Site speed and conversion: which pages should you speed up first?

A slow page or a button that responds badly can cut an enquiry short. To decide where to spend, measure the pages that matter and check how they behave on mobile. A technical score on its own won’t tell you how many sales you are losing.

The Vodafone case study published on web.dev reports a 31% improvement in LCP and 8% more sales in an A/B test. That figure holds for that particular test, and your business has to measure its own.

Another public study deals directly with lead generation. Milliseconds Make Millions, commissioned by Google and carried out by Deloitte and the agency 55, followed 37 sites of large European and American brands for 30 days at the end of 2019. On lead generation sites, a 0.1-second improvement in speed metrics, on every page of the journey, went with a 21.6% increase in visitors reaching the form page and 7% more page views. Keep its limits in mind: large brands, a short period, and a metric tracked at the time, First Meaningful Paint, since dropped in favour of LCP.

The three metrics to understand

Core Web Vitals measure loading, responsiveness and visual stability. The “good” thresholds are assessed at the 75th percentile of visits, with mobile and desktop counted separately.

Metric	What the visitor sees	Good threshold
LCP	The main content appears	2.5 seconds or less
INP	The page responds to an interaction	200 milliseconds or less
CLS	Elements stay where they are	0.1 or less

An average can hide the visitors who get the worst experience. Ask which pages and which devices make up the measurement as well.

INP replaced FID in March 2024. It tracks how the page responds to clicks and typing throughout the visit, so menus, filters and forms are directly affected. CLS measures how stable the page is, and it weighs on conversion too. When the “Request a quote” button drops down just as a banner appears, the visitor clicks the wrong thing, tries again or gives up.

Real data or simulation?

PageSpeed Insights shows field data when there is enough of a sample, then a simulated diagnosis. The field data describes what your visitors experience; the simulation helps you understand the causes.

A single Lighthouse page load can’t measure real INP, because it doesn’t follow what visitors click and type. It can show blocking metrics that help with the diagnosis. If field data is missing, test on real devices and consider measuring Web Vitals on your own site.

The field report aggregates a period of time, so a fix shipped this morning won’t show up in it straight away. Note the dates of your changes before comparing.

The simulated score is still a diagnostic tool. On the Nallian redesign, the home page’s Lighthouse score went from 61 to 95 out of 100. That figure shows a page that is lighter to load. It measures neither the experience of real visitors nor the enquiries received, and the case study attributes no rise in conversion to that gain.

The wrong-number trap

Some tools, such as GTmetrix or Pingdom, put full page load time front and centre. That time counts every tracking script and every third-party widget, including the ones the visitor never sees. A page can finish loading late and still feel pleasant if the main content appears quickly and responds to clicks.

Chasing a perfect score can cost weeks for a gain nobody notices. Work on the LCP, INP and CLS of the pages that get traffic, starting with mobile. That is often where the thresholds are exceeded.

Server response time deserves a look too. A site hosted on a single server, with no content delivery network in front of it, responds more slowly to distant visitors. That fix belongs to hosting.

Which pages should you fix first?

Pick the pages that attract relevant visitors and play a part in an enquiry: service, campaign, product or contact. Check the main content, the menu and the form.

Sort the problems you find before deciding what to fix. An oversized hero image, a third-party script and a slow interaction each call for a different response.

Have the banners and widgets that appear after loading tested as well. A button that moves at the moment of the click can trip up the visitor even when the content appeared quickly.

Struggling metric	Common causes	Possible fixes
LCP	Heavy hero image or one discovered late, slow server, CSS or JavaScript blocking rendering	Image in the right format and dimensions, loaded with priority; caching and a content delivery network; critical CSS inlined in the page
INP	Heavy JavaScript running on click, third-party scripts (chat, ad tags, testing tools)	Remove or defer scripts you don’t need; break up long tasks
CLS	Images without dimensions, banner inserted above the content, font that changes the text size	Reserve the space; show the banner on top of the page; declare media dimensions

This table gives you starting points. The PageSpeed Insights diagnosis names the elements at fault on your page, and that is what settles it.

What should you ask your developer for?

A diagnosis should name the elements responsible and propose costed fixes. Ask for a before-and-after measurement on the same pages, under comparable conditions.

For images, adjust the format and the dimensions. Don’t delay loading the image that makes up the main content without looking at it first. For scripts, separate the ones the journey needs from the ones that can be removed or loaded later. Reserve space for elements likely to arrive late.

After each fix, test login, form submission and your measurement tools. An optimisation that breaks tracking or payment has made the site worse.

Finally, ask for a test on a mid-range phone over a mobile network. Emulation on a desktop computer usually paints a rosier picture than the one your visitors get.

How do you tie the work to your business?

Track the enquiries and opportunities from the pages you fixed. Compare similar periods and similar traffic sources. A change in a campaign can shift conversion even when speed stays the same.

For an estimate, use the margin per client, the share of enquiries that become clients and several gain assumptions. Present them as scenarios. Navigation or accessibility errors can justify a fix even without a financial return you can measure precisely.

A hypothetical example: the pages concerned get 4,000 visits a month and convert at 1.5%, which makes 60 enquiries. One enquiry in five becomes a client, with a margin of €2,000 per client.

Scenario	Enquiries per month	Clients per month	Extra margin per year
No effect	60	12	€0
Conversion +5%	63	12.6	€14,400
Conversion +10%	66	13.2	€28,800

These percentages are working assumptions, to be replaced with your own figures. Nobody can tell you in advance which of these scenarios will play out on your site. Compare them with the cost of the work: if even the cautious scenario doesn’t cover it, your main blocker probably lies somewhere other than speed.

Keeping the result after launch

Check your pages after adding a chat tool, a plugin or a new image. Ask who watches for regressions and how they get fixed.

A performance budget helps you hold the line over time: for example, LCP under 2.5 seconds and INP under 200 milliseconds on your main pages. A tool such as Lighthouse CI can check it on every deployment and flag a breach before it reaches your visitors. If your site has too little traffic to appear in Google’s field data, sending Web Vitals to your analytics tool takes over.

Your choice of technology affects maintenance; speed also depends on images, scripts and everything added after launch. The CRO guide covers the rest of the journey review, and a redesign can build these targets into the quote from the start. A free audit of your site also shows where speed ranks among the other blockers on your pages.

Frequently asked questions

Does site speed affect search rankings?

Yes, to a degree. Google states in its documentation on page experience that its ranking systems use Core Web Vitals. The same page says that good results don’t guarantee a top position and that Google looks for the most relevant content first.

What PageSpeed score should you aim for?

Look at the field data first: LCP, INP and CLS in the “good” range at the 75th percentile, on mobile and on desktop. The score out of 100 is a simulation, useful for following a diagnosis. Pushing it up only has value if it improves the experience of real visitors.

How do you measure without field data?

Test your main pages on a real phone over a mobile network, and time the key actions: opening the menu, clicking the button, sending the form. Add Web Vitals collection to your analytics tool to follow your own visitors.

Where do you start when speeding up an existing site?

With the heaviest element on the pages that matter, often the hero image or a third-party script. Convert images to the right format and dimensions, remove scripts you no longer need, then reserve space for the elements that arrive late.