We measured 28 Chennai business websites on a phone. Most of them fail.

We loaded 32 real Chennai business websites — dental clinics, salons, gyms and medical clinics — on a throttled mobile connection and measured what happened. 28 responded. The median one weighed 2.4 MB, made 122.5 requests, and took 4.8 seconds to show its main content.

Google treats 2.5 seconds as the edge of “good”. 22 of the 28 sites were slower than that. 16 of them took longer than four seconds. One took 43 seconds.

This data did not exist before, so here is all of it, along with the method, so anyone can check or repeat it.

How the measurement was done

Each site was loaded once in headless Chromium on 2026-08-17, throttled to Slow 4G — 1.6 Mbit/s down, 150ms RTT, 4x CPU slowdown, 390x844 viewport. That profile is close to a mid-range Android phone on Chennai mobile data — which is how nearly all of these sites are actually seen.

Sites were found through ordinary Google searches for the kind of business and area a customer would type, excluding directory listings like Justdial and Practo so that only real business websites were measured. The script is in the repository as scripts/audit-chennai-sites.mjs, and the raw results, including every site name, are published in the table below.

Four sites could not be measured: two timed out after 60 seconds and two failed DNS resolution at the time of testing. They are excluded from every figure rather than counted as failures, which is the conservative choice — including a site that cannot load inside a minute would make these numbers worse, not better.

The distribution

The shape is the finding. This is not a handful of bad sites dragging an otherwise healthy median — most of the distribution sits on the wrong side of the line, and the tail is extreme.

What made them slow

Weight, almost entirely. 22 of 28 sites were over 1 MB. 11 were over 3 MB. The heaviest was 10.9 MB — a single salon homepage, which is roughly the size of a feature film trailer, sent to someone who wanted to know the price of a haircut.

The request counts tell the same story: a median of 122.5 separate requests per page, with one site making 300. Each of those is a round trip on a connection where round trips are the expensive part.

Layout shift is worse than the averages suggest

5 sites exceeded a Cumulative Layout Shift of 0.1, which is the point at which content visibly jumps under the reader’s thumb. The worst recorded 1.018 — roughly ten times the threshold, meaning the page rearranges itself substantially after you have started reading. That is the mechanism behind tapping the wrong thing and giving up.

The finding that is not about speed

Only 6 of 28 sites showed a price anywhere on the homepage.

Broken down by trade, it is starker. Of the nine salon websites measured, none published a price. Two of twelve dental clinics did. This is the single most common and most expensive gap in the sample, and unlike page weight it costs nothing to fix.

Only 13 of 28 had a WhatsApp link, in a city where WhatsApp is how people actually contact a business. 20 had a tappable phone number, which means 8 sites expect someone on a phone to copy a number out by hand.

What they got right

The basics are mostly in place, which makes the rest more frustrating: 28 of 28 had a mobile viewport tag, 26 had a meta description, and 23 had structured data of some kind. Only one of the 28 was still served over plain HTTP.

These sites were not built carelessly. They were built by people who knew about viewport tags and schema markup, and who then added a slider, a video header, three font families and an animation library, because that is what a website is supposed to look like.

The comparison

The demo build this studio made for a fictional dental clinic paints in 0.14 seconds and transfers 16.1 KB. Against the median of this sample — 4.8s and 2.4 MB — it is roughly 34 times faster to paint and about 150 times lighter.

That gap is not skill. It is the absence of things: no slider, no video, no framework, no font loaded from another server, no analytics suite. Nearly everything that made the median site slow was optional.

What a business owner should take from this

1. Open your own site on your phone, on mobile data, and count. If your main content takes more than three seconds, you are in the majority — and losing the people who do not wait.
2. Publish your prices. Almost nobody in this sample does. It costs nothing and it is the first thing customers look for.
3. Add a WhatsApp link. Half the sample has no way to start a conversation in one tap.
4. Delete the slider. It is usually the single heaviest thing on the page and almost nobody sees the second slide.

Every site measured

This audit will be repeated with the same method in a year, so the comparison is like for like. If you run one of the businesses above and want the specific reason your site is slow, send the URL — you will get a straight answer whether or not you hire Brightleaf Studio. A build like the ones above costs ₹12,000, but the four fixes in this article are free and worth more than a rebuild.