Web vitals

Install on by default in the browser SDK

The browser SDK measures Core Web Vitals on every page view with PerformanceObserver and sends one small report per view. The Web vitals page shows the p75 of each metric so you see what most visitors get, not the average.

1Nothing to install

Any page with the Blipit script reports vitals. Pass webVitals: false to init to turn it off. Reports are sent with sendBeacon when the page is hidden or unloaded, and on single-page-app route changes (CLS and INP only).

2Read the page

Per project, optionally per device type: p75 of LCP, INP, CLS, FCP and TTFB for the last 24 hours and 7 days, rated good, needs improvement or poor (LCP 2.5 s / 4 s, INP 200 / 500 ms, CLS 0.1 / 0.25, FCP 1.8 s / 3 s, TTFB 0.8 s / 1.8 s), a daily sparkline, and the 25 slowest pages by LCP or INP with sample counts.

3What p75 means

p75 is the 75th percentile: 3 out of 4 visits are faster than this number and 1 in 4 is slower. Google rates a page by its p75, so a page passes when at least three quarters of visits are good. It ignores a few extreme visits but still shows what slower phones and networks get.

4LCP: main content shown

Largest Contentful Paint is the time from the start of the visit until the biggest image or block of text in view has appeared, which is when the page looks ready to the visitor. It matters because people judge speed by it and leave pages that look empty. Good is 2.5 s or less, poor is more than 4 s.

  • Usually slow because of: a slow server answer (see TTFB), large hero images, render-blocking CSS and JavaScript, and web fonts that hold back the text.
  • To improve it: compress and resize the main image and load it early (fetchpriority="high", no lazy loading above the fold), cut or defer scripts, inline the critical CSS and serve files from a CDN.
  • In Odoo: the first load of /odoo downloads the whole web client, so every installed app and its assets add to LCP. Uninstall apps nobody uses, keep asset bundles cached (no ?debug=assets in production) and put a proxy with compression and caching in front.

web.dev/articles/lcp ↗

5INP: reacts to clicks and typing

Interaction to Next Paint is how long the page takes to show a response after a click, tap or key press, taken from the slowest interactions of the visit. It matters because a slow answer feels like a frozen page and people click twice or give up. Good is 200 ms or less, poor is more than 500 ms.

  • Usually slow because of: long JavaScript tasks on the main thread, heavy click and input handlers, large re-renders and third-party scripts competing for the same thread.
  • To improve it: break long tasks up, do less work in event handlers, show feedback first and do the heavy work after, and load third-party scripts later.
  • In Odoo: large list and kanban views, many computed fields on screen and custom widgets that work on every key press raise INP. Show fewer records and columns per page, and move heavy computation to the server.

web.dev/articles/inp ↗

6CLS: page jumping while loading

Cumulative Layout Shift adds up how much visible content moves unexpectedly while the page loads and is used, such as text jumping down when an image or banner appears. It matters because people lose their place and click the wrong thing. It is a score without a unit: good is 0.1 or less, poor is more than 0.25.

  • Usually caused by: images and videos without a width and height, ads, embeds and banners inserted above existing content, and web fonts that swap in at a different size.
  • To improve it: give every image and embed its size or an aspect ratio, reserve space for banners and late content, and use a fallback font of a similar size.
  • In Odoo: website pages with images without sizes, cookie bars and snippets that load late are the usual cause; in the backend, side panels and the chatter appearing after the form has loaded.

web.dev/articles/cls ↗

7FCP: first thing on screen

First Contentful Paint is the time until the first text or image appears, so the visitor sees that something is happening. It matters because a blank screen makes people think the site is broken. Good is 1.8 s or less, poor is more than 3 s.

  • Usually slow because of: a slow server answer, render-blocking CSS and JavaScript in the page head, and web fonts that hide the text until they load.
  • To improve it: speed up the server (TTFB), defer scripts or load them async, keep CSS small and let text show in a fallback font while web fonts load.
  • In Odoo: the web client shows nothing until its asset bundles are downloaded and parsed, so fewer installed apps and cached assets help FCP as much as LCP.

web.dev/articles/fcp ↗

8TTFB: server starts answering

Time to First Byte is the time from the request until the first byte of the page arrives: redirects, DNS, the connection and the time your server takes to build the page. It matters because nothing else can start before it. Good is 0.8 s or less, poor is more than 1.8 s.

  • Usually slow because of: slow database queries, too much work per request, no caching, chains of redirects and a server far from your visitors.
  • To improve it: cache pages and query results, fix slow queries, avoid redirects, and serve through a CDN or a server near your users.
  • In Odoo: server and database time show up here. Pulse shows which requests, buttons and screens are slow and which SQL queries take the time; also check the number of workers and database health.

web.dev/articles/ttfb ↗

Good to know

  • Paths are stored without the query string; the country comes from the IP. Bots are skipped.
  • Samples are kept 30 days. Percentiles are computed over the latest 200,000 samples in the window and the page says when that cap applied.
  • LCP and FCP are skipped for pages loaded in a background tab, and LCP stops at the first key or pointer press, as the web.dev definitions require.

Stuck? Email support@blipit.io. Keys and the project id for each project are under API keys in app.blipit.io.