Web development
Slow website: what are the real business impacts?
Short answer
A slow website can create observable friction; its business effect remains a hypothesis to test.
Slowness can delay access to content, postpone an interaction or interrupt a task. Its business impact depends on the page, the audience, the device, the journey and the outcome being observed. A performance value becomes commercially useful only when it is related to a real population and a confirmed business event.
Lighthouse can reproduce a loading problem. CrUX or first-party RUM can describe experiences in the field. Funnel data can locate a lost or completed step. None of those layers, in isolation, states why a person left or how much revenue speed caused the business to lose.
Measure the experience first. Attribute a business effect only when the population, event, comparison and uncertainty are documented.
Evidence boundary
What a speed measurement establishes — and what it cannot establish.
The useful distinction is not “technical versus business”. It is observed versus inferred. A technical defect may be established precisely while its revenue effect remains unknown. Conversely, falling conversions may be real while speed is only one competing explanation among offer, traffic quality, pricing, trust, accessibility or form failures.
| Observed or testable | Not inferred from that observation alone | Evidence needed to go further |
|---|---|---|
| LCP, INP or CLS for a named field population | The exact cause or the revenue lost | Attribution data, journey context and a business event |
| A reproducible laboratory regression | The share of real visitors affected | CrUX or documented first-party RUM coverage |
| A fall in form or checkout completion | That speed caused the fall | A stable comparison, competing hypotheses and sufficient volume |
| Good Core Web Vitals | A ranking, conversion or revenue guarantee | Search and business measurements interpreted separately |
Three layers
Experience, behaviour and business outcome answer three different questions.
Performance data describes when content appears, when an interaction responds and whether the layout remains stable. Journey data records a step reached, an error or a confirmed submission. Business data records the qualified outcome: accepted lead, sale, appointment or another result defined by the organisation.
These layers can be connected, but never silently merged. A p75 LCP is not a conversion rate. A lower completion rate is not a diagnosis. An increase after deployment is not automatically a causal effect.
Experience
02Journey
03Outcome
≠Automatic causality
Risk map
Seven areas where slowness can create a business risk.
These are investigation areas, not seven guaranteed losses. Each one names the observable signal and the limit that prevents an exaggerated conclusion.
First access
The main content arrives too late for part of the audience.
Observe: field LCP distribution by page type and device, plus eligible volume. Do not infer: that every slow visit was abandoned or that the same threshold has the same commercial value on every page.
Interaction
A click, filter or form action responds after the user expects it.
Observe: INP, the responsible interaction and task errors. Do not infer: that a slow response alone explains every incomplete task.
Stability
A moving interface can make an intended action fail.
Observe: CLS, shifted elements and mis-click or correction signals. Do not infer: a lost order when no order or journey event has been connected.
Conversion
A slow step can coincide with a lower completion rate.
Observe: the same business event and denominator across comparable populations. Do not infer: causality from a simple before/after comparison or from an industry benchmark.
Acquisition
Paid or organic traffic can reach a journey that cannot use it efficiently.
Observe: landing-page performance, campaign population and the first meaningful event. Do not infer: that poor campaign profitability is a speed problem before testing targeting, message and offer.
Search
Page experience can contribute to visibility without replacing relevance.
Observe: Core Web Vitals, indexation, impressions and clicks on the affected URLs. Do not infer: that a green score guarantees rankings; Google explicitly says it does not.
Perception
Waiting may alter perceived quality in a specific context.
Observe: task feedback, interviews or a comparative perception test. Do not infer: trust, seriousness or brand damage from a technical value alone.
Published evidence
The strongest published results are contextual, not universal coefficients.
The available studies support investing in performance, but their percentages cannot be copied into another business case. Deloitte analysed selected mobile brands over four weeks. Rakuten 24 compared an optimised landing page with its original version in a month-long A/B test. Farfetch connected performance and business data in the same sessions, then used controlled experiments for specific page types.
The transferable lesson is the method: name the population, isolate the change, preserve the business denominator and publish the limits. The uplift remains specific to the site, period, audience and protocol that produced it.
Deloitte
+8.4% retail conversions and +9.2% average order value were observed for a natural 0.1-second mobile speed improvement. Scope: four weeks, selected European and US brands, study commissioned by Google and based on Fifty-Five data. Source published 24 March 2020. This is a cross-brand correlation, not a coefficient for another website.
Read the study and its scopeRakuten 24
+33.13% conversion rate and +53.37% revenue per visitor were measured in a one-month, 50/50 A/B test on one high-traffic landing page. The optimised version loaded 0.4 seconds earlier and introduced no functional or visual difference. Source last updated 24 August 2022. These results belong to that experiment.
Inspect the A/B-test caseFarfetch
Beyond 2.5 seconds of LCP, conversion decreased by an average 1.3% for each additional 100 ms in Farfetch’s session-level statistical analysis. Source last updated 12 July 2022. The result belongs to its audience, journey and data model: it is a correlation, not a universal prediction.
Inspect the measurement approachEdikka delivery evidence
Three projects establish technical execution — not an attributed revenue uplift.
The results below are dated Lighthouse measurements already published in Edikka project cases. They show that a fast, stable implementation was delivered under the stated laboratory conditions. No conversion or revenue effect is claimed because no corresponding business dataset is published.
UTH.fr
100 mobile performance, 1.3 s mobile LCP and 0 CLS. Lighthouse measurements recorded on 8 June 2026.
Review the UTH projectLocafinance
100 mobile performance, 1.4 s LCP and 0.001 CLS. Lighthouse measurements recorded on 6 June 2026.
Review the Locafinance projectFertilité.net
99 mobile performance, 0 ms TBT, 0 CLS and 1.5–1.6 s LCP. Lighthouse measurements recorded on 30 June 2026 with Analytics active.
Review the IMF projectDeclared limit
These are laboratory delivery results. They do not describe every field visit and do not establish a commercial effect.
Read the measurement protocolThese measurements will be rerun at the 7 December 2026 review. Any value that is no longer reproducible will be removed from the current evidence and retained only as a dated historical result.
Decision
Fix a proven obstruction; measure before promising an uplift.
Performance work does not need a revenue forecast to be legitimate. A blocked form, an unusable interaction, a severe field failure or a documented regression can justify a correction on experience, reliability and quality grounds. Commercial attribution is a separate question.
Evidence first, exposure second, promise last.
A reproducible defect blocks content, interaction, submission or payment on an important path.
A field distribution is poor and the affected page or journey has meaningful exposure.
Only a laboratory score or an unsegmented average is available.
No comparable business event, denominator or sufficient volume supports an uplift claim.
Editorial boundary
This page frames business risk; the linked resources own measurement and conversion.
Sources and limits
Read every result with its population, period and protocol.
This article publishes the external results with their stated scope; none can be transferred as-is to another website. It does not claim that Edikka’s technical results caused a commercial uplift. It does not treat missing CrUX data as a pass, and it does not reconstruct revenue from an isolated Lighthouse score.
Google states that Core Web Vitals are used by its ranking systems, while also stating that good results do not guarantee top rankings. Field measurement guidance recommends distributions and the 75th percentile rather than an average. Business-impact guidance recommends sufficient traffic and controlled, preferably server-side experiments when causality is the question.
- Google Search Central · Page experience — ranking scope and absence of a ranking guarantee.
- web.dev · Measuring Web Vitals in the field — distributions, percentiles and real-user coverage.
- web.dev · Relating site speed and business metrics — traffic requirements, comparisons and server-side experiments.
- Deloitte · Milliseconds Make Millions — multi-brand mobile study with a stated period and scope.
- web.dev · Rakuten 24 — a one-month A/B test, not a universal benchmark.
- web.dev · Farfetch — session-level correlation followed by page-specific experimentation.
Version 1.1.1, reviewed 7 September 2026. Next scheduled review: 7 December 2026, including a rerun of the three dated Edikka Lighthouse measurements, or earlier if a cited methodology materially changes.
Conclusion
Speed deserves investment without needing an invented revenue figure.
A faster, more stable and more responsive journey is a legitimate quality objective. It can remove observable friction and protect an important task. The business value of that improvement becomes attributable only when performance and outcome data describe comparable populations under a documented method.
The honest decision is therefore not “speed has no value until revenue is proven”. It is: correct verified obstructions, monitor field exposure, connect business events where the volume allows it, and keep unknowns visible.
Performance removes friction. Measurement establishes what that change is worth in this business, for this audience and this journey.
A fast website removes friction. It does not manufacture proof.
Technical performance, observed behaviour and business outcomes belong to the same decision, but they are not interchangeable measurements.
Edikka corrects what can be reproduced, measures what can be observed and leaves the commercial effect open whenever the population or comparison is insufficient.
Remove verified friction
Identify the affected element, interaction or journey and counter-test the correction under comparable conditions.
Name what is observed
Separate laboratory execution, field population, journey events and business outcomes instead of collapsing them into one score.
Promise only what follows
A technical gain is published as a technical gain. A business uplift requires its own denominator, comparison and limits.
Go further on this topic
Additional answers to clarify the key points covered in this article.
