Knowledge Loss Statistics: Every Source Checked

Three figures do most of the work in this category. 42% of institutional knowledge is unique to the individual who holds it. Replacing an employee earning £25,000 or more costs an average of £30,614, and 82% of that is lost output rather than recruitment. 92% of organisations do not consistently capture knowledge from staff approaching retirement. All three are traced below to a primary source, with the date, the sample and the method.

That last part is unusual. Most statistics articles in this field cite (McKinsey) or (Harvard Business Review) with no report name, no year and no link. We checked. Several of the most-repeated numbers in knowledge management do not survive contact with their sources at all.

Below is every figure we could trace. At the end are three numbers you have almost certainly seen and should stop using.

How much knowledge is actually at risk

42% of institutional knowledge is unique to the individual who holds it. Panopto, Workplace Knowledge and Productivity Report, July 2018. Online survey of 1,001 US adults working at organisations with 200+ employees, conducted by YouGov. Sponsored by Panopto, a video platform vendor. This is the number that matters most and the one to treat most carefully. It is eight years old and vendor-commissioned. It remains the most-cited figure in the category, which is an argument for someone funding fresh research rather than for repeating it uncritically. It is also the cleanest summary of what actually disappears when someone leaves.

92% of organisations do not consistently capture knowledge from employees approaching retirement. APQC and eGain, The Great Retirement, October 2025. Survey of 1,000 professionals. 58% of C-suite respondents described themselves as very worried about knowledge loss. 85% had not operationalised AI for knowledge management. Sponsored by eGain, a knowledge management vendor, and the full methodology is not published.

75% of organisations say creating and preserving knowledge is important or very important to their success over the next 12 to 18 months. Only 9% say they are very ready to address it. Deloitte, Global Human Capital Trends 2020. The 66-point gap between importance and readiness is the most quotable thing in the whole dataset.

39% of Dutch employers expect staff outflow through retirement within five years. 45% of those expect it to cause problems. Among that group, 90% name the loss of experience and 81% the loss of knowledge. UWV, Uitstroom door pensionering: werkgeversonderzoek, published 31 March 2026. Fieldwork September to December 2025, 3,551 establishments with two or more employees. Of the employers expecting retirement outflow, 45% expect problems as a result: 16% to a high degree, 29% to a reasonable degree. Note the denominator. The 90% and 81% figures describe the employers who anticipate problems, not all employers. Replacement need is cited most often of all, by 95%, with experience and knowledge sitting immediately behind it. In public administration, 89% expect retirements and 67% anticipate problems. This is the strongest source in this article. It is recent, large sample, government, and it separates loss of experience from loss of knowledge, which almost no research does.

2.4 million people aged 55 and over are in paid work in the Netherlands, an increase of 816,000 since 2015. CBS, De arbeidsmarkt in cijfers 2025. Participation among 55 to 65 year olds rose from 61% in 2015 to 76% in 2025. The ageing-workforce exposure is not a forecast. It has already happened.

What replacing someone actually costs

Replacing an employee earning £25,000 or more costs an average of £30,614, of which £25,181 is lost output and only £5,433 is recruitment and logistics. Oxford Economics and Unum, The Cost of Brain Drain, February 2014. UK data, five sectors. This is the best-methodology figure in the field and it is the one almost nobody quotes, because the headline number is less dramatic than the invented ones. The split is what matters. 82% of the cost is the productivity gap while the new person gets up to speed, not the cost of hiring them. Every euro spent on shortening that ramp attacks the large part of the problem. This is also why a hiring problem is often a knowledge transfer problem wearing a disguise. Note that the data is from 2013 and needs inflation adjusting.

Time to reach optimum productivity: 15 weeks for a same-sector hire, 32 weeks from another sector, 40 weeks for a graduate, 52 weeks for someone returning from unemployment. Same source. Sector variation is wide. Legal £39,887 per replacement, accountancy £39,230, IT £31,808, media £25,787, retail £20,113.

The cost of replacing an individual employee ranges from one-half to two times their annual salary. Gallup, This Fixable Problem Costs U.S. Businesses $1 Trillion, March 2019. Gallup also put total US voluntary turnover cost at $1 trillion a year, found 52% of departing employees said their exit was preventable, and found 51% had no conversation with a manager about their role or satisfaction in the three months before leaving. That second pair of numbers is the case for talking to people long before they hand in their notice. This is the defensible version of the salary-multiple claim. See the debunk section for the version that is not.

Median US employee tenure was 4.1 years in January 2026, up from 3.9 years in January 2024. For workers aged 25 to 34 it is 3.0 years. US Bureau of Labor Statistics, Employee Tenure Summary, released 24 September 2026, reference month January 2026. Private sector 3.9 years versus public sector 5.6. Government data, no sponsorship, freely citable. BLS publishes this every two years, so the next release is due in September 2028.

In the Netherlands, 320,000 employees changed employer in Q2 2026, or 4.0%. The series ran 4.7% in Q2 2022, 3.8% in Q2 2025, and back up to 4.0% now. CBS, Werknemer; wisseling van werkgever, Q2 2026, and CBS news release, August 2025, for the earlier quarters. Worth pairing with the retirement figures. Dutch job mobility fell sharply after 2022 and has now stopped falling, while retirement exposure keeps rising. The shape of the problem is shifting from churn to demographics, but churn has not gone away.

How long the gap actually lasts

Only 12% of employees strongly agree their organisation does a great job of onboarding new employees. Gallup, May 2018.

New employees typically take around 12 months to reach their peak performance potential. Gallup, August 2021. Only 29% of new hires say they felt fully prepared after onboarding, and one in five report poor or no onboarding at all.

41% of respondents said their organisation’s critical knowledge sits with colleagues who will retire within five years. ROVC TechBarometer, 2026. Dutch technical sector only, conducted by a training provider, so treat this as sector-specific vendor research rather than a general finding. In the same study, 37% of organisations felt insufficiently prepared and 37% said new employees are deployed independently too quickly.

Half of professional restorers in the Netherlands are aged 55 to 65. SBB trendrapport, relaying Raadsaam Erfgoedprojecten, 2018. Niche, but a vivid illustration of what demographic knowledge loss looks like when it is concentrated in a craft.

Three numbers to stop using

We found these repeated across dozens of pages, including by people who should check. None of them holds up.

"It costs six to nine months of salary to replace an employee"

Gallup attributes this to SHRM, in "Why the Onboarding Experience Is Key for Retention," 16 May 2018. SHRM's own primary benchmarking research reports an average cost-per-hire of $4,129, drawn from fiscal year 2015 data and published in 2016, from a random sample of 2,048 SHRM members. It produces no salary multiple at all. We could not locate any SHRM study that generates a six-to-nine-month figure. It is a circular citation with no traceable origin.

Use instead: Gallup's own one-half to two times annual salary, which is sourced and defensible.

"Fortune 500 companies lose $31.5 billion a year by failing to share knowledge"

This traces to Babcock, P., "Shedding Light on Knowledge Management," HR Magazine 49(5), 2004, which cites IDC with no report name and no year. The underlying IDC research, if it exists, has never been produced. The figure is 22 years old, describes an economy that no longer exists, and has been repeated continuously since.

Use instead: the Oxford Economics per-employee figure, which has a published method and can be scaled to your own headcount honestly.

"Knowledge workers spend 2.5 hours a day, or 30% of the workday, searching for information"

This one has a real source, which is more than most. Feldman, S. and Sherman, C., The High Cost of Not Finding Information, IDC, July 2001, document 01C2996TELECO2996. The problem is what the report actually says. IDC calls the figure "a general estimate" and plugs it into a cost model that readers are told to adjust for their own organisation. It was never a measurement. Twenty-five years of citation have quietly promoted an input assumption into a finding. Competing numbers confirm how unstable it is. McKinsey's The Social Economy (2012) put it at nearly 20% of the workweek, but for a broader activity: looking for internal information or tracking down colleagues who can help. Interact, an intranet vendor, said 19.8% of the week. Three "authoritative" figures for what sounds like one quantity are measuring three different things.

Use instead: nothing. If you need to argue that finding information is expensive, measure it in your own organisation. The published numbers do not support a precise claim.

Why this matters more than it looks

There is a practical reason to care about sourcing beyond intellectual hygiene. Large language models now sit between a lot of research and a lot of readers, and they preferentially surface content where claims are attached to identifiable sources. A page that says "42% (Panopto/YouGov, 2018, n=1,001)" is more useful to a retrieval system than one that says "42% (studies show)". It is also more useful to a human being deciding whether to put the number in a board deck.

The category currently has an unusual amount of fabricated attribution. That is a problem for everyone writing in it, and an opportunity for anyone willing to do the checking. It is the same reason documentation so often misses the part that matters: the useful detail is the first thing dropped.