Traffic from Google and Yandex but no calls: mapping the drop-off points
Traffic and leads are different metrics. Five bottlenecks sit between them, and usually one specific bottleneck is doing the damage.
First check whether the traffic is relevant at all
Before fixing the site, open the query report in Google Search Console and Yandex.Webmaster. A service page often collects informational traffic like 'how much does it cost' or 'DIY' — those people never call, and redesigning buttons changes nothing.
The second filter is geography. Split sessions by city in Metrica and GA4: if half the traffic is outside Uzbekistan while you only serve Tashkent, low conversion is arithmetic, not a design flaw.
A typical picture: 3,000 visits a month, of which 400 are genuinely commercial. Twelve leads. That is 3 %, not 0.4 % — and the conversation changes immediately.
Speed on mobile 3G/4G
Most traffic in Uzbekistan is mobile and connection quality outside central Tashkent is noticeably worse. Test in PageSpeed Insights on Mobile and watch LCP rather than the overall score — above three or four seconds, part of the audience leaves before seeing anything.
Yandex.Metrica has a page load time report. Compare bounce share for sessions under three seconds against those over five — the gap is usually twofold and explains part of the shortfall with no psychology involved.
The most common local causes: uncompressed two to three megabyte photos in the portfolio, a full slider plugin loaded for a single banner, and fonts pulled from an external CDN that responds slowly in the regions.
The first screen fails to confirm relevance
Someone arrives from 'refrigerator repair at home' and sees the headline 'Comprehensive solutions for your comfort'. There is no match between query and screen, so they return to the results page. This is the cheapest loss to fix and the most frequent.
The rule is simple: the hero headline must literally contain the service wording and the city, and the subheading must answer the main risk question — call-out, timeline, warranty. With multiple landing pages, the headline should change per query group.
Metrica's session recordings show these visits instantly: 10–15 % scroll depth, four to eight seconds, no clicks. If you see many of them, the problem is on the hero and there is no point looking further down the funnel.
The contact channel is not the one people use
In Uzbekistan calls and Telegram cover the vast majority of enquiries, while 'leave a request, we will call you back' reads as a deferred promise. If the phone number is not sticky and tappable on mobile, you lose enquiries that were already decided.
The minimum mobile set: a tappable number, a Telegram button and a call button in a fixed bottom bar. Email barely works here — it is largely a B2B channel and even there not always.
Set separate goals in Metrica for tel: clicks, t.me link clicks and form submissions. It often turns out the form supplies 20 % of enquiries while phone and Telegram supply the other 80 % that nobody was counting.
The form asks too much or silently breaks
Every extra field is drop-off. Name and phone are enough for first contact; everything else is established in conversation. Email, preferred time and comment fields reduce submissions with no benefit to the sales team.
Check the phone mask separately: it must accept +998 XX XXX XX XX and not break when someone pastes a number with spaces from their contacts. This mundane bug quietly eats leads for months.
Then check delivery itself: send a test lead and confirm it arrives. Emails land in spam regularly — duplicating each lead to a Telegram bot and writing it to a database is far safer.
The lead arrived but was never worked
The last drop-off point is off the site. Reply after two hours and the person has already messaged two competitors on Telegram and is talking to whoever answered first. The local norm is a response within 10–15 minutes during working hours.
Read the first messages your team sends. Replying 'hello, how can I help?' to a lead that already contains a filled form restarts the dialogue from zero and gives another reason to disengage.
Before rebuilding the site, count: how many enquiries came in, how many were answered within 15 minutes, how many reached a quote. Often the site is fine and everything is lost at the response stage.