Site speed is a revenue problem wearing an engineering costume

Performance work gets deferred because it is framed as engineering hygiene — worth doing, never urgent. Framed properly, it competes for budget with advertising, and often wins: paying more for traffic that arrives at a page people abandon is the most expensive way to be slow.

Test on the device your customers own

Most performance testing is done on a developer's laptop over office broadband, which is the one condition that does not matter. The number that counts is a mid-range Android phone on a congested mobile network — the median for a large share of visitors in every market we work in.

Where the weight usually comes from

• Images shipped at desktop dimensions to phones.
• Fonts that block text from rendering while they download.
• Third-party tags — chat widgets, heatmaps, four analytics tools, two of them forgotten.
• JavaScript for interactivity on pages that are mostly text.

The third item is worth a specific look. Marketing tags accumulate quietly, nobody owns the list, and the cumulative cost frequently exceeds the entire application.

Set a budget before design, not after launch

A performance budget agreed at the start — a ceiling on page weight and a target for Largest Contentful Paint — changes design and engineering decisions while they are still cheap to change. Retrofitting speed onto a finished build is always more expensive than designing within a limit.