People do not file a ticket that says the page felt slow. They leave, refresh, or go back to the spreadsheet. For a custom web application, speed is not a polish step at the end. It is whether the main action feels immediate.
You do not need a lab to see it. Open the app on a phone, on an ordinary connection, and do the one job a customer does most. If that path hesitates, the architecture or the page is doing too much work before it answers.
A feature nobody waits for is a feature nobody uses.
Amap Infotech
Product & Strategy
Google publishes simple targets for this. The main content should show quickly, the page should answer a tap quickly, and the layout should not jump while it loads. Miss those and the rest of the design is arguing with the clock.
The Three Numbers to Watch
Largest Contentful Paint is when the main block of the page appears. Interaction to Next Paint is how long the page takes to react after a tap or a click. Cumulative Layout Shift is whether buttons move while images and fonts arrive. Together they describe the feeling of the app better than a single “load time” average.
Measure them on the pages people actually use: sign-in, the list they open every morning, and the form they submit. A fast homepage does not help if the working screen is the slow one.
2.5sTarget for the main content to appear.
Google Core Web Vitals, LCP
200msTarget for the page to respond to input.
Google Core Web Vitals, INP
0.1Target for layout stability. Lower is calmer.
Google Core Web Vitals, CLS
What Usually Makes It Slow
The common causes are ordinary. The page downloads images that are far larger than the slot they sit in. The first screen waits on five API calls that could have been one. The database looks up a list without an index, so every extra row costs more time. A script from another site blocks the page before your own content can paint.
Fixes follow the cause. Send smaller images and only the ones on screen. Cache data that rarely changes. Put an index on the columns you filter by. Load third-party tags after the page is usable. And do the heavy report in a background job, then notify the person when it is ready.
Images sized for the screen, not the original camera file
One request for the first screen, not a chain of them
Indexes on the database fields people search and filter
Cache for content that does not change every click
Background work for exports, mail, and large reports
Make the Slow Case Visible
An average can hide the problem. If most requests take 200 milliseconds and one in twenty takes 4 seconds, the average still looks fine and those people still leave. Watch a high percentile, such as the slowest 5 percent, on the actions that matter.
Put that number where the team can see it after each release. A deploy that makes the main action slower should be as visible as a deploy that breaks a test. Speed regressions are bugs. They just do not throw an error.
3 secPast this, 53% of mobile visits are abandoned.
Google research on mobile speed
~60%Of web traffic is already on a phone, where delay is obvious.
Statcounter global stats
p95Watch the slow tail, not only the average response.
A useful production habit
Fast Enough to Finish
A custom web application is fast when the person can start the job and finish it without wondering if the click registered. Set the three targets, find the one slow step, and check it again after every release.