Do Employees Waste 9.3 Hours a Week Searching for Information?

No. The figure of 9.3 hours a week spent searching for information, universally attributed to McKinsey, appears zero times in the McKinsey report it is attributed to. We downloaded the full 184-page report and searched it. The report says 8.8 hours, and that number is IDC's, not McKinsey's. The 9.3 version traces to an uncited 2013 guest blog post written by a knowledge-management vendor's business development manager.

That is the whole finding, and every clause of it is checkable in a browser in about five minutes. This page shows the chain link by link, because the interesting part is not that one number is wrong. It is that the number has been circulating for thirteen years, in pitch decks and board papers and procurement business cases, with nobody in the chain having opened the document.

Do employees waste 9.3 hours a week searching for information?

The honest answer has three parts, and collapsing them is what produced the myth in the first place.

One. There is no measurement anywhere in this citation chain. Not at the top, not at the bottom. The oldest ancestor calls its own figure “a general estimate”. The most-cited descendant has no sample at all because it was invented in a blog post.

Two. There is a real sentence in a real McKinsey report, and it is 19 percent of working hours, not 9.3 hours. It is also not a McKinsey survey. It is McKinsey re-basing IDC estimates against a workweek McKinsey itself chose.

Three. There are modern surveys with real samples, and every one of them was commissioned by a company that sells the remedy. They are citable, provided you name the commissioner and the sample in the same breath. Most published content does neither.

Every version of the number, and where each one leads

Read this by variant, because the variants have genuinely different standing. Two of them are real sentences in real documents. One of them is not in any document at all.

What the McKinsey report actually says

The artefact is The social economy: Unlocking value and productivity through social technologies, McKinsey Global Institute, July 2012. We downloaded the full report PDF, 4,445,882 bytes, 184 pages, and searched the extracted text. Here is the sentence everyone is reaching for, verbatim from page 47:

“Social technologies address the most important aspects of an interaction worker's job. Typically, such a worker spends 65 percent of a workday collaborating and communicating with others. This includes 28 percent of work time reading, writing, or responding to e-mail, and 19 percent of working hours trying to track down information needed to complete tasks.”

Nineteen percent, not 9.3 hours. And the hours version is in the technical appendix, in Exhibit A7, where it reads 8.8. Eight point eight divided by 46.5 is 18.9 percent, which is the 19 percent.

The exhibit is worth reproducing because it shows what McKinsey actually did, which is arithmetic rather than fieldwork. The left column is headed “Activities in IDC report”. The right column is McKinsey's “adjusted activity list”.

IDC's activity list sums to 65.8 hours a week. McKinsey capped the week at 46.5 hours and compressed the seven residual activities into a single 18.3-hour bucket called “role-specific tasks”, while carrying three activities across unchanged. The report says so itself, verbatim:

“We base our breakdown of interaction workers' time spent on various working activities on International Data Corporation estimates, which were based on multiple surveys on how workers spend their time, as well as McKinsey proprietary data. From this data set, we identified the three activities that are performed by almost all interaction workers and for which social technology could have an impact: reading and answering e-mail, searching and gathering information, and communicating and collaborating with colleagues. The data show an average workweek of 46.5 hours, so we aggregated the time spent on other tasks as being 'role-specific tasks.'”

And the entire citation behind the 19 percent and the 28 percent is one footnote, number 60, verbatim:

“McKinsey Global Institute analysis based on proprietary McKinsey data and Susan Feldman, Hidden cost of information work: A progress report, International Data Corporation, May 2009.”

So the 19 percent is McKinsey's re-percentaging of an IDC hours figure against a workweek McKinsey chose, blended with proprietary data that is not described. There is no McKinsey survey, no n, no fielding date and no panel. That is not a criticism of the report, which is transparent about all of this in its own appendix. It is a criticism of thirteen years of people citing the report without reading the appendix.

Where did 9.3 hours come from?

From here, and it is still live. The page is a conference website, the byline reads “August 26, 2023 / May 17, 2023 / by UTR Conf Staff”, and the sentence is verbatim:

“Our employees spend too much time searching for their needed information. How much time? According to a McKinsey report, employees spend 1.8 hours every day, 9.3 hours per week, on average, searching and gathering information. Put another way, businesses hire 5 employees but only 4 show up to work; the fifth is off searching for answers, but not contributing any value.”

House style on this site does not use long dashes. The original wraps “9.3 hours per week, on average” in them where commas appear above. Nothing else in the sentence is changed.

No report name. No year. No page number. No hyperlink: a search of the served HTML for any link whose address contains “mckinsey” returns nothing, while the word McKinsey itself appears five times. The article closes by recommending a knowledge-management product.

Now open the Internet Archive's capture of the same URL from 9 June 2013. It is categorised “Guest Blog”, it carries the identical “1.8 hours every day, 9.3 hours per week” sentence, and it ends with one line that the live version no longer has:

“This is a guest blog post by Jason Shechtman, Business Development Manager at Senexx. Follow Senexx at @senexxInc”

The older ancestor: an IDC white paper sponsored by a search vendor

Before McKinsey there was IDC, and before the 19 percent there was “2.5 hours a day”. The artefact is The High Cost of Not Finding Information, an IDC white paper by Susan Feldman and Chris Sherman, July 2001, ten pages. We downloaded it. The back cover reads, verbatim: “Sponsored by Inktomi”. Inktomi was, at the time, a commercial enterprise-search vendor.

The paper does not present the number as a measurement, and to its credit it says so plainly:

“In these scenarios, IDC uses a salary plus benefits number of $80,000 for a typical knowledge worker. We use a general estimate that the typical knowledge worker spends about 2.5 hours per day, or roughly 30% of the workday, searching for information. This number also needs to be adjusted to reflect the circumstances of each specific enterprise. IDC believes the number represents a general average of time spent searching based on the ubiquity of intranets within organizations.”

No sample. No fielding date. No panel. No instrument. The paper's own words are “general estimate” and “IDC believes”. Every downstream citation of “IDC found that employees spend 2.5 hours a day searching” is converting an explicitly labelled assumption into a finding.

Three years later the analyst described the exercise herself, in KMWorld, and the wording is even franker. She lists studies by IDC, the Working Council of CIOs, AIIM, Ford and Reuters finding that “knowledge workers spend from 15% to 35% of their time searching for information”, gives no sample for any of them, and then writes: “Using those studies as a basis, we set out to quantify the impact that not finding information might have on a typical enterprise of a thousand knowledge workers.” The costed result is attributed, in her own phrase, to “our mythical organization”. The dollar figures also drift between documents: the 2001 white paper says $2.5M, $5M and $15M for the same modelled firm; the 2004 article says $6M, $12M and $15M.

The factor-of-five inconsistency inside that same paper

This one is checkable in thirty seconds and, as far as we can tell, is not published anywhere else. The famous “$2.5 million a year” figure is computed from 2.5 hours per week, while the assumption printed directly above it in the same bullet list says 2.5 hours per day. Verbatim from Scenario 1, with the division signs rendered as they come out of the PDF text layer:

“Assumptions: Knowledge worker salary = $80,000 annual salary plus benefits. 1,000 knowledge workers x 2.5 hours/day searching on average. Calculation of cost: $80,000 ( 52 weeks ( 40 hours/week x 2.5 hours/week searching x 1,000 knowledge workers x 50% unindexed information. Conclusion: An enterprise employing 1,000 knowledge workers wastes $48,000 per week, or nearly $2.5 million per year, due to an inability to locate and retrieve information.”

The arithmetic settles which reading the published figure used. $80,000 divided by 52 weeks and then by 40 hours is $38.46 an hour. Multiply by 2.5 hours, by 1,000 workers, by 50 percent, and you get $48,077, which is the stated $48,000 a week. Using 2.5 hours per day would give roughly five times that.

Be precise about what this does and does not show. The correct statement is that the paper's stated assumption and its published dollar figure disagree by a factor of five. It is not “IDC's number is wrong”. Both lines are in the same bullet list, a reader can confirm it in half a minute, and it means the widely quoted per-day figure and the widely quoted dollar figure cannot both be describing the same calculation.

What about the modern surveys?

They exist, they have real samples, they disclose their commissioners, and they are perfectly citable provided you carry the disclosure with the number. What they are not is independent research about the world. They are self-reported estimates collected by companies selling the remedy.

Coveo, an enterprise-search vendor, commissioned Arlington Research for its 2022 Relevance Report: “The survey comprised a nationally representative sample of the working population across both the UK and USA, with 4,000 adults aged 18+ taking part, evenly distributed between each country. All respondents were people who use a computer for work, as a part of companies which contain more than 250 employees.” The headline: “A survey of 4000 employees found that workers spend an average of 3.6 hours every day searching for information at their jobs.” The 2025 successor, dated 29 April 2025, names the same field house and the same 4,000-person panel but moves the employer-size screen from 250+ to 5,000+ employees, and in the release as served to us on 6 September 2026 it gives no hours figure at all, saying only that employees “waste hours daily searching for needed information”. We have seen a 3.0-hour figure attributed to that report; we could not confirm it in the release itself, so this page does not publish it. Neither release states its fielding window.

Glean, also an enterprise-search vendor, reports that “workers say AI automation saves them roughly 11 hours a week”, from “a survey of 6,000 full-time digital workers across the United States, United Kingdom, and Australia, conducted between December 2025 and January 2026”. No field house is named and the time savings are self-reported. Worth noting that Glean publishes the counter-number in the same programme: “Workers spend 6.4 hours a week botsitting, more time than they spend using AI to produce work.”

Microsoft's Work Trend Index is the best-documented of the family and the one that can be cited without embarrassment, because it names everything: “conducted by an independent research firm, Edelman Data x Intelligence, among 31,000 full-time employed or self-employed knowledge workers across 31 markets between February 6, 2025 and March 24, 2025. This survey was 20 minutes in length and conducted online. 1,000 full-time workers were surveyed in each market”, with an additional 4,500-person US oversample. It is still a vendor survey and its productivity figures are still self-reported.

None of the four carries a confidence interval. None has been replicated by an independent party. That is not a scandal, it is what commissioned market research is, but it does mean a slide that presents any of these as “research shows” is overstating its own evidence.

What should you put in your business case instead?

Your own number, and it is easier to get than people expect.

Any of the published figures will survive about one question from a finance director, because none of them was measured in your company, none has a confidence interval, and most were paid for by a company selling search software. If the case rests on 9.3 hours, it rests on a blog post. That is a bad position to be in halfway through a procurement.

The measurement that does hold up is small and local. Pick ten questions your team actually asks in a normal week, the recurring ones. Time how long the answer takes to find today, including the message to the colleague who knows and the wait for their reply. Multiply by frequency. You now have a figure with a method you can name, a sample you chose, and a scope your finance director recognises, which is exactly what the famous statistics do not have. It will almost certainly be smaller than 9.3 hours, and it will be defensible, and defensible beats large.

The second thing worth doing before you buy anything is to check what the tool would be allowed to see. An assistant indexed across your systems surfaces what an employee could technically already reach, which is not the same as what they were meant to see. We have written that up separately, with vendor documentation behind it, in why an internal AI assistant surfaces documents people were never meant to find. If EU data handling is part of the decision, EU data residency for internal AI assistants compares what nine vendors actually commit to, and does your AI vendor train on your data covers training defaults, retention and admin controls per account tier. If a vendor is selling you an agent rather than a search box, the difference between a retrieval assistant and an agentic one gives you the three questions that separate them.

How to check all of this yourself in five minutes

We would rather you did not take our word for it, so here is the exact procedure. It is short on purpose.

1. Open the McKinsey Global Institute report PDF and use your reader's find function on “9.3 hours”. Then try “1.8 hours”. Then try “8.8” and look at Exhibit A7 in the appendix.

2. Open the live blog post where the 9.3 figure appears and look for a link to a McKinsey report. There is none.

3. Open the 2013 capture of the same page and read the last line.

4. Open the IDC 2001 white paper, read the back cover, then read the Scenario 1 assumptions and the calculation directly beneath them.

If any of those checks comes out differently for you than it did for us on 6 September 2026, the links are here so that you can catch it before we do. This page carries a date because it needs one: the live blog post has already been republished once with its byline changed, and archived captures are the only reason that is visible at all.