Website performance audit: a method for agencies
How to run a website performance audit: which pages to pick, metrics, tools, root causes, prioritization, and how to turn it into a lever for winning clients.

In short
- A performance audit starts from a goal and about ten pages that matter, not from a tool run on the homepage alone.
- The three core metrics are LCP (loading), INP (responsiveness) and CLS (visual stability), read from field data, not just from the lab score.
- Causes are tracked down in order: rendering, server, heavy resources, third-party scripts. Fixing before diagnosing wastes days.
- The useful deliverable is a plan ranked by impact and effort, with before/after measurements on the same pages.
- Auditing every prospect by hand doesn't scale: that is where an automated audit of local small business (TPE/PME) websites changes your prospecting.
A website performance audit measures how a site loads, responds and behaves for real visitors, then ranks the causes of slowness by expected gain and by effort. For an agency or a freelancer, it pays off twice: it gives a client a reliable diagnosis, and it opens the conversation with a prospect whose website is already hurting their business. This article lays out the full method, step by step, then shows how to turn it into prospecting without spending your days on it.
What a performance audit covers
Google's web.dev describes a performance audit as an objective review of what works and what can be improved. It sets a baseline, surfaces quick wins and provides metrics to prioritize the effort, then hard evidence once the improvements are in.
In practice, you look at:
- Core Web Vitals page by page, on mobile and on desktop;
- server response time;
- page weight and request count, including the share of third-party scripts;
- the gap between lab data and field data;
- the journeys that bring in money: contact form, appointment booking, checkout.
Here, "performance" means perceived speed, not an overall verdict on the site. A broader website audit adds SEO, usability and conversion. If your client wants the full picture, performance is one part of it, and often the easiest to put numbers on.
Why audit before fixing
An audit prevents pointless fixes. A fix that saves 40 ms doesn't justify three days of development. It also helps you decide between technical and editorial causes: sometimes the bottleneck is a ten-image carousel that marketing asked for, not the code. Finally, it gives you a measured starting point to prove the gain afterwards, which secures the billing of an optimization project. These are the three reasons OpenMyDiv sets out in its method.

Preparing the audit: goals, pages and scope
An audit starts with a decision, not a tool. Before opening anything, answer three questions.
- What is the site for? Selling online, generating quote requests, booking appointments. Without that answer, the audit turns into generic criticism with no priorities.
- Which metrics serve as the baseline? Organic sessions, conversion rate, revenue by channel, depending on what the client already tracks. Record them on day one so you can compare at three and six months.
- Which pages really matter? About ten is enough: traffic entry points, conversion pages, the most used templates.
For a plumber's or a restaurant's brochure site, the list fits in a few lines: homepage, main service page, contact or booking page, one local page, one article that brings in traffic. For an online shop: homepage, category, product page, cart, checkout.
The classic trap, flagged by Incremys, is mixing incompatible measurements or generalizing from two or three URLs. If you only test the homepage, you will usually miss the template that is losing sales.
Also ask for read-only access to Search Console and Google Analytics. That is enough to get started, and nothing on the site changes while you collect data.
The metrics to collect
The three Core Web Vitals
Google defined three core metrics, detailed on web.dev: LCP measures perceived loading speed (when the main content appears), INP measures responsiveness when the user interacts, and CLS measures visual stability (elements that jump around).
The "good" thresholds are as follows, confirmed by Aspekt based on Google's documentation:
| Metric | What it measures | "Good" threshold |
|---|---|---|
| LCP | Main content rendering | 2.5 s or less |
| INP | Responsiveness to interactions | 200 ms or less |
| CLS | Layout shifts | 0.1 or less |
One clarification that prevents misreadings: web.dev rates a site at the 75th percentile of visits. An LCP of 2 seconds at that level is "good", an LCP of 5 seconds is "poor". In other words, a site isn't judged on its best visit but on what three quarters of its visitors experience.
For context, Seobooster's Core Web Vitals Observatory reports, for April 2026 and across the 145 most visited French websites, a median mobile LCP of 3,140 ms and a median INP of 295 ms. These are benchmarks for large sites, not realistic or sufficient targets for a client's shop: aim for the "good" thresholds.
Supporting metrics
Core Web Vitals don't tell the whole story. Add:
- TTFB (time to first byte), which tells you about the server and the CMS;
- total page weight and request count;
- the share of third-party scripts (chat, ad pixels, widgets);
- on the business side, conversion rate and drop-off rate at each step of the journey, as ZSocial points out.
A fast page that doesn't convert has a UX problem, not a performance problem. One ambiguous line of microcopy can sink a conversion rate without any speed metric showing it.
Tools and what they let you conclude
The web.dev workflow has three stages: assess site health starting with field data (PageSpeed Insights, Search Console), debug with Lighthouse and the Performance panel, then monitor to prevent regressions.
Two kinds of data that you must not confuse:
- Lab data (Lighthouse, DevTools) is reproducible: same test, same conditions. Use it to find what to fix.
- Field data (CrUX, Search Console) comes from real Chrome visitors, aggregated over 28 days. PageSpeed Insights shows it at the top of the report, above the lab test.
The common mistake, noted by Aspekt, is optimizing a lab score when the thresholds apply to field data. A PageSpeed score of 100 doesn't guarantee the site is "good" in CrUX. Conversely, if the field data is green, the site passes, whatever its lab score.
Common tools for an audit:
- PageSpeed Insights: first diagnosis, mobile and desktop separately;
- Search Console: Core Web Vitals report on real-user data;
- WebPageTest or GTmetrix: loading waterfall, tests from several locations;
- Chrome DevTools: main thread profiling, INP debugging;
- a crawler to explore the site the way a search engine does.
Free tools show the symptoms, but they don't rank the causes. That is the auditor's job: read, cross-check, prioritize.
Finding the causes: a funnel method
You can't fix what you haven't explained. Most performance projects fail because they start with the solution: install a caching plugin, remove some extensions, recompress the images, and the site stays slow because the cause was elsewhere.
1. Narrow down the scope
Does the slowness affect one page, one page type, mobile only, or the whole site? Centreon suggests comparing mobile and desktop, geographic regions, before and after a deployment. A slowdown isolated to mobile points to resource weight or render-blocking JavaScript. A site-wide problem points to the infrastructure or the CDN.
2. Read the TTFB
A TTFB above 600 to 800 ms usually signals a hosting issue, according to WebWallaby: saturated shared resources, no server-side cache, or a server located far from the audience. It's a rule of thumb, not a verdict: measure at several times of day before drawing conclusions.
3. Open the loading waterfall
In GTmetrix or WebPageTest, sort resources by weight and by blocking time. A 500 KB file loaded deferred gets in the way of rendering less than a 50 KB file that blocks it. That is why a list of "files that are too heavy" is never enough to set priorities.
4. Go through the usual suspects
The common causes of a slow WordPress site or online shop, listed by Tiaro:
- no caching (reload the page twice and compare);
- images weighing several MB;
- too many plugins and scripts;
- overloaded hosting;
- external embeds (videos, maps, widgets);
- no CDN and a cluttered database.
Each cause comes with a simple test. Without that test, you are delivering an opinion, not a diagnosis.
Ranking fixes and proving the gain
An audit that produces two hundred recommendations in no particular order is unusable. The most practical grid is the impact/effort matrix:
| Low effort | High effort | |
|---|---|---|
| High impact | Quick wins: do these first | Structural projects: schedule them |
| Low impact | Do if time allows | Drop them |
Each line of the plan should put a number on the expected gain and name an owner: the client, their web provider or you. Start with the URLs that combine the most traffic with the worst score, as Aspekt recommends. Typical fixes: WebP or AVIF images with explicit dimensions, deferring non-critical third-party scripts, server cache and a CDN for TTFB, reserving space for dynamic elements to eliminate layout shifts.
For follow-up, re-measure after each batch of fixes on the same benchmark pages. Because field data is a 28-day average, changes only show up after about four weeks: tell the client about this delay before you deliver, or they will judge your work too early.
A worked example of a useful report
Here is what the summary page of a deliverable audit looks like for a tradesperson with a WordPress site. The example is illustrative and the figures are made up.
| Finding | Page | Likely cause | Impact | Effort |
|---|---|---|---|---|
| Mobile LCP at 4.1 s | Homepage | 2 MB header image, no dimensions set | High | Low |
| CLS at 0.28 | Contact page | Cookie banner injected without reserved space | Medium | Low |
| TTFB at 1.1 s | All | Shared hosting, no caching | High | Medium |
| INP at 450 ms | Appointment booking | Blocking third-party widget script | High | Medium |
The client sees in thirty seconds what to do first: resize the image, reserve space for the banner, then deal with the hosting. That clarity sells the rest of the project.
From audit to meeting: the audit as a prospecting tool
For an agency, a performance audit has a second life: it is the most concrete argument for opening a conversation with a prospect. "Your website takes more than four seconds to display its main page on mobile" is a message a business owner understands, unlike a generic redesign pitch.
The problem is time. Building the list of plumbers in Lyon or restaurants in Bordeaux, finding the owner, verifying an email, auditing the site, writing a message that cites a real flaw: every prospect takes time, and that time is taken away from selling. As a result, many agencies fall back on word of mouth, freelance platforms or calls for tenders.
Good audit-led prospecting, which works even by hand, comes down to a few rules:
- Target one trade and one city, so the message speaks about an industry you know.
- Pick a single flaw, the most visible one and the most costly for the prospect's business.
- Show the evidence (the page, the measurement), then the consequence (mobile visitors leaving, missed calls).
- Suggest a short call, not a quote.
Automating prospecting with AutoLeads
This is exactly what we built AutoLeads to do at scale. You pick an industry and a city, for example "plumbers in Lyon". AutoLeads finds these businesses in real time on the web and Google Maps, then audits their online presence: outdated or slow website, weak SEO, Google reviews, social media. Each lead is enriched with the owner, a verified email and a phone number.
For each business, AutoLeads writes a sequence of emails and LinkedIn messages built on the real flaws of its website: you show up with a concrete diagnosis. Sends go out in weekly waves from the agency's own mailboxes, or are pushed to HubSpot. Replies land in a built-in CRM as tasks.
Plans are monthly, with credits consumed according to lead volume and the enrichments you turn on: Starter at €249 per month, Growth at €499 per month, Scale at €899 per month. The "We call for you" option takes care of outsourced appointment setting, with no published price.

Grezy's view
We see the same pattern with agency founders: they have the audit under control, but not the prospecting. They know how to diagnose an LCP or a TTFB, yet they wait for the phone to ring.
Our belief is simple: a web agency's best sales argument is a visible flaw on the prospect's website, not a brochure. That is why AutoLeads detects businesses with a visible need in real time on the web, with no static database or outdated open data, and enriches each lead with concrete audits: SEO, advertising, customer reviews.
We also believe the message should be personalized per prospect, not per segment. When the copy talks about the owner's actual website, the conversation starts from a finding rather than an agency pitch. Our email and LinkedIn sequences are designed to start from those flaws.
The goal isn't to send more messages, but to fill your calendar with meetings with local small businesses (TPE/PME) that genuinely need a new website, without depending on word of mouth.
Tom
Sources
- Performance audit, web.dev
- How the Core Web Vitals thresholds were defined, web.dev
- Core Web Vitals workflows, web.dev
- Website performance audit: measure and prioritize, OpenMyDiv
- Website performance audit, Incremys
- Core Web Vitals Observatory France 2026, Seobooster
- Core Web Vitals 2026: the thresholds, Aspekt
- Core Web Vitals 2026: audit and optimization, Aspekt
- PageSpeed Insights: definition and how to read the report, Abondance
- Web performance audit: a step-by-step guide, WebWallaby
- Slow website: the 4-step method, Centreon
- Slow website: the common causes, Tiaro
- Full digital audit: measuring UX, ZSocial
- 19 SEO KPIs to track, Ahrefs