An earlier version of this post, published in December 2025, said that 1 in 3 shoppers never return after one bad experience. The research behind that headline does not say that, so this version replaces it. Below are three kinds of evidence, kept apart: what customers say they would do, what executives report about downtime, and what companies have actually measured after a failure. The last section covers why retention is worth the effort in the first place.
What the “1 in 3” figure measures
The figure comes from PwC’s Future of Customer Experience Survey 2017/18, published in March 2018. PwC surveyed a representative sample of 15,000 people in 12 countries (4,000 of them in the US) through an online survey and in-field interviews. Its findings are about bad customer experiences in general, not website downtime.
| What PwC reports | Result | Population |
|---|---|---|
| Would stop doing business with a brand they loved after one bad experience | 32% | All customers surveyed |
| Would walk away from a company or product they love after one bad experience | 17% | US |
| Would walk away from a company or product they love after several bad experiences | 59% | US |
| Would walk away from a brand after one bad experience | 49% | Latin America |
Three things limit how far this can be pushed. These are stated intentions (PwC’s wording is “say they will walk away” and “would stop doing business”), not behavior PwC observed. The global 32% applies to brands people already love, and the US figure for one bad experience is 17%, roughly half the headline. And the data is from 2017 and 2018.
A more recent survey is specific to website errors. In a May 2021 Fullstory survey of 1,500 North American consumers, 77% said they would leave without completing a transaction if they hit an error, 65% said they trust a business less after a problem on its website or app, and 60% said they were unlikely to return to a site after one such error. Fullstory is a software vendor, and the article does not give the fieldwork dates, methodology or question wording.
Stated intent does not carry over one for one into behavior. A 2006 meta-analysis of 47 randomized experimental tests (Webb and Sheeran, Psychological Bulletin) found that a medium-to-large change in intention (d = 0.66) led to a small-to-medium change in behavior (d = 0.36). That is general psychology, not a measurement of customers, but it is the reason a survey answer of “I would leave” should not be read as a churn rate.
What executives report about downtime
Splunk and Oxford Economics survey executives at Global 2000 companies, the world’s largest firms by sales, profits, assets and market value. In its 2024 report Splunk defines downtime as any service degradation or outage of a business system, so slow service counts.
| Finding | June 2024 edition | May 2026 edition |
|---|---|---|
| Respondents | 2,000 executives, 53 countries | 2,000 executives, 20 countries |
| Annual cost of unplanned downtime, Global 2000 | $400 billion (about $200 million per company) | $600 billion, up 50% in two years (about $300 million per company) |
| Customer loss | 29% of respondents said they had lost customers because of downtime | 81% of technology leaders cite loss of customers as a consequence of downtime |
| Customers detect it first | 41% of technology executives said customers are “often” or “always” the first to detect downtime | 47% admit customers are often or very often the first to detect service degradation or outages |
The 2024 report also says 44% of respondents claim downtime damages their reputation and 40% of CMOs say it affects average customer lifetime value.
Three limits apply. The dollar totals are estimates calculated from survey answers (for 2024, Splunk says Oxford Economics did the calculation), not audited losses. The customer-loss figures are executives saying they lost customers, with no count of how many or for how long, and the 2024 and 2026 questions cover different groups, so 29% versus 81% is not a trend. And the sample is the largest companies in the world, so it says little about what an outage costs a 50-person SaaS company or a local shop.
One finding appears in both editions and is the useful one: a large minority (41% of technology executives in 2024, 47% in 2026 as Splunk reports it) say customers are often the first to detect a problem.
What has been measured after real failures
Two large failures come with numbers the companies themselves published. Both were far bigger than a three-minute outage, and neither separates one bad experience from everything else going on, but they are the nearest thing to observed behavior in the public record.
TSB, UK bank, 2018. On 22 April 2018 TSB moved to a new IT platform. The UK Financial Conduct Authority says a significant proportion of TSB’s 5.2 million customers were affected and that the bank did not return to business as usual until December 2018. TSB’s 2018 annual report says:
- Around 80,000 customers switched their current account away from TSB in 2018, peaking in the second quarter when the IT problems hit, against around 50,000 in 2017.
- Around 140,000 customers opened a new account or switched an account to TSB in 2018, and TSB had 3.8 million current account customers.
- About 204,000 complaints had been received since the migration by 30 January 2019, of which TSB estimated 25% would have occurred in the ordinary course of business.
- Charges and waived income from the disruption came to £330.2 million, with an estimated £153.0 million recoverable from its IT provider.
That is about 30,000 more switches away than in 2017, under 1% of its 3.8 million current account customers, in a year when about 140,000 others opened or moved an account to the bank. It is a real cost from a failure that lasted months. It is not 1 in 3.
Southwest Airlines, US airline, December 2022. Southwest’s fourth-quarter 2022 results report more than 16,700 flight cancellations in late December and a fourth-quarter pre-tax hit of about $800 million. Its first-quarter 2023 results put a further pre-tax hit at about $380 million, of which about $325 million was lost revenue from “cancellations of holiday return travel and a deceleration in bookings for January and February 2023 travel.” Southwest said the revenue impact was primarily isolated to January and February, that March 2023 saw a robust recovery, and that first-quarter revenue was a record $5.7 billion.
Southwest was a flight-operations failure, not a website outage, so read it as one company’s accounts of how long demand stayed depressed after a visible failure. Together the two cases show a measurable loss: extra customers switching away at TSB, and slower bookings at Southwest for about two months. Neither shows most affected customers gone for good, and neither tells you what happens at a small business, where we found no published data.
Slow pages: the measured version of the question
Slowness does not throw an error, but it can also send visitors away. Two widely quoted figures on this are old. In Google’s analysis of 3,700 mobile sites that opted into sharing Google Analytics benchmark data (March 2016), 53% of mobile site visitors left a page that took longer than three seconds to load. That counts visits that ended, not customers who never came back.
The same Google report cites a 2017 Google and SOASTA model: a neural network trained on bounce and conversion data, with 90% prediction accuracy according to Google, estimated that compared with a 1-second load, the probability of a mobile visitor bouncing is 32% higher at 3 seconds, 90% higher at 5 seconds and 123% higher at 10 seconds. Those are modeled relative increases in bounce probability, not the share of visitors lost. For current Core Web Vitals pass rates, see our guide to page load time benchmarks.
Why retention is still worth protecting
A 2014 Harvard Business Review article says that, depending on which study you believe and the industry, acquiring a new customer is anywhere from five to 25 times more expensive than retaining one. It also cites Bain & Company’s Frederick Reichheld for the claim that a 5% increase in retention increases profits by 25% to 95%. The article names no study for the first range and gives no sample, year or method for the second. The Bain note it links to gives one figure on this: in financial services, a 5% increase in retention produces more than a 25% increase in profit. It does not contain the 95% figure or a study year. Treat the multiples as rules of thumb, not forecasts. None of these numbers measures what an outage costs.
What to take from this
- Treat a visible failure as a measurable, recoverable cost. In both measured cases the loss was visible in company figures, concentrated in the months right after the failure. The evidence does not support “1 in 3 never return.”
- Close the detection gap. Splunk’s respondents say customers often spot problems first. Checking your site from outside your own infrastructure is how you find out sooner.
- Measure your own data. Mark incident start and end times, then compare cancellations, returning-user sessions and orders in the 30 days after an incident with a comparable period without one. Your own numbers will answer this better than any survey above.
Where Site Qwality fits
Site Qwality runs HTTP, TLS and DNS checks against your site and alerts you by email, SMS, Slack, Microsoft Teams, Discord, Telegram or webhook, so you hear about an outage from a monitor instead of a customer. A status page lets you tell customers what is happening. The free tier needs no credit card and includes 100K uptime checks a month at 5-minute intervals (HTTP checks from one location), one status page on a Site Qwality subdomain and 10 SMS credits a month. Paid uptime plans check every 30 seconds. It can tell you when a problem started and how long it lasted. It cannot tell you how many customers left because of it, which has to come from your own analytics. Start free.
Sources
- PwC, Experience is everything: here’s how to get it right (Future of Customer Experience Survey 2017/18, 15,000 people, 12 countries, March 2018), and PwC’s summary page (Internet Archive copy, because the original address now redirects to PwC’s consulting page).
- Fullstory, digital experience consumer survey (1,500 North American consumers, May 6, 2021).
- Webb and Sheeran, “Does changing behavioral intentions engender behavior change?”, Psychological Bulletin 132(2), 2006.
- Splunk, The Hidden Costs of Downtime 2026 and the Cisco release of May 19, 2026 (Oxford Economics survey of 2,000 Global 2000 executives, 20 countries).
- Splunk, The Hidden Costs of Downtime (2024 report; Oxford Economics survey of 2,000 Global 2000 executives, 53 countries) and Splunk’s 2024 summary.
- UK Financial Conduct Authority, TSB fined £48.65m (December 20, 2022).
- TSB Banking Group plc, Annual Report and Accounts 2018.
- Southwest Airlines, fourth quarter and full year 2022 results (SEC Form 8-K exhibit, January 2023).
- Southwest Airlines, first quarter 2023 results (SEC Form 8-K exhibit, April 2023).
- Think with Google, Find out how you stack up to new industry benchmarks for mobile page speed (February 2017; Google Analytics data, March 2016; Google/SOASTA Research, 2017).
- Harvard Business Review, The Value of Keeping the Right Customers (Amy Gallo, October 29, 2014) and the Bain & Company note it links to (Internet Archive copy).