The figure everyone quotes but nobody applies
In 2006, Amazon published a now-famous finding: every extra 100 ms of latency costs 1% of sales. Twenty years later, that ratio still holds, and recent studies confirm it.
Vodafone cut its Largest Contentful Paint (LCP) by 31% in 2024. The result: +8% in sales, from a faster site alone, with no redesign and no new campaign.
The problem: your tech team tracks LCP in Lighthouse or PageSpeed Insights. Your CMO watches ROAS in Meta Ads. Nobody connects the two in euros.
The raw data
Studies agree on one point: beyond 2.5 seconds of load time, every extra second destroys conversion in a predictable way.
| Load time | Average conversion rate | Loss vs. benchmark (1 s) |
|---|---|---|
| 1 second | 3.05% | Reference |
| 2 seconds | 2.44% | -20% |
| 3 seconds | 1.83% | -40% |
| 4 seconds | 1.47% | -52% |
| 5 seconds | 1.08% | -65% |
On mobile, the drop is even steeper: 20% less conversion per extra second according to Portent/Google data. And mobile accounts for 65 to 75% of French e-commerce traffic in 2026.
The euro math nobody does
Take a typical French e-commerce site: €500,000 in monthly revenue, 200,000 sessions/month, an average order value of €85. Its current LCP is 2.5 seconds, within Google's "Good" threshold.
What happens when LCP degrades?
| LCP | Estimated conversion rate | Estimated monthly revenue | Loss vs. 2.5 s |
|---|---|---|---|
| 2.5 s (baseline) | 2.94% | €500,000 | €0 |
| 3.0 s | 2.65% | €450,500 | -€49,500 |
| 3.5 s | 2.35% | €399,500 | -€100,500 |
| 4.0 s | 2.06% | €350,200 | -€149,800 |
| 4.5 s | 1.76% | €299,200 | -€200,800 |
| 5.0 s | 1.47% | €249,900 | -€250,100 |
Going from 2.5 s to 4 s of LCP costs about €150,000 a month for this kind of site. Even a modest 500 ms slowdown means €49,500 in lost revenue.
Why LCP degrades without anyone noticing
LCP isn't a static metric. It fluctuates constantly, and some slowdowns go unnoticed for days:
- Front-end deployment: an unoptimized hero image, a third-party script that blocks rendering, a badly loaded React component
- Saturated CDN: a Friday-evening traffic spike, response times doubling
- Ad script: a new GTM tag that adds 800 ms of latency on mobile
- A/B test: variant B loads a carousel that doubles the LCP
What they have in common: these slowdowns show up neither in Sentry (no server error) nor in GA4 (no LCP metric by traffic source). They stay invisible until the monthly review, when it's too late.
The specific case of paid traffic
Your site's average LCP means nothing for your paid budget. What matters is LCP on paid traffic: the visitors coming from Meta, Google Ads or TikTok.
Why the distinction matters:
- Paid traffic often lands on specific landing pages, not the home page
- Those pages often carry more tracking scripts (Meta, Google Ads, TikTok pixels)
- Paid users are more impatient: they didn't search for your brand, they clicked an ad
- Every bounce on paid traffic is wasted CPC: €0.50 to €2.50 per lost click
On a €30,000/month paid budget with an average CPC of €0.85, an LCP slowdown that raises the bounce rate by 15% means about €4,500 in wasted clicks per month.
What Korvus does differently
Performance monitoring tools exist. Lighthouse, WebPageTest, Chrome UX Report all measure LCP. But they do it for the whole site, with no breakdown by traffic source and no financial translation.
The Korvus perf_drop alert works differently:
- It measures LCP on paid traffic only (sessions identified by UTM, gclid, fbclid)
- It compares real-time LCP against baselines calculated every night
- When the threshold is crossed, it calculates the financial impact: affected sessions x average CPC x estimated bounce rate
- The CMO gets a clear message: "LCP at 4.2 s on Meta traffic: €850 in estimated bounces over 3 hours"
Instead of a Lighthouse chart, the CMO gets a euro amount, the source and the action to take.
How to act
If you don't yet monitor LCP on paid traffic, three immediate actions:
- Measure your LCP by traffic source: not the global LCP, the LCP of your paid landing pages on mobile
- Calculate your cost per second: take your monthly revenue, apply the table above, put a euro figure on it
- Set up a real-time alert: LCP can degrade on a Friday evening and nobody will see it before Monday
LCP is a financial problem that shows up in technical metrics. As long as your tech team sees milliseconds and your CMO sees a falling ROAS with no explanation, nobody connects the two.
Key takeaways
- 100 ms of latency = roughly 1% fewer sales: the order of magnitude Amazon measured on page latency, confirmed on LCP by Vodafone
- On a €500,000/month site, a 1.5 s slowdown costs €150,000/month
- LCP on paid traffic is the only one that should trigger a financial alert
- LCP slowdowns go unnoticed for 3 to 5 days on average without dedicated monitoring
- Turning LCP into euros is the only way to trigger action on the business side