A faster B2B page can earn a lower Lighthouse score and still create more qualified enquiries. In a conversion-focused website redesign, speed is judged through task completion, not the number inside a testing tool. A move from 72 to 96 is worthless when pricing appears later, a form stalls or the CRM never receives the submission.
Lab tools reward technical conditions. Buyers reward an uninterrupted route from interest to a useful commercial outcome.
Speed budgets should follow that route.
Website speed and B2B conversions: measure the task
A buyer arrives to judge fit, inspect a relevant capability and take a commercial action. Loading the top of the page is only the beginning.
Task completion is the percentage of eligible sessions that reach a predefined commercial end state without an error and within an agreed time window. Six measurable proxies expose that journey:
- Primary-action start rate: eligible sessions that begin a form, booking or registration.
- Start-to-success rate: started actions that reach a genuine confirmation state.
- Landing-to-success time: median and 75th-percentile elapsed time from arrival to completion.
- Validation and network error rate: attempts blocked by the interface, server or integration.
- Step-level abandonment: the exact stage where booking, filtering or registration stops.
- CRM-verified enquiry rate: eligible sessions that create a matching, usable CRM record.
Put those six proxies beside Lighthouse and Core Web Vitals, not beneath them. In an explicitly illustrative comparison, route A scores 92/100 in Lighthouse but completes 41% of tasks; route B scores 78/100 but completes 47%. Route B wins.
The distinction starts with architecture. In our work for Lanteria, a broad Microsoft 365 HR offer had to be routed for several stakeholder audiences. On AfriCap Hub, the meaningful path ran through catalogue discovery, filtering and registration. Those case studies document structural decisions, not performance results, so no numerical uplift should be inferred from them.
GA4 should record task starts, successes and failures. Google Tag Manager can carry the events, while CRM identifiers establish whether a reported conversion became an actual enquiry. Consent Mode must remain correctly configured throughout.
A fast interface with an unclear next action needs task-focused UX and UI work, not another image compression pass.
The honest limit here is that task completion is confounded by traffic mix, offer strength and form changes. A before-versus-after dashboard cannot isolate speed by itself.
Task completion outranks a laboratory score.
Four checks for connecting speed with conversion
Budget follows evidence, so every candidate fix must pass four checks.
- Path reach. If a page receives at least 20% of eligible landing sessions or sits directly between paid traffic and an enquiry, include it in the speed backlog. If it receives less than 2% and carries no commercial task, defer it.
- Delay size. If a field trace indicates that the 75th-percentile task journey can be shortened by at least 300 milliseconds, schedule the fix. If the expected saving is below 100 milliseconds and errors will not fall, bundle it into a later release.
- Drop-off shape. If primary-action starts are 62% of eligible sessions but successes are 49%, the 13-percentage-point loss deserves investigation before global page optimisation. Inspect validation, submission requests and third-party dependencies at that step.
- Measurement agreement. If GA4 success events and matching CRM records diverge by more than 5% over the same 28-day window, repair the measurement chain first. In an illustrative reconciliation, 210 reported successes versus 188 actual CRM records is a tracking problem, not evidence of conversion growth.
Personalisation scripts receive no exemption from these checks. The decisions described in our guide to planning useful B2B website personalisation should determine whether their processing cost earns a place on the route.
Performance work must also preserve requirements outside this article’s scope, including the practices covered in our B2B website accessibility guidance.
Commercial friction decides the performance backlog.
Five prioritised speed fixes by milliseconds and developer hours
The pattern we hit most often is excessive work on the buyer’s critical path: scripts execute, assets compete and integrations respond before the visitor can continue.
The five-fix queue below uses planning ranges for audit triage. They are not published study findings, client results or guaranteed savings. Developer cost assumes an established codebase with normal access to deployment and analytics.
| Priority | Fix and measurable trigger | Expected task-path saving | Implementation cost |
|---|---|---|---|
| 1 | Defer non-essential Google Tag Manager, chat and personalisation scripts when they create over 300ms of blocking work | 300–1,200ms | 4–12 developer hours |
| 2 | Right-size the critical image and stop preloading hidden media when the largest first-view asset exceeds 250KB | 250–900ms | 2–6 developer hours |
| 3 | Remove redirect chains and cache the initial response when the route contains at least two sequential redirects or exceeds 800ms response time | 150–600ms | 3–8 developer hours |
| 4 | Split unused JavaScript and CSS by commercial template when more than 40% of transferred code is unused on that route | 200–700ms | 8–20 developer hours |
| 5 | Decouple confirmation from CRM or chat responses when submission-to-confirmation exceeds 1,000ms or technical errors exceed 2% | 200–700ms | 8–18 developer hours |
Rank viable fixes by the lower-bound milliseconds saved per upper-bound developer hour. Override that ranking when a task error rate exceeds 2%, because successful completion matters more than raw latency.
Our falsifiable claim is that fix 1 produces the largest real-world task-completion gain on a script-heavy lead-generation path; a controlled 50/50 path test disproves the claim if another single fix delivers both a larger reduction in P75 landing-to-success time and a larger increase in CRM-matched completions.
Deferral does not mean indiscriminate deletion. Google Ads conversion tags must still fire correctly, and Consent Mode must preserve the intended measurement and consent behaviour.
Distinctive type, motion and art direction should not be removed by reflex either. The principles behind building a recognisable B2B brand personality can operate inside a defined performance budget.
Fix the largest path-level delay first.
If this analysis exposes wider gaps in your site, we can turn the evidence into a focused redesign brief — book a call
What does not matter to B2B conversion
Three common backlog items consume budget without moving the buyer’s task.
1. Chasing a perfect score
Moving Lighthouse from 98/100 to 100/100 is not a commercial outcome. The remaining points may require removing consent controls, analytics or useful interface behaviour while leaving task completion unchanged.
Use a score to locate technical work, never to approve it.
2. Optimising pages nobody reaches
A dormant archive page loading in 3.6 seconds rather than 1.8 seconds still contributes nothing when buyers never enter the commercial journey through it.
Defer any page below 2% of eligible sessions unless it directly supports a live enquiry route, sales conversation or paid campaign.
3. Shaving less than 100 milliseconds from an already responsive step
No buyer will reliably perceive a stable interaction changing from 2.10 seconds to 2.04 seconds. That 60ms difference can also sit inside normal network and measurement variation.
If a sub-100ms improvement changes neither task failures nor interaction feedback, do not fund it as a standalone release.
Invisible gains do not deserve visible budgets.
Website speed to B2B pipeline: an illustrative calculation
Take an illustrative UK consultancy spending £8,000 a month on Google Ads. No figure below is Actualyse client data.
The calculation uses six labelled inputs:
- Media input: £8,000 monthly Google Ads spend, held constant.
- Traffic input: 2,000 eligible paid landing sessions per month, held constant.
- Speed input: 4.2 seconds before versus 2.6 seconds after.
- Conversion assumption: 2.0% enquiry rate before versus 2.4% after.
- Qualification assumption: 50% of enquiries qualify in both versions.
- Pipeline-value assumption: £12,000 per qualified opportunity.
The reproducible arithmetic is:
Enquiries before = 2,000 × 0.020 = 40
Enquiries after = 2,000 × 0.024 = 48
Incremental enquiries = 48 − 40 = 8
Qualified opportunities before = 40 × 0.50 = 20
Qualified opportunities after = 48 × 0.50 = 24
Incremental qualified opportunities = 24 − 20 = 4
Pipeline before = 20 × £12,000 = £240,000
Pipeline after = 24 × £12,000 = £288,000
Illustrative incremental pipeline = £288,000 − £240,000 = £48,000
Cost per enquiry before = £8,000 ÷ 40 = £200
Cost per enquiry after = £8,000 ÷ 48 = £166.67The comparison moves from 4.2 to 2.6 seconds, 40 to 48 enquiries and £240,000 to £288,000 of pipeline. The 2.4% conversion rate is a scenario assumption, not a benchmark or forecast.
Pipeline is not revenue. No close rate, sales-cycle delay or margin has been applied.
What this cannot tell you is whether speed caused the assumed change. Bid strategy, search terms, campaign creative and buyer demand must remain sufficiently stable, followed by a controlled test and CRM reconciliation.
When several templates and journeys need intervention, use conversion-led B2B redesign planning to define the wider scope. A structural problem spanning the site belongs with a website design and redesign team, not an endless sequence of isolated patches.
Pipeline value makes speed investment accountable.
FAQ
Four questions settle the remaining operating choices.
How long should a speed experiment run?
Run it for at least 14 complete days and continue until each version has 100 CRM-matched task completions. Treat that as a minimum evidence gate, not a guarantee of statistical significance.
Should Google Ads traffic be reported separately?
Split brand and non-brand Google Ads traffic when either supplies at least 25% of eligible sessions. A blended conversion rate can hide a campaign-specific landing-page failure.
When should tuning become a redesign?
Choose redesign when the same structural bottleneck appears across three or more commercial templates, or when a safe patch exceeds 40 developer hours. Tune the existing system below those triggers.
Who should own the speed backlog?
Marketing should own the task definition, development the latency diagnosis and operations the CRM reconciliation. Appoint one decision owner and require acceptance or rejection within 48 hours of receiving the evidence.
Clear ownership keeps performance work commercial.
Summary
Apply these five rules:
- Approve speed work against CRM-verified task completion, never Lighthouse alone.
- Repair tracking before optimisation when GA4 and CRM outcomes differ by more than 5%.
- Prioritise commercial-path changes saving at least 300ms; bundle changes below 100ms.
- Test deferring non-essential JavaScript first, then accept whichever fix wins on P75 time and completion.
- Defer pages below 2% of eligible sessions unless they carry a direct commercial task.
Task completion remains the final performance test.
Actualyse designs and rebuilds B2B websites that turn research visits into qualified pipeline. Book a call to talk through where yours stands.
