Site Speed as a Revenue Lever: What Vodafone, Rakuten and T-Mobile Measured

Performance work is usually sold on a score and then abandoned when the score plateaus. The published case studies argue for a different framing. Vodafone ran an A/B test on a 31% improvement in Largest Contentful Paint and recorded 8% more sales, alongside a 15% improvement in lead-to-visit rate and 11% in cart-to-visit rate.

Rakuten 24 reported that a good LCP corresponded to a conversion rate increase of 61.13%, and that its investment in Core Web Vitals increased revenue per visitor by 53.37% and conversion rate by 33.13%. T-Mobile, publishing a data-driven performance programme, reported a 20% reduction in user-reported site issues and a 60% improvement in visit-to-order rate.

These are large operators with large samples, which is exactly why they are useful: the effects are measured against revenue, not against a synthetic audit. The mechanism is not mysterious. Slow pages lose people at the moment of highest intent, and the people you lose are disproportionately on mid-range devices and congested mobile networks — which in the UK means a meaningful share of your market outside city centres.

The trap for mid-market businesses is optimising the wrong page. Home page performance is the easiest thing to measure and rarely the thing that costs you money. The pages that matter are the ones a buyer reaches with intent: the product or service detail page, the pricing or quote page, and the form itself. Fix those first and measure the funnel, not the score.

The second trap is regression. Almost every performance problem we inherit was introduced after launch by a tag, a chat widget, an unoptimised hero image, or a marketing script added without review. Speed is not a project; it is a constraint you enforce on every subsequent change.

Treat page speed as a commercial metric with an owner, hold third-party scripts to the same standard as your own code, and instrument field data rather than lab data. The published evidence says the return shows up in orders.