Page speed and revenue: building a case you can defend

Every page speed pitch deck contains the same slide: a wall of spectacular percentages implying that shaving a second off load time will transform revenue. Some of those numbers are real. Some are mangled beyond recognition. And almost none of them are presented with the caveats that a finance reviewer will immediately probe. Here is what the evidence actually supports, and how to build a speed case that survives contact with a CFO.

The evidence, with its caveats attached

The strongest recent study is Deloitte and Google's "Milliseconds Make Millions", which tracked 37 European and US brands across 30 million sessions for four weeks. It found that a 0.1-second improvement in mobile site speed was associated with +8.4% conversions and +9.2% average order value in retail, and +10.1% conversions in travel (Deloitte/Google, 2020). Note the phrasing: associated with. The study was observational, not a randomised experiment, and some vertical subsamples were small. That doesn't make it useless — it makes it an anchor to be used honestly, not a guarantee to be multiplied through a spreadsheet.

Akamai's analysis of roughly ten billion retail visits found a 100-millisecond delay hurt conversion rates by up to 7%, a two-second delay more than doubled bounce rates, and 53% of mobile visitors abandoned pages taking longer than three seconds (Akamai/SOASTA, 2017). Again the qualifiers matter: correlational data, published by a vendor that sells performance products — and "up to 7%" is a ceiling, not an average.

Walmart's older study found every one-second improvement produced up to a 2% conversion increase (Walmart, 2012). The "up to" is routinely dropped when this figure is quoted; keep it. It is a single company's correlational data from 2012.

Then there are the case studies: Vodafone improved largest contentful paint by 31% and reported +8% sales; Rakuten 24 reported +53.4% revenue per visitor after a Core Web Vitals programme (Google web.dev). These are company-reported single cases with obvious survivorship bias — nobody publishes the optimisation that did nothing. Treat them as an illustrative range of what has happened, never as an expected value.

One more discipline point: the famous "Amazon found 100ms of latency costs 1% of sales" figure has no verifiable primary source — the magnitude exists only in old conference slides. Leave it out. A business case is judged by its weakest citation.

Why the honest version still clears the bar

Notice what the caveats do not do: they do not reverse the direction. Every credible study points the same way — faster pages convert better, slower pages bleed visitors. The uncertainty is about magnitude, and magnitude uncertainty is exactly what scenario ranges are for.

That is the core move: instead of quoting the biggest number you can find, build three cases from the evidence's spread.

• Conservative: something like Walmart's ceiling read pessimistically — well under 2% conversion uplift per second saved. If the project pays back here, it is close to unarguable.
• Moderate: an anchor in the Deloitte/Google findings, scaled honestly to how much speed improvement your engineers actually commit to, in your vertical.
• Optimistic: the upper end of the observational range — presented explicitly as the upper end, not the plan.

Then translate the uplift into money with your numbers, not benchmark ones: your traffic, your current conversion rate, your average order value. For context on where you stand, global e-commerce conversion averages 2.74%, ranging from 5.37% in beauty down to 0.71% in luxury (Dynamic Yield) — and in that panel mobile now converts at 2.86% against desktop's 2.46%, which matters because the strongest speed evidence is specifically about mobile.

As an illustrative example: a retailer with 300,000 monthly sessions, a 2.5% conversion rate and a £70 average order takes about £525,000 a month. A conservative 3% relative conversion uplift — 2.5% to 2.575% — is worth roughly £15,700 a month, or £189,000 a year. If the engineering cost is £120k, even the conservative case pays back within a year, and the moderate case is comfortably better. That is the shape of an approvable case: modest assumptions, visible arithmetic, upside held in reserve.

What separates a defensible case from a deck

Four habits make the difference:

• Cite the study, the year and the design. "Observational study of 37 brands, 2020" is more persuasive to a finance audience than an unsourced "+8.4%!", because it shows you know what the number can and cannot carry.
• Keep the "up to". Ceilings presented as averages are the fastest way to lose a reviewer.
• Tie uplift to a committed speed change. The studies describe what happened when speed actually improved by a stated amount. Your case needs engineering to commit to a target — milliseconds, not vibes.
• Plan the measurement. Speed work is unusually measurable: you will know your before-and-after load times and conversion rates. Promising that comparison up front converts a speculative case into a testable one.

---

The website speed calculator turns your traffic, conversion and order-value numbers into exactly this structure — conservative, moderate and optimistic scenarios, a full assumptions audit trail, and multi-year NPV — and the online conversion calculator does the same for broader conversion work. Your first calculator is free, with unlimited re-runs.