Estimating the impact website loading times have on conversion rates
- Published
- 2025-07-08
Table of Contents
A business website is supposed to convert visitors into leads or customers. The rate at which that happens is called the conversion rate.
Base Rate: If a website had 0ms loading time and rendered instantly on any device, the loading time couldn't impact the conversion rate, right? Let's call that the Base Rate. It'd fully depend on factors other than loading time, such as who you're sending to your website and how convincing your offer is for them. Things like website design and your brand factor in here as well.
Burn Rate: If you have a Base Rate of 10% and your loading speeds turn that into 8.5%, you have a Burn Rate of 15%, because you burned 15% of your potential conversions.
Survival Rate: The opposite of the Burn Rate. The Survival Rate is the percentage of your potential conversions that you didn't lose because of your website's loading speed.
Spark Point: The Spark Point is the loading time at which your website still loses practically no (0.1%) visitors due to loading speeds. The conversion rate at the Spark Point is practically equal to the Base Rate, and reducing loading times below this doesn't make much of a difference. Beyond that, you start burning conversions.
Now let's take a look at the flip side: If your website takes ten minutes to load practically everyone would assume it's broken and move on to a competitor. But ten minutes feels pretty high. Maybe you'd lose all your visitors at five minutes, or three, or one. When's the last time you've waited longer than 30 seconds for a website to load?
Ashen Point: Let's call the loading time at which 99.9% of visitors bounce off of the website the Ashen Point. That's the time at which your website is considered broken or otherwise unuseable, and practically everyone leaves. You've burned all your conversions, you might as well not have a website at that point.
It's obvious that loading speeds have some impact on conversion rates. But how does it evolve? Are you good as long as you're under five seconds? Is it gradual? Or do visitors stay with all websites that load in under three seconds and leave for everyone slower than that?
Loading time: "Loading time" can mean a lot of things. It can mean the time it takes for the server to send the first byte. It can mean the time until all files are downloaded that the website could need. You could define loading time as the time it takes for the website to render its first few things, maybe a loading screen. The one most relevant in my opinion though is Largest Contentful Paint: It measures the time it takes for the website to show its biggest blocks of content. This is usually the core offer of the website. You can check your own website on PageSpeed Insights.
We want to know how loading times map to Burn Rates, so that we can factor that out of the current conversion rate and calculate an optimal conversion rate based on what's possible with current technology. Because, let's be honest, "faster websites are better" isn't that helpful for decision makers in businesses.
So let's do the math.
Priors
We can make the following assumptions:
- There's the Spark Point at which practically none (0.1%) of your conversions are burned.
- There's the Ashen Point at which practically all (99.9%) of your conversions are ash.
- Everything in between the Spark Point and the Ashen Point maps to some Burn Rate between 0.1% and 99.9%.
Just knowing this, let's make a few first guesses:
- Faster is always better than slower. It's unlikely that the 500ms Burn Rate costs you fewer conversions than the 400ms Burn Rate.
- Most burn probably comes from initial load time. Once you've loaded a website for the first time, not only are you already on there looking around and reading, your browser also cached a lot of files it just downloaded. Clicking around on there will go faster because there's fewer files to download the second time around.
- The Spark Point is higher than 0ms. The average human reaction time is between 200 and 300 milliseconds. If you could get your LCP to under that time, odds are that your Burn Rate is super close to 0%.
- The Burn Rate probably doesn't change that much in the first 100 or so milliseconds over Spark Point. Then it probably accelerates a bit as it moves into the seconds, and beyond that it's probably falling off again.
Based on my own personal behavior here's what I would estimate a few key Burn Rate points to be:
- 0.1%: Anything below 500ms is practically instant. Let's put 500ms down for this one.
- 5%: Two seconds is pretty high, I think I leave more than 5% of websites there. At one second... That about rings true. Let's call it 1s.
- 50%: I practically never wait five seconds for a website to load unless I open it in the background. At three seconds I generally stay for more than half the websites, unless again I open several tabs. Maybe four seconds? Let's go with that.
- 95%: At seven or eight seconds I even shoot down tabs I opened in the background. I'll go with seven seconds here.
- 99.9%: One in a thousand... Excluding websites I have to use, e.g. government websites, you're practically guaranteed to lose me at fifteen seconds. At that point I assume your website is broken.
This would be on a desktop computer. What about mobile? PageSpeed Insights groups LCP times into buckets for mobile and desktop: How Lighthouse determines your LCP score
An LCP of 2.5 is considered moderate on mobile, and 1.2 for desktop. 4 seconds on mobile map to 2.4 on desktop for slow. That's a 2.083 multiplier for moderate and a 1.667 multiplier for slow. Let's average that out to 1.875 for a combined mobile / desktop multiplier, and factor that in at 50/50:
But my own wild guess isn't exactly a good model on its own. Let's take a look at a few actual studies:
Portent
Source: Site Speed is (Still) Impacting Your Conversion Rate (2022)
Portent measures the Goal Conversion Rate based on Page Load Speed, and they give us a percentage for each 1s interval:
- 1s: 39% conversion rate
- 2s: 34% conversion rate
- 3s: 29% conversion rate
- 4s: 24% conversion rate
- And from here it flattens out: 10s clocks in at 17%
Interesting... So, no sigmoid curve. Just exponential decay. And quite a lot of people sticking around for ten seconds. If we look at how the data plots out, we can estimate a Base Rate of about 42%, maybe 43%. That's a 10s Burn Rate of 60%.
Portent Goal Conversion - Burn Rate by Loading Time
Back to the e-commerce. How does that pan out?
- 1s: 3.05% conversion rate
- 2s: 1.68% conversion rate
- 3s: 1.12% conversion rate
- 4s: 0.67% conversion rate
- Things get weird here again: At 5s the conversion rate is back up to 1.08%.
A rough guess for the Base Rate here would be what, maybe 4%? That puts the 1s Burn Rate at 25%, the 2s Burn Rate at about 58%, and the 3s Burn Rate at 72%. That's a lot. Just losing 25% alone is rough - that means your customer acquisition cost increases by 33%. At 2s your CAC is now more than double.
Portent E-Commerce - Burn Rate by Loading Time
Looking at the two charts though, I can't help but notice that both level out under 100%. The E-Commerce is obviously just a really rough guess, considering that the data is really wacky in any case. But the Goal Conversion looks pretty clear. It looks like there's no big difference between nine and ten seconds loading time, and fifteen seconds loading times probably don't matter much either?
Except there's gotta be a point at which everyone gives up on the website. There's no online store where if the loading time is ten hours, people still sit it out. Which means, at some point the Burn Rate has to accelerate again to get to the 100%. Maybe the sigmoid curve still makes sense here?
Not quite, but a double sigmoid should do it:
Portent Goal Conversion - Double Sigmoid Model
Yeah, that looks far more reasonable.
But my own estimate doesn't fit on it:
My Estimated Burn Rates - Double Sigmoid Model
Maybe I'm just really impatient? I can imagine 'ten seconds until burn' to be at the far bottom of the patience distribution. Mhm... Let's look at some more data.
Google Mobile Site Speed Playbook
Source: Google Mobile Site Speed Playbook
Some interesting findings here: "Up to 27%" increase in conversion rates for a single second loading speed increase. This is most likely an improvement from 1.5s to 0.5s - the lower improvements seem to be the ones strongest correlating with bounce rate, so far. A 27% improvement thus translates to:
- Base Rate = 1.27 x 1.5s Survival Rate
- 1 / 1.27 = 1.5s Survival Rate
- 0.79 = 1.5s Survival Rate
- 0.21 = 1.5s Burn Rate
Google Mobile Site Speed Playbook - Double Sigmoid Model
Slightly higher than expected, but overall very close!
Money.co.uk: Google mentions that their bounce rate goes from 80% to 52% for a change in load time from 11.7s to 3.5s. Let's add that:
Google Mobile Site Speed Playbook (All Data Points) - Double Sigmoid Model
Generally higher than our model based on the Portent Goal Conversion data, but the shape checks out so far. Unfortunately, the other data in the Playbook is too abstract to be mapped to a proper model. In fact, most "studies" online are just quick before and after. Most of the data in the Site Speed Playbook is in the form of "we optimized our website according to the PageSpeed Insights recommendations, and got a conversion rate increase of XY%."
That doesn't cut it. There's gotta be better data on this.
Maybe something like a fast website slowed down on purpose, just to see what happens?
A Study at Bing (2009)
Source: The Book of Speed - Chapter 1
Turns out, yes! Microsoft, an eternity ago, decided to do some experiments on Bing and randomly slowed down the search results page:
A Study at Bing (Any Clicks) - Double Sigmoid Model
Okay, that doesn't help. This was in 2009, but loading times were pretty comparable to today: The State of Web Performance Optimization 2010.
Greg Linden, 2006
Source: Greg Linden's Blog - Marissa Mayer at Web 2.0
There've been people doing studies on this in 2006! And even then a half second delay added 20% to their Burn Rate.
Source: Conductor - Amazon Page Speed Study
In his "Make Data Useful" class Greg Linden then also mentions that every 100ms delay costs Amazon about 1% of sales. But no data set I can plot.
Pingdom's Study (2018)
Source: Pingdom - How Page Load Time Really Affects Burn Rate
This is data that's actually usable. They mapped out their bounce rates across page loading speeds, which is almost exactly what we've been looking for. Not perfectly what we've been looking for, but we can make it work.
Even at a loading speed of 0ms, a visitor still has a chance of bouncing off of your website (leaving before clicking around). The opposite of that would be the stay rate, that is, the chance of them not bouncing off of the website at a loading speed of 0ms.
The same way, at 0ms loading speed a visitor still has a chance of not converting. We called that the Base Rate.
If we call the rate at which visitors do anything at all on our website the stay rate, and the rate at which visitors do what we want them to do (that is, convert) the Base Rate, we can see that it's very likely that loading speed has a somewhat similar impact on the both of them: In either case, with higher loading speeds there's less doing. The only difference is that to not be considered a bounce, a visitor only has to take a single action. To convert, the visitor has to take several.
Let's take the data and calculate how much of the stay rate is burned due to loading speeds. For that we'll translate their bounce rate into their stay rate first, then calculate how much was burned in relation to an estimated base stay rate at 0.5s loading time:
| Loading Time (s) | Bounce Rate (%) | Stay Rate (%) |
|---|---|---|
| 1 | 7 | 93 |
| 2 | 6 | 94 |
| 3 | 11 | 89 |
| 4 | 24 | 76 |
| 5 | 38 | 62 |
| 6 | 46 | 54 |
| 7 | 53 | 47 |
| 8 | 59 | 41 |
| 9 | 61 | 39 |
| 10 | 65 | 35 |
| 11 | 62 | 38 |
| 12 | 67 | 33 |
| 13 | 69 | 31 |
| 14 | 66 | 34 |
| 15 | 69 | 31 |
| 16 | 73 | 27 |
Based on this, the base stay rate can be estimated to be about the 94% (the stay rate is pretty constant across the first two seconds, let's just take the higher one).
Pingdom's Study - Double Sigmoid Model
Huh, that's almost directly confirming our model here. But it appears that in this study, the burn rate only really starts picking up after the 2s mark. Except... what we're interested in is the burn rate of conversions, not the burn rate of visitors.
As mentioned above, this study only measures how many people take any action on your website. A conversion usually takes several actions. And while the visitor benefits from files being cached after the first click, at least some files are usually loaded when navigating around.
How does this look if we move it 2s to the left?
Pingdom's Study (Shifted by 2s) - Double Sigmoid Model
Pretty damn close. Just moving the data around is obviously not the best way of science, let's be honest. But it's giving us some info: The shape of the curve that's describing the impact loading speeds have is correct.
Walmart (2012)
Source: Walmart PageSpeed Slide - SlideShare
Walmart's Cliff Crocker gave a presentation on their conversion rate vs load time, exactly what we're looking for. Except... they didn't label their y axis.
We can make this work as well, though: We don't actually care about their real conversion rate, just the burn rate that's associated with their website's loading speed. It doesn't matter if the top of the y axis is 100, or 55, or 12. How much it's going down relative to that point is the same either way.
Here's their data in table form, as best as it could be read from the graph:
| Loading Time (s) | Conversion Rate (relative) | Burn Rate (%) |
|---|---|---|
| 0.5 | 90 | 0.0 |
| 1.5 | 36 | 60.0 |
| 2.5 | 22 | 75.6 |
| 3.5 | 13 | 85.6 |
| 4.5 | 12 | 86.7 |
| 5.5 | 7.5 | 91.7 |
| 6.5 | 8 | 91.1 |
| 7.5 | 8 | 91.1 |
| 8.5 | 9 | 90.0 |
| 9.5 | 8 | 91.1 |
| 10.5 | 8 | 91.1 |
| 11.5 | 1 | 98.9 |
| 12.5 | 3 | 96.7 |
| 13.5 | 8 | 91.1 |
| 14.5 | 8 | 91.1 |
| 15.5 | 1 | 98.9 |
We're just assuming a conversion rate at the y axis of 100 to make the math easier.
Let's plot it:
Walmart - Double Sigmoid Model
Okay, wow. That's significantly worse. Our chart should look more like this:
Walmart Fitted - Double Sigmoid Model
What's going on here? Let's review:
The Portent goal conversion data was just, well, goal conversion. Getting a visitor to just enter their email for a newsletter is far easier than getting them to fill a shopping cart and then enter all their payment details. The Portent e-commerce data made the impression that it's pretty similar to the goal conversion one, but... it is significantly stronger up to the 4s mark, and afterwards it gets weird.
The Pingdom study is the burn rate for visitors that would potentially do anything on a website, but don't, because of the loading time. The impact being less there makes sense.
But the Walmart study is for online shopping. That's a lot of clicks to browse around, look at products, click on the reviews, read the texts, add some of them to the shopping cart, log into your account, enter payment details, and then finally order.
That's a lot of steps. And each of them is impacted by the loading speed.
Accounting for conversion complexity
So... we know for sure that the loading speed of a website has an impact, and a surprisingly strong one. But how strong, that depends on the amount of hoops the visitor has to jump through:
Combined Data: Portent and Walmart
And interestingly enough, the Portent e-commerce data suddenly seems more reasonable. Of course, it gets weird after the 4s mark - but before that it matches the Walmart data pretty well.
One more thing, though... mobile users are more tolerant to loading times than desktop users. But the studies above don't tell us what devices were used to visit their websites. We don't have any good data on that... But we can make a rough estimate based on the way Google judges website performance differently for mobile and desktop: A really rough estimate here is that one second on desktop is about as impactful as 1.875 on mobile. With the assumption that the studies above had a 50/50 split, we can adjust the chart accordingly:
Device-Weighted Burn Rate Model
Conclusion
The data available isn't amazing. But the few useful sources that can be found online paint a similar picture: Your website's loading speed has a massive impact on its conversion rate. And just a few seconds make the difference between your website being a dead or a flourishing sales channel - the first few seconds in particular.