Build a Dollars-Per-Second Model for Your Own Website
A Tampa Bay accounting firm generates $2.4 million in annual revenue through their website. Their average load time, measured from PageSpeed Insights field data, is 4.2 seconds. They have never calculated what that might be costing them in leads they did not receive.
This article shows you how to do that math, and tells you exactly how much to trust the result.
That second part matters as much as the first. A model that pretends to be a measurement produces false confidence. A model that names every assumption and gives you a range instead of a single figure is genuinely useful for a capital-allocation decision. You can take it into a budget conversation and say: based on industry data, the estimated cost of this problem falls between X and Y, here is where each number came from.
That is what you are building here.
What You Need Before You Start
Four inputs. All of them are either numbers you already know or numbers you can get for free in under ten minutes.
1. Monthly website visits. Pull this from GA4, it is on the homepage dashboard. Use a rolling three-month average to smooth out seasonal spikes.
2. Current conversion rate. In GA4 (updated in 2024 to use the term “key events” for behavioral conversions), divide your key events by your total sessions. A service business’s conversion event is typically a contact form submission or a phone call click. An ecommerce site’s is a completed purchase. Use the conversion that directly maps to revenue.
3. Average revenue per conversion. For a service business, this is your average client value. For ecommerce, it is average order value. You know this number.
4. Current load time from field data. Open PageSpeed Insights and run your URL. Look at the field data section, not the Lighthouse score, and find your LCP (Largest Contentful Paint: the time until the biggest visible element on your page finishes rendering, which is the closest technical proxy for “the page loaded”). Use the 75th-percentile figure, which is what Google uses for the Core Web Vitals assessment. This is your current performance baseline.
Once you have these four numbers, you are ready to build the estimate. Note that diagnosing performance without paid tools, including where to find field data and how to interpret it, is covered in detail at diagnosing performance without paid tools.
Step One: Establish Baseline Revenue Per Visit
This is the simplest step and the anchor for everything that follows.
Revenue per visit = (Monthly visits × Conversion rate × Revenue per conversion) / Monthly visits
Or more simply: Conversion rate × Revenue per conversion.
For the Tampa Bay accounting firm: 4% conversion rate × $12,000 average client value = $480 in expected revenue for every 100 visitors. That is the baseline. Every percentage point change in conversion rate is worth $4.80 per 100 visits, or $576 per month at 800 monthly visits.
Write this number down. It is the multiplier that gives the rest of the model its meaning.
Step Two: Apply the Conversion-Impact Range
This is where the model requires honest handling of the evidence, because the research varies in quality and the numbers are not interchangeable.
Here is what the published literature actually says, with the methodology labeled for each:
The “7% per second” heuristic. Commonly attributed to Akamai and Aberdeen studies from the 2010s and included in Edmonds Commerce’s 2025 synthesis. Directionally plausible, but the methodology is old. Treat it as an upper-bound heuristic, not a measured fact.
Deloitte Ireland / Think with Google, “Milliseconds Make Millions” (2020, HISTORICAL, correlational). Found that a 0.1-second improvement correlated with an 8.4% increase in retail conversion rate. Correlational, it shows an association across a large industry dataset, not an isolated cause-and-effect. Also five years old.
Yottaa 2025 Web Performance Index (CURRENT, vendor study, 500 million-plus visits across 1,300-plus ecommerce sites, September–October 2024). Yottaa found that saving one second of mobile page load time increases mobile conversions by 3%. More recent and larger than the Deloitte study, but produced by a platform vendor with a commercial interest in performance. Treat the 3% figure as a data point within a range, not a universal rate.
The adjudicated range. Together, these sources suggest conversion rate changes of 3 to 8 percent per second of improvement, depending on industry, starting load time, and device mix. Use the range, not a single number.
The one truly causal study. The Vodafone A/B test (web.dev / Google case study, 2021, HISTORICAL) is the cleanest published controlled study: a 31% improvement in LCP (Largest Contentful Paint, the time until the main content finishes rendering) led to 8% more sales and 15% higher lead-to-visit rate, measured through a randomized A/B test running approximately 100,000 clicks and 34,000 visits per day per version.
That scale matters. A statistically valid A/B test on a speed change requires enough traffic that daily conversion variation does not swamp the signal. An SMB with 800 monthly visitors cannot run one. You cannot achieve causal proof for your own site without that traffic base. What this model gives you instead is a directional range informed by the causal evidence from large controlled studies, not certainty, but a defensible frame.
The Rakuten 24 A/B test (2022, cited in Edmonds Commerce WPO Stats database) produced a 53.37% increase in revenue per visitor, the highest per-site result in the published literature. Treat it as an exceptional upper bound, not a typical expectation.
Step Three: Run the Arithmetic
Take your baseline revenue per visit and apply the conversion-impact range to your estimated improvement.
The Tampa Bay accounting firm scenario:
- Monthly visits: 800
- Conversion rate: 4% (32 leads per month)
- Average client value: $12,000
- Current LCP (field data, 75th percentile): 4.2 seconds
- Target LCP after optimization: 2.7 seconds
- Improvement: 1.5 seconds
Applying the 3–8% per-second range to a 1.5-second improvement:
- Low estimate: 1.5 × 3% = 4.5% conversion improvement → 32 × 1.045 = 33.4 leads/month → 0.96 additional leads/month × $12,000 = $11,520 additional annual revenue
- High estimate: 1.5 × 8% = 12% conversion improvement → 32 × 1.12 = 35.8 leads/month → 2.56 additional leads/month × $12,000 = $30,720 additional annual revenue
Directional annual revenue range from this improvement: $11,520 to $30,720.
This is a directional estimate, not a guarantee. Label it that way when you use it. It is the range you need to decide whether the cost of optimization is justified, not the number you commit to your projections. The conversion research this model draws on is explained in detail at what one second of delay costs in conversions, including which studies are observational and which involved controlled tests.
For context, Edmonds Commerce’s 2025 illustrative model, which applies the 7% heuristic to a £1M/year business, estimates approximately £105,000 in additional annual revenue from a comparable 1.5-second improvement. That model uses the same 7% heuristic as the high end of this range. It is also labeled explicitly as illustrative.
Step Four: State Your Confidence Level
Every estimate needs a stated confidence level. Here is this model’s:
High confidence: The direction, faster sites convert better, is supported by multiple independent studies, a vendor study with 500 million-plus visits, and two randomized controlled A/B tests.
Medium confidence: The 3–8% per second range. This is the best defensible range given available evidence. It could be narrower with site-specific data, or fall outside this range for your specific industry or device mix.
Low confidence: That this specific range applies to your site without testing. The only way to replace “directional estimate” with “measured outcome” is your own real-user monitoring data, linked to conversion tracking through a stable session ID. The CrUX 28-day measurement window is a key part of understanding how long it takes post-fix improvements to show up in field data, relevant when you are tracking whether an optimization actually landed.
What this model cannot tell you: Whether your result lands at 3% or 8%, or whether factors like checkout flow complexity or form design are a bigger bottleneck than load time.
The Objection Worth Preempting
“How do I know this will actually happen for my site?”
You don’t, without an A/B test with sufficient traffic. No correlation study and no external case study can substitute for a measurement of your own site with your own users.
What this model gives you is a defensible range for a capital allocation decision. If the low end of your range exceeds the project cost with margin to spare, the optimization is worth doing even if the actual result is below the modeled expectation. If the numbers are close, you need better data before committing. That is also a valid output of the model.
Running the same arithmetic against your paid media spend: running the numbers on your own budget, often reveals a harder case for urgency: you are already paying to drive traffic to a page that converts below its potential.
What to Do Next
Pull your four inputs from GA4 and PageSpeed Insights. Run the low-end and high-end arithmetic. State your confidence level. Decide whether the range justifies the investment.
A free assessment from iServU runs this model against your actual data, real visit counts, real conversion rates, real field-data load times, and produces a range estimate with its assumptions fully stated. Gabriel Lopez-Seco runs it directly. No account manager, no handoff.
Schedule a free assessment at iServU.
Sources
- Deloitte Ireland / Think with Google (2020, HISTORICAL). “Milliseconds Make Millions”, cited in Edmonds Commerce (2025). https://edmondscommerce.co.uk/research/performance/load-time-impact/
- Edmonds Commerce (2025). Load time impact synthesis and illustrative models. https://edmondscommerce.co.uk/research/performance/load-time-impact/
- web.dev / Google (2021, HISTORICAL). Vodafone case study: A/B test, LCP improvement and revenue lift. https://web.dev/case-studies/vodafone
- Yottaa (2025). 2025 Web Performance Index (500M+ visits, 1,300+ ecommerce sites, Sept–Oct 2024). https://www.businesswire.com/news/home/20250128945228/en/Faster-Load-Times-Higher-Conversions
