Real-user monitoring

The speed your visitors actually get.

One npm package reports LCP, INP, CLS, FCP and TTFB from real visits. The report names the slow pages, components and visitors, and the deploy that made it worse.

Monitor real users free Performance testing

Capabilities
Use cases
Benefits
Works with
FAQ

StorefrontSuitesEnvironments & domainsRequests & loadReal usersMonitoring & alertsproduction1 h6 h24 h72 h7 d Receiving data

LCPiGoodLargest Contentful Paint2.1 sp75 · 1,240 visits

INPiNeeds workInteraction to Next Paint240 msp75 · 1,240 visits

CLSiGoodCumulative Layout Shift0.04p75 · 1,240 visits

FCPiGoodFirst Contentful Paint1.4 sp75 · 1,240 visits

TTFBiGoodTime to First Byte620 msp75 · 1,240 visits

iINP · Interaction to Next PaintAfter a click, a tap or a key press, how long until the screen shows the response.Good ≤ 200 msPoor > 500 msWhen it is badHeavy JavaScript on the main thread, long React renders, or long lists rendered at once.See Slowest components →

Slow pagesLCP p75 · events

/search3.4 s860

/orders/:id2.8 s312

/product/:id2.4 s2,140

/checkout2.3 s540

/shoes2.0 s1,310

/1.8 s3,020

/orders/123 → /orders/:id

Slowest componentsp95 render

<ProductGrid>64 ms2,300 renders

<CartDrawer>41 ms1,120 renders

<ReviewList>22 ms880 renders

<SearchBox>18 ms3,410 renders

One frame is 16 ms

Frontend errors4.1 per 1,000

Cannot read properties of undefined (reading 'price')41 times

ChunkLoadError: Loading chunk 812 failed12 times

Trendsproduction1 h6 h24 h72 h7 dMark a deploy

LCP p752.1 sGoodone point per 6 h

3.0 s2.5 s2.0 s1.5 s

good ≤ 2.5 sv2.4.0v2.4.1v2.4.2

v2.4.1Got worseLCP +18 % vs the 24 h before2.1 s → 2.5 s

6 Oct8 Oct10 Oct12 Oct

Deployseach against the 24 h before it

v2.4.211 Oct, 06:10 · from CILCP 2.5 s → 2.1 s · −16 %Got better

v2.4.19 Oct, 06:05 · from CILCP 2.1 s → 2.5 s · +18 %Got worse

v2.4.07 Oct, 12:30 · marked by handLCP +1 %No change

Better or worse only past 10 %, with 5 samples or more on each side

INP240 msp75

CLS0.04p75

FCP1.4 sp75

TTFB620 msp75

Errors4.1per 1,000

deploys marked · sample project

Install the SDK on your siteproductionAny websiteReactNext.js

Packages

@nacew/performance-webAny site — Web Vitals, pages, errors

@nacew/performance-reactAdds page and component render timings

@nacew/performance-nextThe wrapper for Next.js

Events kept30 days · 1 to 90

1Install the packagenpm install @nacew/performance-webCopy2Start it once, on every pagethis environment’s key, filled inproduction3Check it arrivedCheck now Receiving data

one npm package

Measured from real visitors: Web Vitals at p75 from the visits your site actually got, each rated Good, Needs work or Poor.
One npm package: For any site, for React, or for Next.js. The install steps arrive with your environment’s key filled in.
Bots out, IP never stored: Bots, headless browsers and Lighthouse runs are dropped before anything is counted.
Deploys marked: Each deploy is compared with the 24 hours before it, so the one that made things worse is named.

Core capabilities

From the number to the cause

Pick a capability to see it drawn from a sample project.

LCP, INP, CLS, FCP and TTFB, each Good, Needs work or Poor. Every metric carries an ⓘ: what it is, where good and poor start, and what usually makes it slow.

StorefrontSuitesEnvironments & domainsRequests & loadReal usersMonitoring & alertsproduction1 h6 h24 h72 h7 d Receiving data

LCPiGoodLargest Contentful Paint2.1 sp75 · 1,240 visits

INPiNeeds workInteraction to Next Paint240 msp75 · 1,240 visits

CLSiGoodCumulative Layout Shift0.04p75 · 1,240 visits

FCPiGoodFirst Contentful Paint1.4 sp75 · 1,240 visits

TTFBiGoodTime to First Byte620 msp75 · 1,240 visits

Slow pagesLCP p75 · events

/search3.4 s860

/orders/:id2.8 s312

/product/:id2.4 s2,140

/checkout2.3 s540

/shoes2.0 s1,310

/1.8 s3,020

/orders/123 → /orders/:id

Slowest componentsp95 render

<ProductGrid>64 ms2,300 renders

<CartDrawer>41 ms1,120 renders

<ReviewList>22 ms880 renders

<SearchBox>18 ms3,410 renders

One frame is 16 ms

Frontend errors4.1 per 1,000

Cannot read properties of undefined (reading 'price')41 times

ChunkLoadError: Loading chunk 812 failed12 times

Web Vitals · sample project

Common use cases

Where real visits disagree with the lab

After a release, for one group of visitors, on one page.

Trendsproduction1 h6 h24 h72 h7 dMark a deploy

LCP p752.1 sGoodone point per 6 h

3.0 s2.5 s2.0 s1.5 s

good ≤ 2.5 sv2.4.0v2.4.1v2.4.2

v2.4.1Got worseLCP +18 % vs the 24 h before2.1 s → 2.5 s

6 Oct8 Oct10 Oct12 Oct

Deployseach against the 24 h before it

v2.4.211 Oct, 06:10 · from CILCP 2.5 s → 2.1 s · −16 %Got better

v2.4.19 Oct, 06:05 · from CILCP 2.1 s → 2.5 s · +18 %Got worse

v2.4.07 Oct, 12:30 · marked by handLCP +1 %No change

Better or worse only past 10 %, with 5 samples or more on each side

INP240 msp75

CLS0.04p75

FCP1.4 sp75

TTFB620 msp75

Errors4.1per 1,000

seven days of LCP · sample project

After a release

Find the deploy that made LCP worse

Every deploy is compared with the 24 hours before it on LCP, INP, CLS and error rate. In the sample project, v2.4.1 made LCP 18 % worse, and v2.4.2 took it back.

Who has it worstproduction · 24 hPick a row to filter the whole report

ShowingCountry: Brazil ✕LCP p754.3 sINP p75410 msCLS p750.08Visits136against LCP 2.1 s overall

DeviceLCPINPCLSShare

Mobile2.3 s250 ms0.0561 %

Desktop1.5 s120 ms0.0234 %

Tablet1.9 s180 ms0.045 %

BrowserLCPINPCLSShare

Chrome2.0 s230 ms0.0455 %

Safari2.2 s210 ms0.0327 %

Samsung Internet2.9 s380 ms0.068 %

Firefox1.8 s160 ms0.027 %

Edge1.9 s170 ms0.033 %

CountryLCPINPCLSShare

Germany1.8 s190 ms0.0334 %

United States2.0 s230 ms0.0429 %

United Kingdom1.9 s200 ms0.0418 %

BrazilPoor4.3 s410 ms0.0811 %

Iceland–––<1 %3 visits · no verdict

Dropped before counting✕Bots✕Headless browsers✕Lighthouse, our PageSpeed runs included✕The IP — never storedNo verdict under 5 samples

by device, browser and country · sample project

Across your audience

Visitors in one country have it worst

An overall p75 can look fine while one group waits. In the sample project LCP is 2.1 s overall and 4.3 s in Brazil — pick that row and every chart on the report follows it.

What to fix firstBuilt from measured dataproduction · 24 h✦ Order and explain with AIoptional

1VisitorsBrazil visitors have it worstLCP p75 4.3 s for Brazil, against 2.1 s overall11 % of traffic

2Page/search loads slowlyLCP p75 3.4 s on this page860 events here412 KB of JavaScriptAnalyse this page with PageSpeed ↗Watch people use this page →

3Component<ProductGrid> renders slowlyp95 render 64 ms over 2,300 renders38 renders of 50 ms or moreseen on /search

4Web VitalInteraction to Next Paint (INP)INP p75 is 240 ms — needs work1,240 real visits measured

5ErrorFrontend error: Cannot read properties of undefined (reading 'price')41 times4.1 per 1,000 events

AI can reorder and explain these items. It cannot add one.

UsabilityNew studyOpened from “Watch people use this page”

Live URLhttps://shop.example.com/search

First missionFind trail shoes in your size and open one

Who tests itAI testersPeople, by linkCreate study

49PageSpeed · /searchMobile · LCP 5.1 s in the lab↗

what to fix first · sample project

Into usability

From a slow page to people using it

“Watch people use this page” opens a new usability study on that URL. The page the numbers point at is the one testers try next.

Human testersOne share link, consent built in — for the people who wait for that page.

Benefits

Measured, then explained

p75of real visits, not a lab run
What visitors measured
Three visits in four were this fast or faster — the figure the Good, Needs work and Poor thresholds are set against.
+18 %Got worsev2.4.1 · sample project
A deploy named, not suspected
Better or worse only past a 10 % change with at least 5 samples on each side. Anything less says no change, or not enough data.
Measured itemsAI order — optional
AI that cannot invent
“Order and explain with AI” can only reorder the measured items it is given. It never adds a page, and its answer is saved.

Works with what you have

Three packages and your CI

Already published on npm. Pick the one that matches the site.

@nacew/performance-web

For any website. Web Vitals, pages and frontend errors.

@nacew/performance-react

Adds page and component render timings for React apps.

@nacew/performance-next

The wrapper for Next.js apps.

Your CI

Marks each deploy with one POST and a deploy token — or mark it by hand.

Discover more

The rest of the pillar, and beside it

FAQ

What a front-end lead asks first

What does the SDK collect?

LCP, INP, CLS, FCP and TTFB from each visit, the page it happened on, frontend errors, and the visitor’s device, browser and country — plus page and component render timings with the React package. Ids in URLs are grouped, so /orders/123 counts as /orders/:id. Events are kept for 1 to 90 days, 30 by default.

Do you store visitors’ IP addresses?

No. The IP is never stored. Bots, headless browsers and Lighthouse — including our own PageSpeed runs — are dropped before anything is counted.

What counts as a regression?

Each deploy is compared with the 24 hours before it on LCP, INP, CLS and error rate. It is called better or worse only past a 10 % change with at least 5 samples on each side.

Can the AI make up a fix?

No. It is given the measured items and can only reorder and explain them; anything it returns that was not on the list is dropped. It is optional, and its answer is saved.

The tester app

Test from your pocket

The uTestMe tester app runs a study on a phone. Taps and screens recorded, consent first, results in the same view.

Coming soon onApp StoreGet it onGoogle Play

Scan to install

Tap tracking
Screen recording
Consent first

Test code · K7F2-9QA