Ask what the ERP implementation failure rate is and the answer lands somewhere between 10% and 84%, depending who you ask. That spread isn't measurement noise. It's twelve studies running five incompatible tests and all calling the result "failure." I spent a day opening every source I could reach, including the SEC filings of the companies behind the famous disasters. Here is what each number actually measures.
The numbers, and what sits behind each one
| Study | Reported failure or overrun rate | Definition of "failure" used | Sample | Source | Year |
|---|---|---|---|---|---|
| Standish CHAOS Report | 16.2% success, 52.7% challenged, 31.1% cancelled. Challenged projects cost 189% of the original estimate | Success = on time, on budget, with all features originally specified. Anything else is challenged or impaired | 365 respondents representing 8,380 applications, US | Standish Group, CHAOS Report | 1994 |
| Eveleens & Verhoef, replication of the Standish method | Standish definitions judged "misleading, one-sided" and producing "meaningless figures" | Tests the definition itself: a project under budget with more functionality than promised scores as a failure under Standish's rules | 5,457 forecasts across 1,211 real projects | IEEE Software, Vrije Universiteit Amsterdam | 2010 |
| Panorama Consulting ERP Report, self-assessment | 60% call the project a success, 30% neutral or don't know, 10% call it a failure | The respondent picks the label. No budget or schedule criterion applied | 172 respondents polled on Panorama's website, Sept 2012 to Jan 2013 | Panorama Consulting, 2013 ERP Report | 2013 |
| Panorama Consulting ERP Report, budget and schedule | 53% over budget, 61% over schedule | Actual cost or duration above the planned figure | Same 172 respondents | Panorama Consulting, 2013 ERP Report | 2013 |
| Panorama Consulting ERP Report | More than half stayed inside their expected budget; more than half inside their expected timeline | Self-reported adherence to the organisation's own expectation | 131 respondents, Aug 2022 to Dec 2023, median cost $450,000, median duration 15.5 months | Panorama Consulting, 2024 ERP Report | 2024 |
| McKinsey & University of Oxford | 45% over budget, 7% over time, 56% less value than predicted | Three-way deviation from the business case, restricted to projects with an initial price tag above $15 million | More than 5,400 IT projects as of June 2012; total cost overrun $66bn | Bloch, Blumberg & Laartz, McKinsey Digital | 2012 |
| Flyvbjerg, Budzier, Lee, Keil, Lunn & Bester, ERP subset | Mean cost overrun ratio 1.3, median 0.9, max 79.8 | Actual cost divided by estimated cost. Below 1.0 means under budget | 1,612 ERP projects inside 5,392 IT projects completed 2002–2014, $56.5bn total | Journal of Management Information Systems 39(3) | 2022 |
| "55–75% of ERP projects fail to meet their objectives" | 55–75% | Undefined | Undisclosed | Attributed to Gartner across dozens of vendor pages; primary source not found | undated |
| Hershey Foods, Q3 order fulfillment disruption | Net sales down 12%, from $1,217.2M to $1,066.7M year on year | Reported financial impact, not a failure rate | One company, one quarter | Form 10-Q filed with the SEC, 12 Nov 1999 | 1999 |
| Nike, i2 demand and supply planning | Q3 FY2001 guidance cut from $0.50–$0.55 to $0.34–$0.38 per share | Reported earnings impact | One company, one quarter | Form 8-K filed with the SEC, 27 Feb 2001 | 2001 |
| Revlon, SAP go-live at Oxford, N.C. | Approximately $50 million of lower net sales in the first six months | Reported revenue impact | One company, one half-year | Form 10-Q filed with the SEC, Q2 2018 | 2018 |
| Lidl, SAP programme | €500m spent, programme abandoned after seven years | Cancellation before full rollout | One company, 10,000 stores in scope | The Register, 12 Dec 2019 | 2018 |
Five studies, five meanings of the same word
Standish says 16.2% of projects succeeded in 1994. Read the definition and the number changes shape. Type 1 success means on time, on budget, every feature specified. Deliver a system three weeks late that everyone ends up using happily and you land in "challenged," reported downstream as not-success. So 83.8% failed a test of estimation accuracy, not of whether the software worked.
Eveleens and Verhoef checked. They applied the Standish definitions to their own data, 5,457 forecasts across 1,211 real projects, and published the result in IEEE Software in 2010. Their verdict: misleading, one-sided, meaningless. Their words. A project that lands under budget with more functionality than promised scores as a failure under Standish's rules, because the forecast-to-actual ratios point the wrong way. Steer on that and you're training your PMO to pad estimates.
Panorama asks something else entirely. In 2013, out of 172 respondents, 86% said they were satisfied with the software, 60% called the project a success, 10% called it a failure. Failure here means someone ticked the box marked failure. Same report, same sample: 53% went over budget, 61% ran long. Three figures, one group of people, contradicting each other. By 2024 the same firm reports most of its 131 respondents inside both budget and timeline. Both cohorts filled in a form on a consulting firm's website, which is not nothing, but it isn't a census.
Then the paper that rearranged my thinking. Flyvbjerg and colleagues, 2022, Journal of Management Information Systems: 5,392 IT projects, 1,612 of them ERP. Mean cost overrun ratio for the ERP subset, 1.3. Median, 0.9. The typical ERP project in the largest assembled sample came in under budget. The mean is dragged up by a tail reaching 79.8, a project that cost eighty times its estimate. The authors state the average cannot be calculated at all, because the tail has infinite variance, and that overruns and underruns are equally frequent.
As for the Gartner figure everyone quotes, I went looking for the original. Gartner's own ERP page carries different claims, and every citation trail ended at a vendor blog quoting another vendor blog. It sits in the table flagged unsourced.
What the companies said in their own filings
For the famous cases I pulled the filings instead of the retellings. Hershey attributes its 12% sales drop to "problems in order fulfillment (customer service, warehousing, and shipping) encountered since the July start-up of a new integrated information system." Philip Knight names "complications arising from the impact of implementing our new demand and supply planning systems." Revlon reports "approximately $50 million of lower net sales."
Worth noting what isn't there. The figures everyone repeats about Hershey, $100 million in unfilled orders and a $112 million project cost, appear nowhere in the 10-Q, and I found no primary source for either. None of the three reports a failure rate, because that isn't something a company measures about itself.
A warehouse in October
A mid-sized distributor I worked with went live on a new ERP in March. By the Friday of go-live week: no blocking incidents, orders flowing, payroll out the door. The integrator closed the project. In every study above, that's a success. Under Standish, possibly a Type 1.
I was back on site in October for something else. The receiving team was scanning inbound pallets into the system, then writing the same line onto a clipboard sheet pinned by the dock door. The supervisor said the putaway suggestion sends them to a slot that's often already full, so they log where the pallet really went on paper and fix it later, when there's time. There usually isn't time. Stock accuracy in that zone had drifted since June and nobody upstream knew, because the system reported itself as accurate.
Different floor, same company: a buyer had rebuilt her reorder logic in a spreadsheet, because the ERP's minimum-order-quantity field couldn't handle a supplier's price breaks. Her sheet was better than the system, and lived unbacked-up on her laptop.
Neither shows up as an overrun or a slip. The project was on time, on budget.
Where I part ways with the standard reading
The usual conclusion: ERP projects fail because they're too big, so phase them, govern them harder, hire change management.
I don't buy it, and not for contrarian reasons. Look at what each study measures. Standish, estimation accuracy. Panorama, a feeling at survey time, or budget adherence, depending on the page. McKinsey, deviation from business case. Flyvbjerg, a cost ratio. The filings, one quarter of revenue. Five instruments pointed at five different things, and one thing unites them: not one measures whether people use the system.
The whole value of an ERP is that the data in it reflects the business. A project that lands on time, on budget, with full scope, and whose stock figures are quietly wrong from month four scores as a success in every study above. The clipboard by the dock door is invisible to all of them.
Which means the low failure rates are as suspect as the high ones. Under budget and unused is still a write-off.
If I could replace the table with one metric: six months after go-live, what share of intended users complete the target task in the system, with no parallel file. Nobody publishes that, I suspect because the number would be ugly, and because it means measuring after the consultants have left.
FAQ
What is the real ERP implementation failure rate?
There isn't one, because "failure" means something different in each study. Standish's 1994 definition puts 83.8% of projects outside success. Panorama's 2013 self-reported figure is 10%. Flyvbjerg's 2022 sample of 1,612 ERP projects has a median cost overrun ratio of 0.9, meaning the typical project came in under budget. All three are defensible; they measure different things.
Did Hershey's ERP failure really cost $100 million in unfilled orders?
Hershey's own Q3 1999 10-Q reports a 12% drop in net sales and attributes it to order fulfillment problems after the July start-up of the new system. The $100 million figure circulates widely but I could not trace it to a primary source.
Are there ERP success stories with numbers behind them?
Rarely with published figures. Companies file disclosures when a rollout hurts earnings, not when it goes well, so the public record is biased toward disasters. The closest thing to a positive dataset is Flyvbjerg's 2022 ERP subset, where roughly half the projects came in at or under their estimated cost.
Sources
- Standish Group, The CHAOS Report, 1994 — https://personal.utdallas.edu/~chung/SYSM6309/chaos_report.pdf
- J. Laurenz Eveleens & Chris Verhoef, The Rise and Fall of the Chaos Report Figures, IEEE Software, 2010 — https://www.cs.vu.nl/~x/chaos/chaos.pdf
- Panorama Consulting Solutions, 2013 ERP Report — https://web.archive.org/web/20130228102235if_/http://panorama-consulting.com:80/Documents/2013-ERP-Report.pdf
- Panorama Consulting Group, 2024 ERP Report — https://4439340.fs1.hubspotusercontent-na1.net/hubfs/4439340/Reports/ERP%20Report/2024-erp-report-panorama-consulting-group.pdf
- Michael Bloch, Sven Blumberg & Jürgen Laartz, Delivering large-scale IT projects on time, on budget, and on value, McKinsey & University of Oxford, 1 October 2012 — http://web.archive.org/web/20251013095742/https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/delivering-large-scale-it-projects-on-time-on-budget-and-on-value
- Bent Flyvbjerg, Alexander Budzier, Jong Seok Lee, Mark Keil, Daniel Lunn & Dirk W. Bester, The Empirical Reality of IT Project Cost Overruns: Discovering a Power-Law Distribution, Journal of Management Information Systems 39(3), 2022 — https://arxiv.org/pdf/2210.01573
- Hershey Foods Corporation, Form 10-Q for the quarter ended 3 October 1999, filed 12 November 1999 — https://www.sec.gov/Archives/edgar/data/47111/0000047111-99-000081.txt
- NIKE, Inc., Form 8-K filed 27 February 2001 — https://www.sec.gov/Archives/edgar/data/320187/000032018701000003/0000320187-01-000003.txt
- Revlon, Inc., Form 10-Q for the quarter ended 30 June 2018 — https://www.sec.gov/Archives/edgar/data/887921/000088792118000011/rev2018q210-q.htm
- Lindsay Clark, ERP disaster zone: The mostly costly failures of the past decade, The Register, 12 December 2019 — https://www.theregister.com/2019/12/12/erp_disaster_zone_the_mostly_costly_failures_of_the_past_decade/
- Gartner, Enterprise Resource Planning (ERP) topic page — https://www.gartner.com/en/information-technology/topics/enterprise-resource-planning (consulted for the origin of the widely quoted 55–75% figure; the page was not reachable at the time of writing and the figure could not be traced to any Gartner publication)
I run AdoptionLayer, which works on software adoption inside companies. The field scenes in this article are composite; the figures are not, and each one links to its source above.