Web performance agency

The scope of a web performance engagement

A web performance agency engagement typically covers five blocks: a Core Web Vitals audit on real-user data, a prioritized action plan, technical implementation in the codebase, continuous monitoring, and internal team training. That is exactly the scope I have been running for the past 8 years as an independent consultant.

I am Paul Delcloy. I work with demanding brands (CHANEL, April Moto, Leroy Merlin, Rimowa, Fenty Beauty, SiriusXM) on this complete scope, from diagnosis to knowledge transfer. Every engagement is scoped and priced before commitment, measured in RUM, and documented for your technical team.

Context varies from one project to another: a redesign in progress that starts drifting on LCP, an e-commerce site whose TTFB explodes under load, a strategic SEO site that just dropped into the red zone on Search Console. The method, however, stays the same. We measure, we prioritize, we ship, we transfer.

What web performance changes concretely for your business

Web performance impacts three business drivers that are directly measurable.

SEO: Core Web Vitals have been a Google ranking signal since 2021, folded into the Page Experience weighting. On a competitive query with equivalent content, the site with green CWV takes the lead. The official mobile LCP threshold is 2.5 seconds at P75 to stay in the green zone, and Google Search Console silently penalizes URLs that drift above it.

Conversion: the Deloitte 2020 "Milliseconds Make Millions" study, commissioned by Google and run on 30 retail and travel brands, measures that a 0.1-second gain in mobile speed lifts retail conversions by 8.4% and travel conversions by 10.1%. On a site with €500k monthly revenue, the order of magnitude of gains from an LCP optimization engagement is counted in tens of thousands of euros recovered per quarter.

Advertising cost: Google Ads factors landing page experience into the Quality Score calculation, which weights the actual CPC. Core Web Vitals are not cited by name in the Ads documentation, but the measured mobile experience enters the evaluation, with CPC drops observed in practice when a landing page moves from red to green. The mechanism is indirect but measurable in post-optimization A/B tests.

The Core Web Vitals levers I ship in engagements

On a production site, three families of levers determine whether Core Web Vitals pass to green.

Critical rendering path and LCP

The work targets the LCP image, critical CSS, and fonts. The identified LCP image must be served in AVIF or WebP with an explicit preload, never lazy-loaded:

<link rel="preload" as="image" href="hero-800.avif"
      imagesrcset="hero-400.avif 400w, hero-800.avif 800w, hero-1200.avif 1200w"
      imagesizes="(max-width: 600px) 100vw, 50vw"
      fetchpriority="high">

<img src="hero-800.avif"
     srcset="hero-400.avif 400w, hero-800.avif 800w, hero-1200.avif 1200w"
     sizes="(max-width: 600px) 100vw, 50vw"
     width="1200" height="600"
     fetchpriority="high"
     alt="Featured product">

On the font side, two failure modes must be addressed: FOIT (Flash of Invisible Text: text stays invisible until the web font loads) and FOUT (Flash of Unstyled Text: a visible switch from system font to web font). font-display: swap protects against FOIT; the size-adjust, ascent-override, and descent-override descriptors on the fallback font protect against the CLS caused by FOUT.

JavaScript and INP

On a React SPA, INP stabilizes by working three axes: effective bundle splitting (routes with React.lazy, dynamic imports), unnecessary re-renders (React.memo, useMemo, useCallback used surgically, not by carpet bombing), and third-party scripts that saturate the main thread. A tag manager loading 15 synchronous tags can single-handedly push an INP into the red above 200 ms.

Back-end and infrastructure

TTFB is driven by page cache (varnish, redis, CDN cache), SQL queries, and CDN strategy. On a high-traffic Magento e-commerce site, dividing TTFB by 4 is common with a well-orchestrated Varnish cache and a database optimized on catalog queries. Cache dilution by tracking URL parameters (utm_*, gclid, fbclid) is a recurring cause that any audit must always check.

Measuring on real-user data, not Lighthouse alone

The Lighthouse score is a starting point, not an objective. Google does not rank on Lighthouse; it ranks on Core Web Vitals measured in RUM from real Chrome users, aggregated over a 28-day rolling window and exposed via the Chrome UX Report (the CrUX definition on pauld.fr summarizes the methodology). I have audited sites at Lighthouse 95 failing on CrUX, and the reverse: CHANEL is a known case, with a Lighthouse score around 51 but CWV in the green.

The metric that matters is the 75th percentile (P75) of mobile LCP on your priority URLs, month over month. That is the metric I target in engagements, and the one I commit to in the action plan. A serious engagement cross-checks three sources:

Optimizing on a single source means optimizing blind or celebrating too early. A real regression seen in RUM can be invisible in lab, and a lab win may not reflect in field if the problem's origin lies elsewhere (traffic mix, device distribution, connection).

Field notes: the levers that actually gain LCP

A recent case, April Moto, a high-traffic WordPress site. Mobile Lighthouse score red, LCP at 3.2 s at P75, INP acceptable but fragile. The internal team had already tried two optimization passes without noticeable success.

The audit revealed three combined causes: LCP image as an 800 KB JPEG with no preload and no fetchpriority, page cache fragmented by tracking URL parameters multiplying stored variants, and a tag manager loading 14 synchronous scripts in the <head>. Three weeks of implementation, LCP down to 1.3 s (59% improvement), CWV in the green, mobile bounce rate down two points over the following 30 days.

◆Field notes·8 years · 35+ clients

My field note — E-commerce and cache by postal code.

I audited a WordPress e-commerce site where the browser cache was nearly inoperative because shipping costs were computed from the homepage using the entered postal code. Result: a distinct cache per postal code, cache hit ratio at 40% instead of the expected 88%. The fix was to defer the shipping computation to a post-load AJAX call, keeping the homepage fully cacheable. LCP unchanged on the homepage, but TTFB divided by 3 across all internal pages and CDN cache back to being effective.

Another case, a Drupal enterprise site. TTFB at 1.1 s on average, identified cause: Drupal modules reloading JS, CSS, and running thousands of SQL lines on every page, while the data was only used on a single view. Refactor spread over three sprints, TTFB down to 180 ms, with direct impact on mobile LCP and on the TTFB metrics in Search Console.

A more recent one, on a React SPA with strong SEO stakes. Mobile LCP drifting at 4.1 s because the hero component waited for the end of hydration before rendering the image. Targeted migration to React Server Components on the layout and hero (Next.js App Router), with no global application refactor. LCP down to 1.7 s in three weeks, without touching the rest of the codebase.

Stacks and tools I cover

On the front side, I work on React (Next.js Pages Router and App Router, RSC), Astro, Angular, Vue, WordPress, and Shopify Liquid. The most frequent stacks are React and WordPress, but the optimization patterns differ radically by stack: on React, INP and bundle size dominate; on WordPress, it is the LCP image, cache, and parasitic plugin cleanup.

On the back and e-commerce side: WordPress/WooCommerce, Drupal, Magento, Salesforce Commerce Cloud, PrestaShop. On Magento and SFCC, the personalization layer is almost always the first workstream: several megabytes of JavaScript injected for conditional content that could be rendered server-side. On WooCommerce, the combination of caching plugin and CDN cache must be finely orchestrated to avoid collisions.

On the monitoring side: SpeedCurve, DebugBear, Datadog RUM, Dynatrace, Akamai mPulse depending on the stack in place. The tool matters less than what you do with it: a RUM without performance budgets in CI and without regression alerts stays a decorative dashboard. The discipline that guarantees a score holds for six months is enforced budgets on every merge request.

What matters most is not the stack itself, it is the target metric. A green LCP on 4G mobile at P75, an INP stabilized below 200 ms, a CLS close to zero. Those are the engagement objectives, reached with the stack in place rather than by forcing a technical rewrite nobody budgeted for.

How we get started

The first scoping conversation happens within 48 hours. We clarify the stakes (SEO, conversion, acquisition cost), the current state of Core Web Vitals on Search Console, and the technical constraints (code freeze, release cycles, internal team autonomy).

The full initial audit is delivered within 10 business days. It includes the metric-by-metric quantified report (LCP, INP, CLS, TTFB at P75 and P95), the prioritized action plan with gain and effort estimates per lever, and a presentation to your technical team and leadership. From there, the trajectory is chosen together: a 3-week express sprint before a campaign, a rebuild spread over several quarters, or monitoring alone if the foundation is already sound.

Every engagement is measured before and after on RUM data, documented for your team, and leaves in place the performance budgets that let the score hold over time. Renewal is discussed quarter by quarter, only if it delivers measurable value.

Sources

• Deloitte, Google (March 2020). Milliseconds Make Millions: How mobile page speed drives consumer engagement. Study on 30 retail, travel, luxury and lead-gen brands (Europe and US), over 4 weeks of field data. Correlation between mobile speed, funnel progression, average order value and bounce rate. Deloitte Ireland — dedicated page
• Google Chrome team. Chrome UX Report — methodology. Official documentation on Core Web Vitals P75 aggregation over a 28-day rolling window from opt-in Chrome users. developer.chrome.com/docs/crux
• Google Search Central (June 2021). Understanding Page Experience in Google Search results. Announcement and documentation of the Page Experience ranking signal, of which Core Web Vitals are a component. developers.google.com/search/docs/appearance/page-experience
• web.dev, Google Chrome team. Core Web Vitals — definitions, thresholds, fixes. Official technical reference for LCP, INP, CLS metrics and their P75 thresholds (2.5 s / 200 ms / 0.1). web.dev/articles/vitals