Ten numbers every marketing deck repeats. We went looking for where they came from.

Somewhere this quarter, one of these numbers will end up on a slide with your name on it. Nobody who builds a deck has time to chase a statistic through 20 years of blog posts, so we did, and we opened everything we could. Where a document was paywalled or gone, we say so. “No source we could find” is the honest limit here, because the absence of a source isn’t proof that none exists.

1. “Your attention span is eight seconds”

The Microsoft Canada report of spring 2015 shows the figure on a slide, credited to Statistic Brain. Microsoft’s own fieldwork, 2,000 Canadians online and 112 by EEG, didn’t produce it, and the report opens with “Think digital is killing attention spans? Think again.” [1] Time still called it “a new study from Microsoft Corp.” [2] Statistic Brain credited the US National Library of Medicine and the Associated Press. The BBC asked both in 2017, and “neither can find any record of research that backs up the stats.” [3]

Say instead: there’s no evidence human attention has shrunk to eight seconds, and it depends on the task.

2. “People see 4,000 to 10,000 ads a day”

The trail reaches back to a 1960s calculation of 1,500 ads, for a family of four, as told on the BBC. The 5,000 came from Yankelovich’s J. Walker Smith, and at the time it was “up to” 5,000 for a person living in a city. [4] He told the BBC it was “a high estimate” and “a speculative number.” [5] Think about the arithmetic: over 16 waking hours, 10,000 ads is one every 5.8 seconds. One published estimate, Media Dynamics in 2014, put US adults at about 360 exposures across five media, with 150-155 noticed. [6] It’s a consultancy’s figure and it predates most social and app inventory.

Say instead: nobody has counted this well, and one 2014 estimate was about 360 a day.

3. “Acquiring a customer costs five times what keeping one does”

HBR’s 2014 version says “anywhere from five to 25 times,” which is quite a range for a law. [7] Keiningham and colleagues, tracing it in 2005, called it a “platitude” whose exact origin is hard to determine, and pointed to the Technical Assistance Research Project in the late 1980s. [8] We couldn’t open that report. Those authors sold loyalty work themselves, so they aren’t neutral either. The direction is probably right for many businesses. The multiple is decoration.

Say instead: keeping customers is often cheaper, and here’s what acquiring one costs us.

4. “Email returns $36 for every $1”

This number has two parents. The UK’s DMA reports in pounds: £42.24 for 2018, from an online survey of 197 respondents, of whom 62% “think they are measuring.” The question asked for the “approximate return,” and averages were built from bucket midpoints. [9] Currency got swapped along the way, and a trade summary rendered the £38 in the DMA’s 2015 report as “$38 for each $1.” [10] Litmus reported $36 from its 2020 survey. Its own 2025 re-ask found 35% at $10-36, 30% at $36-50, 5% above that, and 21% not measuring. [11] So $36 is now the edge of a bucket, and the largest group sits under it. We found no peer-reviewed randomized holdout, only vendor material, to tell you what’s incremental, which is the question that a lifecycle program has to answer.

Say instead: in Litmus’s 2025 survey, the largest group reported $10 to $36 back per $1, and one in five didn’t measure. These are self-reports.

5. “A 5% gain in retention lifts profit 25-95%”

Reichheld and Sasser’s 1990 article, “Zero Defections: Quality Comes to Services,” said 25% to 85%, as quoted by Ipsos. [12] We couldn’t read the paywalled original. The 95 appears, in what we could open, in a 2000 article about e-commerce companies. [20] Then there’s the unit. Byron Sharp points out that the “5%” is five percentage points of defection, from 10% to 5%, which is a halving, and that the cost of achieving it is assumed to be zero. [21] Sharp’s school argues the opposite position, so weigh that. With a flat margin and a 10% discount rate, our own calculation has customer lifetime value rising 20% when retention goes from 80% to 85%, and 14.3% from 70% to 75%.

Say instead: 1990 calculations suggested halving defection could lift profit substantially in some service firms, before the cost of getting there. Model your own.

6. “Google Ads returns $8 for every $1”

Google publishes the formula, to its credit: 2 + 0.7 × 5 × 2 - 1 = 8. The $2 is an assumption drawn from Hal Varian’s analysis of a proprietary sample of auctions, inferring value by assuming advertisers maximize profit. [13][15] The 5 is the ratio of organic to paid clicks from a study of the search engine Dogpile, and the 0.7 is Google’s choice. [16] So $7 of the $8 is organic clicks. Watch the wording too: in 2009 the report said “$2 in revenue for every $1,” which it called $1 of profit. [14] In 2025 it’s “$2 in profit,” from the same formula. And when eBay ran field experiments, a naive return of over 4,100% became -63%. [17]

Say instead: Google estimates this by assumption, mostly from organic clicks. It isn’t a benchmark for your account, so measure with a holdout, as with brand search.

7. “68% of online experiences begin with a search engine”

BrightEdge’s 2019 report is accurate: “68% of all trackable website traffic,” organic 53.3% plus paid 15.0%, and it states that it excludes direct traffic. [18] The sample is thousands of domains from an SEO vendor. Then the roundups got hold of it, and “trackable website traffic” became “online experiences begin.” [19] That’s a population swap, and BrightEdge’s own PDF doesn’t make it. We couldn’t find who did first.

Say instead: in one SEO vendor’s 2019 sample, search supplied 68% of website traffic once direct was excluded.

8. “92% trust recommendations, and 88% trust reviews like a friend”

Nielsen’s 92% comes from more than 28,000 respondents in 56 countries, in fieldwork from 2011. [22] It has stayed in range, at 83% in 2015. [23] BrightLocal’s 88% from 2014 is another matter. It came from an opt-in panel, and 62 of those points were conditional: 32% “as long as there are multiple ones to read,” and 30% “if they believe it to be authentic.” [24] Its 2026 wave reads 49%, from a representative panel of 1,002 US adults. [25] Part of that fall is likely a change of sample and question, so it isn’t a pure trend. Review vendors still quote the 88.

Say instead: people say they trust people they know more than any ad, and about half of US consumers say they trust online reviews that much.

9. “B2B buyers are 57% through before they talk to sales”

This one holds, with limits. CEB and Google surveyed more than 1,500 customers of 22 large B2B suppliers in 2011, and 57% is the average of recalled progress before first serious sales contact. [26] Its popular sibling is a different statistic altogether. SiriusDecisions’ 67% is the share of the journey “done digitally,” and its own executive wrote that it “in no way says that no one talks to a salesperson.” [27] A 2025 vendor survey from 6sense lands at 61%, down from 69%. [28]

Say instead: first serious contact comes around 57-61% of the way through, on recalled averages with wide variation.

10. “100 milliseconds costs 1% of sales”

Greg Linden described Amazon’s A/B tests in 2006, in which small delays caused “substantial and costly drops in revenue.” The 1% comes from his slides, which are gone, as cited by Ron Kohavi and colleagues. [29][30] Bing measured 0.6% of revenue per 100 milliseconds. [30] Google found 0.2% to 0.6% fewer searches per user. [31] The same Kohavi paper shows late-loading elements with no detectable effect, so it isn’t universal.

Say instead: the direction is well established, and your own number needs your own test.

What the survivors have in common

The failures come in a handful of types. Some have no measurement at all. Some are assumptions presented as results, like the $8. Some are self-reports presented as measurements, like email. Others swap the population, as with the 68%, or drift in unit: pounds to dollars, points to percent, revenue to profit.

Almost every number here was published by someone who sells the thing it flatters: Google, Litmus, BrightEdge, Bain, Microsoft’s ad team, 6sense, BrightLocal. That doesn’t make them wrong. BrightEdge’s PDF is accurate, and the damage happened downstream. It does tell you which denominator to check first. Most of the directions survive, too. Email is cheap and it works, search is large, and waste is real. Tearing down the number doesn’t tear down the idea.

Before a figure goes in your deck, ask:

• Who counted, and do they sell the thing being counted?
• What’s the denominator, and did it survive the copying?
• What year is it, and has the same source updated it?
• Was it observed, or did people report it?
• Can you open the original?

Our own version of this exercise is the 70% cart abandonment rate and the 60/40 split. The replacement sentences above are less exciting than the numbers they replace. They’re also the ones you can defend when a sharp person in the room asks where it’s from.

Sources

1. Microsoft Canada · Attention spans, 2015third-party mirror, since Microsoft’s own link no longer resolves to it; the slide credit is on page 6
2. Time · You now have a shorter attention span than a goldfishMay 2015, the misattribution in print
3. BBC · Busting the attention span mythMarch 2017, the sources contacted directly
4. The New York Times · Anywhere the eye can see, it’s likely to see an adJanuary 2007, archived copy with the “up to” wording
5. BBC More or Less · Do we see 10,000 adverts per day?January 2024, an unchecked machine transcript of the interview
6. Media Dynamics · Adults spend almost 10 hours per day with the mediaSeptember 2014 press release, consultancy estimate for five US media
7. Harvard Business Review · The value of keeping the right customersOctober 2014, by Amy Gallo
8. Keiningham and colleagues · Loyalty Myths, excerpt2005 book excerpt from authors who sold loyalty work
9. DMA · Marketer email tracker 2019archived copy, 197 respondents, sponsored by an email vendor
10. Emailmonday · DMA national client email report 2015the currency swap, in a trade summary
11. Litmus · The ROI of email marketing2025 survey of nearly 500 marketers, from a vendor selling email tools
12. Ipsos · Shattering the myths of customer loyaltyquotes the 25% to 85% wording of the 1990 article
13. Google · Economic impact methodologythe published formula, current edition
14. Google · Economic impact report 2009the earlier revenue wording
15. Hal Varian · Online ad auctionsAmerican Economic Review 2009, on a proprietary sample of auctions
16. Google · Economic impact FAQJansen and Spink’s Dogpile ratio of 5.3 to 1, rounded down
17. Blake, Nosko, and Tadelis · Consumer heterogeneity and paid search effectivenesseBay field experiments, one advertiser
18. BrightEdge · Channel report 2019May 2019 data, direct traffic excluded
19. Ahrefs · 107 SEO statistics for 2026an example of the reworded version credited to BrightEdge
20. Bain · E-loyalty: your secret weapon on the webarchived summary of the July 2000 HBR article
21. Byron Sharp · Loyalty mythscritique from a rival school of thought
22. Nielsen · Consumer trust in online, social and mobile advertising grows2012 release on 2011 fieldwork
23. Nielsen · Global trust in advertising 201583%, completely or somewhat
24. BrightLocal · Local consumer review survey 2014opt-in panel, 2,104 entries, conditional answers
25. BrightLocal · Local consumer review survey2026 wave, 1,002 US adults
26. CEB and Google · The digital evolution in B2B marketingmore than 1,500 contacts at 22 suppliers
27. SiriusDecisions · Three myths of the “67 percent” statisticJuly 2013, archived copy of the originator’s correction
28. CustomerThink · 6sense study provides critical insights on B2B buyer behaviorsecondary account of a vendor survey with nearly 4,000 responses
29. Greg Linden · Marissa Mayer at Web 2.0November 2006, Amazon’s delay tests with no figure on the blog
30. Kohavi and colleagues · Seven rules of thumb for web site experimentersKDD 2014, citing Linden’s slides and reporting Bing’s result
31. Google Research · Speed matters2009, experiments with 100 to 400 milliseconds of delay

MS

Editor-in-chief

Mikołaj Salecki

Writes about media, tech, and AI business for people who actually run digital. Former agency lead. Skeptic of frameworks that read better than they perform.

LinkedIn ↗Facebook ↗salecki.digital ↗

More articles →