Every guide on this subject tells you the same thing: get under three seconds. It is not wrong, and it is not the most useful way to think about it — because a threshold invites exactly one reply, "we're at 2.8 seconds, we're fine," and that reply is where the argument stops.

The better frame is a slope. Speed is not a pass mark you clear once. It is a dial that keeps paying all the way down.

The measurement worth quoting

Deloitte Digital and 55, commissioned by Google, tracked 30 million user sessions across 37 brand sites in Europe and the US, matching real changes in load time against real funnels over four weeks. Not a survey, not a lab simulation — actual users.

In travel and booking, a 0.1-second improvement in mobile load time raised conversion by 10.1%.

+10.1%

conversion in travel & booking, per 0.1s saved

Deloitte Digital + 55 for Google, "Milliseconds Make Millions" (2020) · 30M sessions, 37 brands

+1.9%

average order value, same measurement

Deloitte Digital + 55 for Google, "Milliseconds Make Millions" (2020)

−6.5%

homepage bounce rate, same measurement

Deloitte Digital + 55 for Google, "Milliseconds Make Millions" (2020)

That is the strongest speed number available to anyone writing about this, and it is the only one whose headline vertical is travel rather than retail.

The caveat, stated before the argument rather than after it

Those 37 sites were enterprise brands. An owner-run small business has different traffic volume, a different funnel, and no testing infrastructure. What transfers with confidence is the shape of the finding — that conversion responds continuously to speed, in tenths of a second, well inside the range everybody already calls "fast enough". What does not transfer is the exact 10.1%. Cite it as what a Google-commissioned study measured across 30 million sessions. Never as what your business will get.

Three seconds is the floor, not the goal

The familiar threshold still has a job. Booking-engine vendor Exely names three seconds as the point where hotel sites start losing bookings, and a Ryte and HubSpot audit guide puts 40% of users leaving past that mark — a 2020 figure, and we cite it as one, for all devices rather than mobile alone.

Two things happen above three seconds at once. The visitor closes the tab before the page renders — and the visitors who do wait quietly read the delay as a business that does not have its act together.

The under-discussed half. A slow site does not only lose the impatient booker. It quietly lowers everyone else's confidence in how the place is run.

But treat that threshold as a floor. The curve below it still slopes:

Two independent measurements of the same curve. Neither is specific to any one industry, and the conversion column is e-commerce — they describe the shape of the relationship, not your business.
Load timeBounce probability vs 1sE-commerce conversion
1.0sbaseline3.05%
2.0s1.68%
3.0s+32%1.12%
4.0s0.67%
5.0s+90%

Bounce probability: Think with Google / SOASTA, 900,000 mobile landing pages (2016). Conversion: Portent, 20 sites and 5.6 million sessions (approximately 2022). The steepest part of the conversion curve sits between one and three seconds — inside the range a threshold marks as already passing.

E-commerce conversion against page load time: 3.05% at one second, 1.68% at two, 1.12% at three, 0.67% at four.The line falls most steeply between one and two seconds — to the left of the dashed three-second threshold, inside the range a threshold marks as already passing.steepest part of the curve0%1%2%3%the familiar threshold3.05%1.0s1.68%2.0s1.12%3.0s0.67%4.0s
The four measured points from the table above, joined — nothing is interpolated and nothing is extrapolated past four seconds. Instrument: Portent, 20 sites and 5.6 million sessions (approximately 2022), e-commerce rather than your industry. It describes the shape of the relationship, not your business.

The arithmetic, with every input labelled

Here is where most articles on this topic multiply a benchmark by a room rate and print a revenue figure. We are going to do the arithmetic and then not do that, and the reason is the point of the piece.

Take a small hotel as the worked example — the arithmetic transfers to any small business that sells through its site. Every number in this paragraph is an assumption, not a measurement of anything: 20,000 website sessions a year, an average rate of €120 a night, and 2% of sessions ending in a direct booking of one night.

400

direct bookings a year at a 2% conversion rate

Illustrative assumption — not measured, not a benchmark

€48k

direct room revenue those bookings represent

Illustrative assumption — 400 bookings × €120, one night each

€12k

what half a percentage point of conversion is worth, per year

Illustrative assumption — 0.5pp of 20,000 sessions × €120

Half a percentage point — 2.0% to 2.5% — is €12,000 a year at these assumed numbers. That is the useful output: a sense of how little the conversion rate has to move before the money is real. Every 0.1pp is worth €2,400 a year to this imaginary hotel, and nobody would notice 0.1pp in their analytics.

What we are not going to tell you is that fixing your load time delivers that. We could. The studies above would let us multiply something by something and produce a confident euro figure, and page one of this search is full of people who have. We will not, for three reasons: the elasticity was measured on enterprise brands, the conversion curve was measured on e-commerce, and neither was measured on your property in your season with your traffic mix. What the research establishes is sensitivity, not a forecast.

Sensitivity, not a forecast. The research says conversion responds to speed. It does not say what fixing yours would earn — that number does not exist yet, for your property.

The honest version of the claim is smaller and survives scrutiny: conversion responds to speed continuously across the range your site is probably in, and at a small business's volumes a change too small to notice is worth more than it looks.

What this has to do with the rest of your site

In our own audit rubric, speed on a phone carries 18 points out of 100 — the heaviest single thing we measure. This piece is why. It is not because speed is the most interesting problem a website can have; it is because it is the one most likely to be costing a small business customers today, silently, while the site looks completely fine.

The people you lose to it never contact you. There is no bounced email and no complaint. That is exactly why it goes unfixed.

What to actually do

  1. Test on a phone, on mobile data, with the cache cleared. The desktop-on-wifi number is the optimistic case; your customers see the other one.
  2. Read the Performance score, not the SEO score. They are different numbers and only one of them measures how long a visitor waits.
  3. Aim under 1.5 seconds, not under 3. Three seconds is where the damage becomes obvious, not where it starts.

See what your site scores — free

Our free check measures how fast your site loads on a phone and nine other things that decide whether customers and AI assistants can find you. Plain language, measurable issues, about a minute.

See what your site scores — free

No sales call attached. If your site is already fast, the check will tell you that too.