The real cost of a four-second load, in revenue not milliseconds

A freight brokerage came to us in early 2025 with a straightforward brief: their website looked dated and they wanted it redesigned. Forty thousand sessions a month, a healthy ad budget, and a conversion rate they described as "probably about average."

It was 0.4%.

We asked to look at the analytics before agreeing to anything. What we found had almost nothing to do with how the site looked.

The number nobody was watching

Their Largest Contentful Paint on a 4G connection was 4.2 seconds. Two-thirds of their traffic was mobile. Which meant that for the majority of the people they were paying to acquire, the page was blank for longer than most people are willing to wait.

The site wasn't converting badly because it was ugly. It was converting badly because a large share of visitors never saw it at all.

Here is what was actually on the page before anything useful rendered:

Roughly three megabytes to render a headline, a paragraph and a button.

What we changed, and nothing else

This is the part that makes the result worth writing about. For six weeks after launch we changed nothing else. Same copy. Same offer. Same ad spend, same campaigns, same targeting. The only variable was delivery.

• Statically rendered pages, served from the edge.
• A hard 90 KB JavaScript budget enforced in CI — a pull request that exceeds it fails the build.
• AVIF images with WebP fallback, correctly sized per breakpoint, priority-loaded for the hero only.
• Self-hosted variable fonts, subset to Latin, preloaded.
• Third-party tags deferred until after first interaction.

LCP went from 4.2s to 0.8s.

What happened

Traffic did not increase. Sessions were flat to slightly down. Conversions rose 2.4× in six weeks.

Mobile conversion rate went from 0.29% to 0.81%. Desktop, which had never had a serious speed problem, moved from 0.71% to 0.79% — a small lift attributable to noise and a marginally faster interaction time.

The mobile number is the whole story. Nearly three-quarters of the improvement came from a segment that had been silently failing.

The arithmetic that matters

Speed work gets discussed in milliseconds, which is exactly why it loses budget arguments. Nobody approves a project to save 3,400 milliseconds.

Convert it to the unit the person signing the invoice actually thinks in:

• 40,000 sessions/month × 0.52 percentage-point lift = ~208 additional conversions per month
• At their close rate and average contract value, that was worth roughly $180,000 in annualised gross profit
• The rebuild cost a fraction of that and paid for itself inside the second quarter

What this does not mean

Speed is not a growth strategy. If your offer is wrong, or you're buying traffic from people who were never going to buy, a fast site just delivers that failure more efficiently.

What speed is is a precondition. It's the floor below which nothing else you do can be measured properly, because a meaningful share of your audience is being excluded from the experiment entirely.

If you have not measured your LCP on a real mobile connection — not on your desktop, on your office fibre, with a warm cache — you do not currently know what your conversion rate is. You know what it is among the people patient enough to wait.

Those are different numbers, and the gap between them is usually where the money is.

---

We wrote up the full engagement, including the form redesign that came afterwards, in the Northbeam Logistics case study. If your own numbers look like the ones at the top of this post, tell us about it.