spreadsheet error statistics

Spreadsheet Errors and Risks: Statistics and Notable Failures (2026)

Last updated: October 2026

Spreadsheets sit underneath most of the world's financial decisions. Decades of academic audits and a string of public failures show how often they go wrong, and how much it can cost. Below is what the research and the record actually say, with the primary source for every figure.

How common are spreadsheet errors

94% of audited operational spreadsheets contained at least one error. The figure is Panko's weighted average across the seven field audits conducted from 1995 onward, the cut-off he applies on the grounds that auditing procedures improved after that date. Eighty-eight spreadsheets are represented in that sample. It is reproduced here from an independent peer-reviewed review of the literature rather than from Panko's own summary. (Powell, Baker & Lawson, "A critical review of the literature on spreadsheet errors", Decision Support Systems 46(1), 2008, §4.4)

94% again, on a separate and larger sample of intensive inspections. Panko's 2015 review tabulates 85 operational spreadsheets examined through inspections taking several days each, most performed by commercial spreadsheet auditing firms, one by a government tax unit using cell-by-cell inspection. The weighted average share containing errors is again 94%. Two independent compilations reaching the same figure on different samples is the strongest support this number has. (Panko, "What We Don't Know About Spreadsheet Errors Today", EuSpRIG 2015, arXiv:1602.02601)

5.2% average cell error rate — the most-cited figure on this page, and the weakest. Same Panko compilation, but cell error rates were available for only 43 of the 88 spreadsheets. Powell, Baker and Lawson publish three reasons to distrust it: three of the seven sources are unpublished, most gave little or no information on how they defined or found errors, and the single study accounting for 70% of the sample used to compute it did not actually report a cell error rate. Cite the 94%; treat the 5.2% as an order of magnitude, not a measurement. (Powell, Baker & Lawson, 2008, §4.4)

2.2% and 2.5%: the cell error rates actually measured in operational spreadsheets. In Panko's 2015 table of 85 inspected spreadsheets, only two studies reported a cell error rate at all, and both landed near 2%. For comparison, 14 laboratory development studies covering 967 participants average 3.9%. These are the defensible numbers when a per-cell rate is needed. (Panko, EuSpRIG 2015, Tables 1 and 2)

20,277,835 formula cells across 9,120 Enron spreadsheets — why a low per-cell rate still matters. Hermans and Murphy-Hill analysed the spreadsheet corpus released through Enron's legal discovery. Panko uses scale figures like this to make the compounding argument: at cell error rates of 1% to 5%, he writes, "the probability of a calculation error is overwhelming", and even rates an order of magnitude lower would give error probabilities "that any company would view as unacceptable". (Hermans & Murphy-Hill, 2015, cited in Panko, EuSpRIG 2015)

How much finance still relies on spreadsheets

96% of FP&A professionals use spreadsheets for planning. The 2025 AFP FP&A Benchmarking Survey reports "96% of survey respondents using them for planning and 93% using them for reporting purposes on a daily or weekly basis", despite years of investment in dedicated FP&A platforms. Based on 362 FP&A and finance practitioners, fielded autumn 2024. AFP's phrasing leaves it ambiguous whether the daily-or-weekly qualifier applies to both figures or only to reporting; it is reproduced here as published rather than resolved. (Association for Financial Professionals, 2025 FP&A Benchmarking Survey)

58% of finance leaders still rely on spreadsheets as their primary automation tool. Reported in Rossum's Document Automation Trends 2025, even as AI and dedicated tools become available. (Rossum, Document Automation Trends 2025, as reported by The CFO)

67% of UK CFOs say their teams use Excel to analyse and justify accounts during the close, as many as use an ERP system (67%). Multiple answers allowed: 61% also cite collaborative tools (Teams, email), 38% custom internal tools, and 3% a dedicated account reconciliation solution. Only 15% analyse and reconcile accounts without relying on Excel or collaborative tools. Odoxa survey for Sixthfin, a vendor of account reconciliation software; 303 CFOs of UK private companies with 250+ employees, all involved in the monthly or periodic close; quota sample by sector, size and region, weighted by company size; online, 11-23 March 2026; published 18 May 2026. (Odoxa for Sixthfin, Accounting Issues Observatory: United Kingdom, May 2026)

What spreadsheet errors have cost

$6.2bn: JPMorgan's "London Whale" loss, with a spreadsheet flaw among the contributing operational failures (2012). The bank's own management task force found the Value-at-Risk model ran through a chain of spreadsheets requiring manual copy-paste between them, and that one step divided by a sum instead of an average, understating risk. The Excel error was one of several operational flaws behind the loss, not its sole cause. (JPMorgan Chase Management Task Force Report, 16 January 2013, archived by the Yale Program on Financial Stability)

−0.1% became +2.2%: the Reinhart-Rogoff spreadsheet error that undercut a pillar of austerity policy (2010). The influential paper "Growth in a Time of Debt" reported that average growth turns negative (−0.1%) above a 90% debt-to-GDP ratio. Economists Herndon, Ash and Pollin found that a formula not extended across five rows excluded five countries from the average. Corrected, average growth above the threshold was +2.2%, not negative, weakening the empirical case the figure had been used to support. (Herndon, Ash & Pollin, 2013, via The Conversation)

15,841 COVID-19 cases dropped from England's contact tracing because of a spreadsheet row limit (2020). Public Health England pulled testing data into a legacy .XLS file, whose 65,536-row cap silently truncated results once exceeded. Nearly 16,000 positive cases were not passed to contact tracers during the affected window. (The Register, October 2020)

Sources

  • Stephen G. Powell, Kenneth R. Baker & Barry Lawson, "A critical review of the literature on spreadsheet errors", Decision Support Systems 46(1), 2008, 128-138. Open copy hosted by the Tuck School of Business, Dartmouth. Link
  • Raymond R. Panko, "What We Don't Know About Spreadsheet Errors Today: The Facts, Why We Don't Believe Them, and What We Need to Do", Proceedings of the EuSpRIG 2015 Conference. arXiv:1602.02601. Link
  • Association for Financial Professionals, 2025 FP&A Benchmarking Survey. Link
  • Rossum, Document Automation Trends 2025 (reported by The CFO). Link
  • Odoxa for Sixthfin, Accounting Issues Observatory: United Kingdom, results report (18 May 2026; 303 UK CFOs, fielded 11-23 March 2026). Link
  • JPMorgan Chase, Report of the Management Task Force Regarding 2012 CIO Losses (16 January 2013). The bank's own copy is no longer online; the report is archived by the Yale Program on Financial Stability. Link
  • Herndon, Ash & Pollin (2013), on the Reinhart-Rogoff error (via The Conversation). Link
  • The Register, "Excel spreadsheet blunder ... 16,000 COVID-19 cases" (2020). Link

Changelog

  • 2026-10: Added a close-specific reliance figure: 67% of UK CFOs use Excel to analyse and justify accounts, on par with ERP, and only 15% work without Excel or collaborative tools (Odoxa for Sixthfin, n = 303, March 2026). The survey's sponsor sells reconciliation software, which is stated inline.
  • 2026-09 (second pass): Re-sourced the prevalence section and corrected its scope. Panko's own site (panko.shidler.hawaii.edu) is no longer reachable — the whole host is down, not just the page — and it carried all three citations in this section. The 94% and 5.2% are now sourced to Powell, Baker and Lawson's peer-reviewed review of the literature (Decision Support Systems, 2008), which reproduces Panko's compilation and states its scope precisely: seven field audits from 1995 onward, 88 spreadsheets, cell error rates available for only 43 of them. A second, independent confirmation of the 94% was added from Panko's 2015 EuSpRIG paper on arXiv (85 spreadsheets, intensive inspections). Three substantive corrections came out of the re-sourcing: the page previously gave "~5%" where the compilation says 5.2%; it attributed that rate to "the operational spreadsheets audited" when it covers 43 of 88; and it carried no caveat at all, where the reviewing authors publish three, the hardest being that the study behind 70% of that sample never reported a cell error rate. The 5.2% is kept but demoted and explicitly qualified. Two figures were added: the cell error rates actually measured in operational spreadsheets (2.2% and 2.5%), and the Enron corpus scale that makes the compounding argument concrete.
  • 2026-09: Re-sourced the JPMorgan "London Whale" figure. The bank's own copy of the Management Task Force report had 404'd and no Wayback capture of it exists, which under this hub's rules would have meant pulling the figure. The report itself is archived in full by the Yale Program on Financial Stability, so the citation now points there. The figure and its wording are unchanged. Every other link on this page was re-checked and resolves.
  • 2026-07: Corrected the AFP figure. The page previously paraphrased the frequency as "at least weekly"; AFP's published wording is "on a daily or weekly basis". Added the sample size (362 practitioners, fielded autumn 2024), which was missing, and flagged the ambiguity in AFP's phrasing as to whether the frequency qualifier applies to both the planning and reporting figures.
  • 2026-06: Initial version. Seven figures across prevalence, finance reliance, and notable failures. Citigroup ($900M) and the "24% of large businesses" figure were reviewed and excluded (no primary source / not a spreadsheet error).

Further reading

Compiled by Layerz.

Related reading