Web Performance4 min read
Core Web Vitals for UAE e-commerce
The 2026 thresholds, why a Dubai storefront served from Europe starts half a second behind, how Arabic fonts and instalment widgets break LCP and INP, and the diagnostic order we use to fix a failing product page.
Danial AhmedFounder & engineering manager
On this page
A UAE storefront can pass every Lighthouse check in a Dubai office and still fail Core Web Vitals in the field, because the score that counts is the 75th percentile of real visits on real devices over 28 days — and a meaningful share of those visits come from mid-range Android handsets on networks that never touch your test environment. This is the diagnostic sequence we use on e-commerce builds, in the order that actually finds the problem.
The thresholds, and the number that matters
Three metrics, each with a good and a poor boundary, evaluated at the 75th percentile and reported separately for mobile and desktop.
| Metric | Good | Needs work | Poor |
|---|---|---|---|
| Largest Contentful Paint (LCP) | ≤ 2.5s | 2.5–4.0s | > 4.0s |
| Interaction to Next Paint (INP) | ≤ 200ms | 200–500ms | > 500ms |
| Cumulative Layout Shift (CLS) | ≤ 0.1 | 0.1–0.25 | > 0.25 |
A useful diagnostic that is not itself a Core Web Vital: Time to First Byte above roughly 800ms means the problem is server or network, and no amount of front-end work will fix it. Interaction to Next Paint replaced First Input Delay in 2024 and is far harder to pass, because it measures every interaction across the visit, not just the first one.
Where your site is served from is a Core Web Vital decision
Round-trip time from Dubai to a European origin is typically 110–140ms, and to the US east coast 230–260ms. A cold request needs several round trips before the first byte moves — DNS, TCP, TLS, then the request itself — so an origin in Ireland can spend half a second on physics before your application does any work.
The options that remove it: AWS me-central-1 (UAE) or Azure UAE North for the origin, or a CDN with a Dubai point of presence serving statically rendered HTML so that the origin is not in the critical path at all. Google Cloud has no UAE region at the time of writing; the nearest are in Qatar and Saudi Arabia. If you cannot move the origin, pre-render and cache at the edge — it achieves most of the same result.
The four things that break LCP on UAE storefronts
LCP decomposes into four parts, and knowing which one is large tells you what to fix.
- Time to first byte. Usually origin location or an uncached database query on the product page.
- Resource load delay. The hero or product image is discovered late because it sits behind JavaScript, a carousel, or a lazy-loading attribute applied indiscriminately. Never lazy-load the LCP image; preload it.
- Resource load duration. Uncompressed or oversized product photography. Serve AVIF or WebP with explicit dimensions and correct
sizes, and cap the delivered resolution. - Element render delay. A font or a blocking script holds the paint. Preload one weight of your primary face and let the rest swap.
Arabic-language storefronts carry an extra font cost: Arabic families are large, contextual shaping limits how aggressively you can subset, and loading a display weight and a body weight for both scripts is easily 300KB before a product image appears. Use a variable Arabic face, preload exactly one file, and set font-display: swap.
Third-party scripts are why INP fails, and BNPL widgets are usually first
Interaction to Next Paint is the sum of input delay, processing time and presentation delay — and on UAE storefronts the input delay is almost always a third-party script occupying the main thread. The usual suspects, in order of damage: tag managers with a dozen tags, live chat, heatmap and session recording tools, and instalment-plan widgets.
The instalment widget is the interesting one, because it is commercially non-negotiable and technically avoidable. A Tabby or Tamara badge showing "4 payments of AED 187.25" does not need a script to compute a division. Render the instalment price server-side from the product price, reserve its exact box, and load the vendor script only where the interaction genuinely requires it — the checkout.
Cumulative Layout Shift: the UAE-specific causes
CLS is mostly solved by reserving space, but three causes recur locally. Currency and instalment badges injected after load, on a product card grid, shifting every row. Right-to-left stylesheets applied after first paint, mirroring the entire layout in view. And cash-on-delivery fee lines appearing in the order summary once a payment method is chosen.
All three have the same fix: render the final layout on the server, reserve the box, and never let a client-side script decide the height of something above the fold.
Measure the field, not the lab
Lighthouse is a lab tool and does not report INP at all in the way the field does. Use the Chrome User Experience Report for the p75 numbers, and add real-user monitoring with the web-vitals library so you can segment by device, connection and page template.
Segment your own data before you optimise. A storefront that passes overall can be failing badly on the product listing page on mid-range Android — which, for most UAE retailers, is where the revenue is. Fix the template with the worst p75 and the most traffic, in that order, and re-measure after 28 days rather than after the deploy.