barchard & Pace, 2011

• 0.06% of errors looked wrong

  Only 0.06% of the errors were blank or outside the allowed range, so range checks would have caught almost none. Barchard & Pace, 2011

A 2020 follow-up with 412 people reached the same verdict. Double entry was significantly and substantially more accurate than the alternatives. The typing study adds a detail worth knowing, which is that 6.31% of all keypresses were corrections. People fix a lot as they go and still leave about one character in a hundred wrong.

Our take. A single error rate hides where the errors land. Measure accuracy on your own documents by counting the fields a person had to fix, and count separately for the fields that matter, such as amounts, dates and account numbers.

Error rates in healthcare and clinical data#

Clinical research has measured data entry errors for decades, usually as errors per 10,000 fields. A 2025 meta-analysis pooled 93 papers published from 1978 to 2008, with rates ranging from 2 to 2,784 errors per 10,000 fields.

• 976 vs 14 errors per 10,000 fields

  Published audits from source records to the database averaged 976, against 14 when keying from a finished form. Most errors enter when someone copies from the source. Nahm et al., 2008

• 2.3% to 26.9% error rates

  That's the range double entry uncovered across clinical research databases at one hospital. Goldberg et al., 2008

• 2.8% of pathology fields wrong

  Hand-entered data for 421 patients had a 2.8% error rate, and only 76% of patients' records were fully accurate. Hong et al., 2013

• 3.7% of glucose results mismatched

  In 260 of 6,930 manually entered point-of-care results the value didn't match the instrument. About 5 per 1,000 were clinically significant. Mays & Mathias, 2019

• 7.4 errors per 100 words

  in speech-recognition drafts of clinical notes, falling to 0.3% once signed. 96.3% of the drafts had errors. Zhou et al., 2018

• 0.370 vs 0.046 errors per 1,000 fields

  Single keying against double keying on patient questionnaires. Automated reading of check boxes matched double keying, while handwriting recognition ran at 6.734. Paulsen et al., 2012

Spreadsheet errors in finance and operations#

Much manually keyed data ends up in spreadsheets, and spreadsheets have their own well-studied error problem. Ray Panko's 2015 review counted errors in 94% of the spreadsheets examined in 85 intensive inspections. Across 14 lab studies with 967 people, the average cell error rate was 3.9%.

Field data tells a similar story. Powell, Baker and Lawson audited 50 working business spreadsheets with 270,722 formulas and found errors in 0.9% to 1.8% of formula cells. That sounds small until you see that 43 of the 50 workbooks, 86%, had at least one error that produced a wrong result, and one had a cell error rate of 28%.

People also underrate their own mistakes. In an experiment Panko reports, spreadsheet developers put the chance they had made an error at 10% (median). In fact, 86% had.

What bad data costs#

• $12.9 million a year

  The average cost of poor data quality to an organization, according to Gartner research from 2020. Gartner

• $3.1 trillion a year

  IBM's estimate of what poor-quality data cost the US economy in 2016. Redman, Harvard Business Review

• 15% to 25% of revenue

  The cost of bad data for most companies, by Thomas Redman's estimate, built on Experian research and two consultants' work. MIT Sloan Management Review, 2017

• 47% of new records

  had at least one critical error across 75 data quality assessments, and only 3% of the scores met the 97% bar for acceptable. Nagle, Redman & Sammon, 2020

• 10% to 25% of B2B records

  carry critical data errors, from well-run marketing databases to typical ones. SiriusDecisions

• 92.0% of invoices error-free

  The median share of invoices processed error-free the first time, across 2,342 organizations, so about 8% need rework. APQC

Where the 1-10-100 rule comes from

The 1-10-100 rule says an error costs $1 to prevent at entry, $10 to find and correct later, and $100 if nobody fixes it. It comes from Labovitz, Chang and Rosansky's 1992 book Making Quality Work, where it described quality costs in general. SiriusDecisions later applied it to records ($1 to verify a record as it's entered, $10 to clean it, $100 if nothing is done). Treat it as a rule of thumb about when to catch errors, not a measured cost.

The calculator below counts only the errors someone finds and fixes, which is the cheap end of that rule.

For the wider business case, see how to measure IDP ROI and accounts payable automation.

How much teams trust their data#

• 67 data incidents a month

  on average in 2023, up from 59 in 2022. Resolving one took 15 hours on average. Monte Carlo and Wakefield Research, 200 data professionals

• 31% of revenue affected

  by data quality issues, on average, up from 26%. In 74% of cases, business stakeholders found the problem first all or most of the time. Monte Carlo, 2023

• About a third of customer data

  is inaccurate in some way, by organizations' own estimate. Experian, 700 business leaders, 2021

• 55% lack trust

  in their data assets, and 95% have seen impacts from poor data quality. Experian, 2021

Data entry jobs: BLS numbers#

• 127,080 data entry keyers

  were employed in the US in May 2025. BLS OEWS

• $19.88 an hour median pay

  ($41,340 a year). The mean is $20.82 an hour ($43,310 a year). BLS OEWS, May 2025

• 25.5% fewer jobs by 2035

  BLS projects 131,800 jobs in 2025 falling to 98,200 in 2035. BLS Employment Projections

• 7,700 openings a year

  are still projected through 2035, with a high school diploma the typical entry requirement. BLS Employment Projections

Popular statistics we couldn't trace#

These figures appear across many articles on data entry errors. We looked for the original study behind each one and didn't find it, or found that it says something else.

Where automation changes the numbers#

The studies agree on two things. Keying from a source document is where most errors start, and checking by eye barely helps. Double entry works but doubles the keying.

Document AI removes the keying. It reads the page, extracts each field with a confidence score, and sends only low-confidence fields to a person, with the source line highlighted. Docsumo reports 99% field-level accuracy on 250+ document types and 95%+ straight-through processing. That applies to invoices, bank statements and the other documents teams still retype. To compare tools, see the best data entry software and how OCR accuracy is measured.

Citing these numbers#

You're welcome to use any figure here. Link to the original study, and to this page if it saved you the search. Quote the population and year with the number, since a 0.29% rate for clinical trial data and a 6.57% rate for chart abstraction describe very different work.

If your team keys data from documents today, book a demo with a few of your own files, or start a free trial.

Frequently asked questions#