digi.ai.in
← digi.ai.in / blog
2026-09-198 min readPerformance

The Version of "Done" That Actually Matters to Users

A site your vendor calls "done" can still fail the only test that matters: does it feel fast and stable to a real visitor.

Your development team can close every ticket, pass every QA test, and deploy the site — and it can still feel broken to the person visiting it on a train in Nagpur. "Done" for a developer and "done" for a user are two different things, and the gap between them is where most website disappointments come from.

Two Definitions of "Done," and Why They Diverge

Agile teams work off a Definition of Done (DoD) — a checklist a feature must pass before release. The Scrum Guide describes it as a formal description of the state of the increment, met the moment a backlog item satisfies it, at which point "an increment is born" Future Processing. That checklist is real and necessary. It is also entirely internal.

The distinction that matters here is between internal and external quality. As one engineering analysis puts it, external quality is how users perceive the software — whether it's easy to use, works as intended, and does things quickly — while internal quality, like code maintainability and readability, is something developers care about but users never see directly Matti Lehtinen.

A vendor can hand over a site that is 100% "done" by their internal checklist — features built, bugs closed, staging approved — while a visitor on a mid-range Android phone experiences something that stutters, jumps around as it loads, and takes a beat too long to respond to a tap. Both things are true at once. That's the gap this post is about.

Google's Answer: A Definition of "Done" Built Around the User

Google has effectively published its own external Definition of Done, and it calls it Core Web Vitals. These are a set of three metrics focused squarely on user experience — perceived load time, visual stability, and responsiveness to input — rather than on anything a developer sees in their build logs web.dev. The point is explicit: Core Web Vitals are user-centric metrics, in contrast to lab-based diagnostic tools like Lighthouse, which test in a controlled environment rather than on a real visitor's real device and network web.dev.

The three metrics and their "good" thresholds, per Core Web Vitals reference data corewebvitals.io:

Metric What it measures "Good" threshold
Largest Contentful Paint (LCP) Loading speed Under 2.5 seconds
Interaction to Next Paint (INP) Responsiveness Under 200 milliseconds
Cumulative Layout Shift (CLS) Visual stability Under 0.1

Two details matter more than the thresholds themselves. First, Google doesn't grade a single page load — it grades the aggregate experience, using the 75th percentile of all page views, so a site only counts as "good" if at least 75 percent of real visits meet the threshold web.dev. Second, this data comes from real Chrome users in the field (the CrUX report), not a lab test run once from an office in Pune Google Search Central.

That second point is the whole argument. If your vendor's "done" is based on a Lighthouse score run once on office Wi-Fi, it is measuring the wrong thing. Core Web Vitals is measuring what actually happened to actual visitors.

Why "Feels Done" Beats "Is Done"

There's a psychological layer underneath the metrics. MDN's performance documentation distinguishes perceived performance — a subjective sense of how fast a site seems — from actual speed, and argues perceived performance is harder to quantify but arguably more important, because while a site's actual speed may be fixed, how fast it feels can often still be improved MDN. Elsewhere MDN makes the same point more bluntly: how fast a website seems to the user has a greater impact on user experience than how fast the website actually is MDN.

A concrete illustration: a page might take ten seconds to fully load a large background video, yet if the headline and navigation appear almost instantly, it creates good perceived performance ecorn.agency. Technically that page isn't "done" loading for several more seconds. To the user, it already is.

This cuts both ways for anyone commissioning a build. A site can be technically complete and still feel slow if the wrong things load first. And a site that is still fetching data in the background can feel entirely finished if the visible, interactive parts appear fast.

What It Costs to Get This Wrong

The commercial stakes are documented, if not perfectly consistent across studies. Google's own research found that 53% of users will abandon a page if it takes more than three seconds to load ecorn.agency. A separate Google neural-network study of 11 million mobile landing pages found that as load time goes from one second to ten seconds, the probability of a mobile visitor bouncing rises 123% Google.

Worth flagging: these numbers aren't fully standardized. Different studies, sample sizes, and years produce bounce-probability figures anywhere from 32% to 123% for similar load-time increases, because methodologies vary Google. Treat any single stat as directional, not exact — the direction is consistently the same: slower costs visitors.

There's also a question of how much faster is fast enough to notice. Usability research cited by SitePoint suggests a performance improvement must be at least 20% before users perceive it at all SitePoint. That's a relative bar. Google's Core Web Vitals, by contrast, set fixed absolute thresholds — 2.5 seconds is 2.5 seconds regardless of where you started. A site can clear Google's absolute bar while still feeling, subjectively, no faster than it did last year. Both measures are valid; they're just answering different questions.

Why This Gap Is Wider in India

Network conditions in India make the developer/user "done" gap larger than in markets where testing and real-world use happen on similar connections.

TRAI's 2024–25 report shows India's telecom subscriber base at 1.2 billion as of March 2025, with overall teledensity at 85.04% — but rural teledensity trailing badly at 59.06%, against 131.45% in urban areas Light Reading. Internet subscriber growth has also flattened, rising just 1.54% year-on-year, from 954.4 million to 969.10 million subscribers between March 2024 and March 2025 Light Reading.

There's real progress too. A government release notes India's global ranking in average internet download speed improved by 114 ranks in a decade, from 130th in 2014 to 16th, with an average download speed of 105.85 Mbps, and more than 95 percent of villages now have 3G or 4G coverage PIB. Overall penetration is now roughly 7 in 10 Indians online as of 2025 Wikipedia.

But averages hide the gap that matters for a specific site. One secondary source ran Google's "Test My Site" tool on an identical landing page and found mobile load time was 2.8 seconds from India versus 1.7 seconds from the UK Ignite Visibility. That's not a lab artifact — it's a real difference in what "done" means depending on where the visitor is sitting. A site that comfortably clears Google's 2.5-second LCP threshold when tested from a Pune office on fibre can miss it for a meaningful share of real Indian visitors on mobile networks, especially outside metro areas.

A Practical Checklist for Commissioning Work

Before you sign off a project as "done," ask your vendor to show you, not tell you, the following:

  1. Field data, not just lab scores. Ask for CrUX or real-user-monitoring numbers, not a single Lighthouse run from the office web.dev.
  2. The 75th percentile, not the best case. A site is only "good" if at least 75% of real visits meet the threshold — ask what percentile your vendor is reporting against web.dev.
  3. Testing on a throttled mobile connection, not just office Wi-Fi, given the documented India-vs-UK load time gap on identical pages Ignite Visibility.
  4. What loads first. Ask whether the headline and navigation appear before secondary content like video or background imagery finishes loading ecorn.agency.
  5. INP, not the retired FID metric. INP formally replaced First Input Delay as a Core Web Vital in March 2024 — some guides and even reference pages haven't caught up, so confirm your vendor is measuring the current metric Wikipedia.
  6. A number for "how much faster," not just "faster." Since a 20% improvement is roughly the threshold for users to notice a change at all, ask what percentage improvement a proposed fix will actually deliver SitePoint.

Questions People Also Ask

Is a fast Lighthouse score the same as good real-world performance?

No. Lighthouse is a lab-based diagnostic tool run in a controlled environment. Core Web Vitals are user-centric metrics based on real Chrome user data in the field, and the two can disagree web.dev.

What Core Web Vitals thresholds should a new site meet before launch?

LCP under 2.5 seconds, INP under 200 milliseconds, and CLS under 0.1, measured at the 75th percentile of real visits rather than a single best-case test corewebvitals.io web.dev.

Why does the same site feel slower in India than in the UK or US?

Rural teledensity in India is still 59.06% against 131.45% in urban areas Light Reading, and one comparative test found identical landing pages loading in 2.8 seconds on mobile from India versus 1.7 seconds from the UK Ignite Visibility. Network conditions, not just site code, drive the difference.

Written by the studio behind digi.ai.in. Questions? hi@digi.ai.in or WhatsApp +91 98605 96905
WhatsApp us