Agencies love to talk about “fast websites.” Most of them will not show you the scoreboard for the site they built for themselves. We will — because we ran the experiment on our own house, in public, across three platforms: WordPress, then Webflow, then Hydra.
This is not a vendor roast. It is the receipt for why CI Web Group stopped renting someone else's architecture and built our own. If you are deciding what to put under your business's digital front door, these are the numbers I wish someone had put in front of me before we spent years optimizing a foundation that could not become what the market needed next.
Definition — what “performance” means when money is on the line
Performance is not a Lighthouse vanity gauge you screenshot for a pitch deck. It is whether a homeowner with a broken AC at 9pm gets a page that paints, answers, and lets them call or book before they hit Back. Google's own research has said for years that as mobile load time moves from one second toward three, bounce risk climbs hard. Speed shows up in rankings, in Quality Score on paid traffic, in whether an AI agent can even read your page before its time budget expires, and in whether your site survives the weekend heat-wave spike without a database falling over.
We teach buyers how to read the scoreboard in our Lighthouse page-speed guide — field data first (real Chrome users), lab data second (the engineering diagnosis). The rule on that page is the same rule I run the company by: no agency will build you a better website than the one they built for themselves.
When to use each chapter of the story
Use this essay when someone tells you “WordPress is fine if you hire the right developer,” or “Webflow is modern enough,” or “static sites are a fad.” Use it when a peer agency flashes a desktop-only 95 and hides the mobile field section. Use it when your own team is tired of plugin patches and still losing the Core Vitals fight. The sequence matters: we did not skip steps. We earned the scar tissue.
Chapter one — WordPress: the industry default we outgrew
For years, WordPress was how we shipped. Themes, plugins, a database on the open internet, a login page waiting to be probed. It was also how most of the trades web still ships. On the comparison we publish on our own platform pages, a typical WordPress load sits around 3.5 seconds — and the HTML that arrives is a pile: on the order of 1,500+ lines of theme, plugin, and builder noise before the customer gets to the offer.
That is not a moral failing of WordPress as a blogging CMS. It is what happens when you force a blog engine to be a lead machine, a booking surface, a compliance surface, and a security surface at once. Every plugin is another attack surface and another main-thread tax. We wrote about the structural problem for trades businesses at ciwebgroup.com/learn/wordpress-problems. Living it was worse than writing it.
Chapter two — Webflow: the fifty-plus percent lift that still wasn't enough
So we moved. Webflow was the right next bet for a company that needed cleaner hosting, fewer plugin fires, and a real design system. The performance story was not subtle. On Core Web Vitals, the move from WordPress to Webflow picked up fifty-plus percent — the same receipt I put on the record in Why Does the Industry Call Anything Better Than Them a Scare Tactic?. Average load on the published comparison drops from 3.5 seconds to about 1.8 seconds. The code diet improves too — on the order of 600+ lines instead of fifteen hundred.
Webflow made us better. It did not make us finished. It was still a hosted CMS with designer dependency, still heavier than what AI crawlers and impatient mobile buyers reward, still not the architecture you build an operating system on. Faster mediocrity is still a ceiling. When your clients' jobs depend on the same foundation, “good enough for marketing sites” is a quiet way to lose.
Chapter three — Hydra: under 100ms, because the page is already HTML
Hydra is what we built when Webflow's ceiling became the problem. Decoupled publishing. Pre-rendered static HTML. CDN edge delivery. No WordPress admin login to brute-force. No plugin directory to hijack. No database query on the request path for a brochure page. On the same comparison bars we show buyers: under 100 milliseconds — and the illustrative HTML footprint collapses to on the order of 69 lines for a clean page shell.
That is not a caching plugin someone remembers to renew. That is architecture. LCP, CLS, and INP become properties of what you ship, not a monthly report item. Spikes do not “take the site down” the same way, because there is nothing fragile to crash on the edge path. I have told the company story of rebuilding around an intelligence layer in From 320 to 38 and From Marketing Agency to Technology Company. Hydra is the public website face of that same decision.
Failure modes — what people get wrong when they read the scoreboard
- Desktop-only bragging. Desktop lab is the friendlier number. Mobile is where local-service money lives. If an agency will not show mobile field data, they are choosing the flattering gauge.
- Deleting revenue tools to chase a 100. ServiceTitan, chat, call tracking — load them wrong and they punish Total Blocking Time. Load them after first paint, on intent, and you keep the booking. The game is not the gauge.
- Confusing lab with field. Lab explains. Field decides. Our own marketing site is the honesty test: on July 19, 2026, ciwebgroup.com passed Core Web Vitals in the field (LCP 1.9s · INP 104ms · CLS 0) while the throttled mobile lab performance score sat amber at 68 because of the analytics and session stack we deliberately run on ourselves. Desktop lab that same day: 92 · 97 · 96 · 100. We publish the amber. Anyone who hides theirs is selling a story, not a standard.
- Stopping at Webflow and calling it transformation. A better CMS is not an operating system. If your “AI strategy” is still a theme and a plugin list, you are one platform hop behind the market you are pitching.
Proof — fleet numbers, not vibes
The journey is not only our homepage. Across the Hydra launches we publish with Lighthouse receipts, the before/after averages (mobile lab, as of the July 19, 2026 guide snapshot) look like this:
- Performance: typical pre-rebuild trades site ~59.8 → Hydra clients ~95.8 (22 of 24 at 90+)
- Accessibility: ~88.8 → ~96.4
- Best Practices: ~62.5 → ~92.9
- SEO (technical crawlability): ~88.5 → ~98.5
Those “before” sites were the WordPress / Wix / Webflow stack the industry still sells as modern. The after numbers sit on named customer stories with domains and dates — not a composite invented for a webinar. Best published launch on that board: a clean 100 on performance.
And the load-time ladder we put on our own platform pages stays the simplest sentence in the company:
WordPress ~3.5s → Webflow ~1.8s → Hydra <100ms.
Action — what to do with this if you run trucks, not servers
- Open pagespeed.web.dev, paste your homepage, read the field section first.
- Paste your agency's homepage. Same test. If their house fails the standard they sell you, you have your answer.
- Ask one architecture question: is the page HTML at the edge, or a live app assembling itself after the request? That single answer predicts most of the scoreboard.
- If you want the longer buyer education, read the Lighthouse guide and the Hydra overview on ciwebgroup.com. If you want the cultural version of why improvement gets called “fear,” read the scare-tactic essay.
We did not invent a faster plugin. We changed platforms until the numbers matched the promise we make clients — and then we published the numbers. WordPress taught us the ceiling. Webflow raised it. Hydra removed it. That is the performance story. Everything else is branding.