Skip to content
Get Ranked On TopIllustrative case studySignal plan

A fictional business. The figures are a modelled scenario, not a client result.

Viungo House · Stone Town, Tanzania

How a Stone Town rooftop restaurant cut its mobile load from 11 seconds to under 2

Mobile LCP from 11.2 s to 1.8 s, then +133% organic bookings

9 months · Middle East & Africa · Signal plan

Rooftop restaurant & spice cooking school, 60 covers · Stone Town

Every bolt drawn · modelled mobile LCP

The challenge

A beautiful door that would not open.

A page-builder site with a 9 MB hero video and the menu in a PDF gave an 11-second mobile LCP. Every page failed Core Web Vitals and no dish was indexed.

Viungo House sells two things: a table on the roof at sunset and a seat in the afternoon spice class. Both are booked on phones, often on hotel Wi-Fi or a patchy 4G signal in the alleys of Stone Town. The page builder loaded an autoplaying 9 MB drone video before anything else, then a chat widget, four font files and a script bundle for effects nobody saw. The menu, the very thing guests search for, lived in a PDF that Google barely read.

Mobile load, before · simulated 4G, p75

Diagnosis

Four bolts on one door.

Hero media

9 MB

An autoplaying drone video of the dhow harbour, 9 MB, was the largest element on every page. On a mid-range phone it painted at 11.2 seconds, by which time the visitor had usually gone.

Bolt 1 of 4

Menu

0/46

All 46 dishes and the class programme sat in a 6 MB PDF behind a single button. Search Console showed the PDF indexed, but no dish or price appeared on any HTML page, so a search for “octopus curry stone town” found the competitors.

Bolt 2 of 4

Render-blocking

2.4 s

The page builder put 1.6 MB of JavaScript in the head: sliders, parallax and animation libraries on pages that used none of them. The main thread was blocked for 2.4 seconds before first paint.

Bolt 3 of 4

Third parties

38

38 third-party requests on the homepage: the chat widget alone pulled 1.2 MB and 19 requests on load, before anyone had opened it. Four font files added another 0.7 MB.

Bolt 4 of 4

Homepage weight, by source
  • Hero video 9.0 MB
  • Render-blocking script 1.6 MB
  • Webfonts 0.7 MB
  • Chat widget 1.2 MB
  • Everything else 0.9 MB

The strategy

Draw the bolts, then keep them out.

Built on our Signal plan: the technical engagement for sites where speed, crawling and rendering decide the outcome.

  1. Performance budget

    Before any fix, a written ceiling: mobile LCP under 2.5 seconds at p75 and under 1.5 MB per page. Every asset had to earn its place against it, and the budget still runs in CI on every deploy.

  2. Hero media rebuild

    The drone film became a still, served as responsive AVIF with explicit dimensions and high fetch priority. The video survives behind a play button and streams only when someone asks for it.

  3. Indexable HTML menu

    The PDF became a real page: 46 dishes in HTML with prices in TZS and USD, allergens, and Menu, MenuSection and MenuItem markup. The cooking classes got pages of their own, with Event schema for every session.

  4. Script & font discipline

    The page builder was retired for static templates, the fonts cut to one subset file and the chat widget swapped for a facade that loads on tap. Third-party requests on the homepage fell from 38 to 4.

Unbolt the door

Every bolt is an asset. The door opens as slowly as the page loads.

Draw the bolts in any order, then load the page. The door takes the modelled mobile LCP to open, in real seconds.

Modelled LCP

11.2 s

Performance budget

LCP budget 2.5 s11.2 s · Over by 8.7 s
Weight budget 1.5 MB13.4 MB · Over by 11.9 MB

4 bolts still in: LCP 11.2 s, 13.4 MB. Over budget.

Leave any single bolt in and it breaks the budget on its own: one ceiling, shared by every asset.

Model: the rebuilt page on its own (1.8 s) plus the cost of each remaining asset on a mid-range phone over 4G. Illustrative figures, played in real time.

Month by month

9 months, bolt by bolt.

Field LCP each month · p75 mobile
  • Within the 2.5 s budget
  • Over budget
  1. Month 1

    Audit & budget

    Field data, waterfall traces, Search Console, and a performance budget signed off by the owners.

  2. Months 2–3

    Rebuild

    Hero video to poster image, PDF menu to a 46-dish HTML menu with schema, and the site moved off the page builder to static templates.

  3. Month 4

    Strip

    Builder bundle removed, fonts subset and self-hosted, the chat widget loading on tap.

  4. Month 5

    Launch & verify

    New site live; lab data and CrUX field data watched, the budget wired into CI on every deploy.

  5. Months 6–9

    Hold the line

    Every change goes through the budget; new class and dish pages, a monthly review.

Results

The door swings open in under two seconds.

Mobile LCP (p75)

11.2 s → 1.8 s

9.4 s faster

Organic bookings / month

+133%

52 → 121 bookings a month

Menu dishes indexed

0 → 46

every dish in HTML with MenuItem markup

URLs passing Core Web Vitals

0 → 28 of 28

every URL passes on field data

Modelled scenario for a fictional business — not a client result. Rankings and traffic are never guaranteed; see our Terms.

Mobile LCP, field data (p75)
0 s4 s8 s12 sbudget 2.5 sM1M2M3M4M5M6M7M8M911.2 s1.8 s
Organic bookings per monthTablesCooking classes
0408012052M150M255M363M474M586M698M7110M8121M9

Searches that now open the door

QueryMonth 1Month 9
rooftop restaurant stone town#14#2
spice cooking class stone town#9#1
zanzibar cooking class#23#4
sunset dinner stone town#31#6
octopus curry zanzibar#48#7
corso di cucina zanzibar#27#5
kochkurs sansibar#35#8

Modelled mobile positions for a searcher in Stone Town. Queries are shown as typed, in the searcher's language.

Get a plan like this

Is your site beautiful but slow to open? Start with the Signal plan, or tell us what is weighing your pages down.

← All case studies