Website analysis

We analyzed our own website first. The expensive mistake was invisible.

8 min read

We ran the analysis on our own site before anyone else's. Lighthouse, the raw HTML, and one mistake no design review would have caught.

The old Omni2 homepage hero as it stood in Framer, headline Design, Build, and Grow High-Performing Digital Experiences
The Omni2 homepage as it stood, built in Framer. Everything visible here was fine. The expensive mistake is in the copy.

A website analysis should find what quietly costs Architecture, Engineering and Construction (AEC) firms leads. Somebody else’s practice would have been an easier place to start, and would have taught me nothing I was willing to be judged on. So the first analysis is ours.

Our site ran on Framer until the rebrand in September 2026. We built our original website, and were pleased with the results in early 2025. The old site came down when the new one went live, so in October 2026 I republished the original Framer project to a temporary address, ran Lighthouse, and read the HTML a browser actually receives. The copy came back down the same day.

Six findings. Five of them are an afternoon of work. The sixth took a rebrand.

What the numbers said

Lighthouse is a code review for a website. Google’s tool loads the page the way a visitor would, scores the result out of 100, and lists every failure with an estimate of the cost. An architect does not need to know how an inspector calculates egress width to understand a failed inspection, and the same holds here.

Our homepage scored 60 out of 100, on mobile and on desktop alike. Something appeared on screen inside two seconds. The headline a visitor actually came to read took 12.7 seconds on a phone, and 10.3 on a desktop. Google has a name for the second measurement, Largest Contentful Paint, and a threshold: 2.5 seconds is good, anything past four is poor.

In other words, the page started drawing almost immediately and then left a visitor looking at a nearly empty screen for ten more seconds.

I had never seen the following happen. I always arrived at our site with every asset already cached, which is the condition no prospective client is ever in.

The homepage ran 15,417 pixels tall, roughly seventeen phone screens, and pulled 2.6 MB across 88 separate requests, 26 of them scripts.

omni2design.com · the whole homepage, 15,417px · scroll inside
The entire old Omni2 homepage captured as one tall strip, 15,417 pixels from top to bottom, with its only H1 near the bottom in the team section
The full homepage, 15,417 pixels tall. The only H1 sits near the bottom, in the team section.

I blamed the images first

Here is the step most analyses skip. Skip the measurement and an analysis becomes fiction.

Before blaming the images, I measured them. The source files were enormous, PNGs running up to 9.8 MB, roughly 55 MB in total. Framer never served the originals. The platform resized every image and negotiated WebP, so the same asset reached a visitor as a 27 KB file at 1024 pixels wide, or 10.7 KB at 512.

Compress your images, the opening line of every analysis ever published, was already done for us.

The dimensions were a different story. Framer sent images around 2,000 pixels wide into slots a fraction of the size, and Lighthouse put the waste at roughly 1.6 MB and five seconds. The platform solved file format and compression, and left image dimensions alone.

And the images were still not the cause. Lighthouse names whatever took longest to appear, and on our homepage the slowest element was the headline itself. Plain text, no picture involved. The server delivered the words in under a second, and the browser then spent the next twelve running scripts instead of drawing a headline already sitting in memory.

The old Omni2 homepage hero on a phone, the headline Design, Build, and Grow High-Performing Digital Experiences

Where the twelve seconds went

ServerBrowser waiting
Seconds until the headline appeared on screen
DeviceServer responseBrowser waitingTotal
Mobile0.60 seconds12.13 seconds12.73 seconds
Desktop0.60 seconds9.74 seconds10.34 seconds

Google’s threshold for good is 2.5 seconds, and both bars run far past it.

The hero a visitor saw on a phone, and where the wait went. The server delivered the headline in 604 milliseconds. The browser spent the next 12,129 running scripts before drawing the words.

I nearly wrote the 55 MB figure down as a finding. The number was real and the conclusion would have been false, because 55 MB is the weight of the source files rather than the weight a visitor downloads. Notice which way the mistake pointed. A 55 MB homepage makes a better story than a 2.6 MB homepage, so the error was exactly the kind I wanted to believe. An analysis confirming only what the author already suspected is marketing, not an audit.

The heading nobody reads

Every page carries one heading a search engine reads as the title of the whole page. Ours said “Our Team”, and sat 13,112 pixels down a 15,417 pixel page, almost at the footer.

The headline a visitor actually saw, “Design, Build, and Grow High-Performing Digital Experiences,” sat one level down, near the top of the page. So the strongest label on the homepage announced a team page, and the sentence saying what the studio does carried the same weight as a project caption. Lighthouse flagged the heading order on its own, knowing nothing about the content.

Nothing about the page looked wrong, because heading levels are invisible. A heading can be styled any size at all regardless of the level underneath, so a design review never catches the error. The client sites got the levels right. Ours did not.

The metadata written for a bigger box

The title tag ran 73 characters. Search results display about 60.

The meta description ran 248 characters. Search results display about 155, so roughly 90 characters were written, approved, and never shown to a single person.

Metadata written longer than the box

Shown in searchWritten, never shown
Characters written versus characters a search result displays
FieldWrittenDisplayed
Title tag73 characters60 characters
Meta description248 characters155 characters
The solid portion is what a search result displays. The rest was written, approved, and never shown.

Both were composed carefully. Both overflowed the only container they would ever appear in.

The placeholder we never replaced

The About section sat under the same heading, next to a photograph of two people.

We needed a team image and had no time to arrange a shoot, so we used a stock photograph and moved on. Stock libraries and AI are legitimate tools, and a small practice without a photographer on call will reach for both. The source was never the failure. The failure was stopping there, because a sourced or generated image needs enough post-production to hold up at full size, and needs replacing the moment a real photograph exists. Ours got neither, and the placeholder became permanent, the way placeholders do.

A practice asking a client to trust the work, illustrated with an image we never finished. On a client project we would have caught the same gap inside an hour.

The old homepage's About section: the Our Team heading and the About paragraph beside a stock photograph of two people
The page's only top-level heading, the placeholder photograph, and the sentence giving the whole argument away.

The mistake no tool can find

Everything above is cheap. Rewrite a heading, trim a description, book a photographer, done in an afternoon.

The sixth finding took a rebrand.

Our old homepage sold to “founders, startups, and brands, through UX-driven web design, product strategy, and social-first digital marketing.” The About section described “a Philadelphia-based branding and digital creative studio helping small businesses and founders look professional, modern, and trustworthy online.”

Not one word about architecture. Not one word about the built environment. An architect trained at Penn, selling undifferentiated design services into the most crowded market on the internet, competing with every studio on earth for the same buyer.

The background was always load-bearing. An MArch and an MLA from Penn, years inside design-first practices, projects carried from schematic design through construction administration, a working knowledge of what a Statement of Qualifications is for and what a submittal log has to hold. A studio without the same history has to learn the vocabulary on somebody’s live project. I already spoke the language.

Marketing has a term for what I was missing. Founder-market fit names the overlap between what a founder has actually done and the market the business serves. Mine was close to total, and our homepage named none of the background.

Anyone can copy a service list in an afternoon. Nobody can copy eight years of drawing sets, and an architect can tell inside one conversation whether you have sat through a coordination meeting. The only durable advantage we had was the one we left off the page.

And the answer was already on the page. The same sentence continues: “We combine architectural clarity with...”

Architecture appeared exactly once on our homepage, as an adjective describing a visual style. The discipline now defining the entire practice carried no weight at all.

No crawler flags a misaimed page. No Lighthouse audit scores positioning.

A page can be fast, accessible, correctly structured, and aimed at nobody.

Before, the Framer homepage
Design, Build, and Grow High-Performing Digital Experiences

The headline a visitor actually saw, selling to founders, startups, and brands.

Designing websites and web apps for AEC firms, built by architects.

The first heading on the page.

No audit tool would have flagged the first heading. Any architect reads the second and knows within four words whether to keep reading.

What an analysis should actually check, in order

Measure before diagnosing, because the platform may have already solved the problem you are about to write up.

Read the code a browser receives rather than the design file, because the invisible mistakes live there and no design review will ever find them.

Check what truncates. Titles, descriptions, anything written for a box with a fixed size.

Then ask the question no tool asks. Who is the page for, and does the page say so in the first sentence?

Go read your own About paragraph

Five of the six findings cost an afternoon. The sixth could not be found with a tool, and fixing the other five would never have reached the positioning. A fast, accessible page selling the wrong service to the wrong people is a fast, accessible page selling the wrong service to the wrong people.

Architects, engineers and contractors already audit their own work as hard as any industry does. Every drawing set gets a redline. Every calculation gets checked and stamped. Every submittal gets reviewed, every RFI gets answered, every punch list gets walked.

Your own website is the one surface never getting the same treatment, because you have never once read your own pages as a stranger. Nothing leaves your office without a redline. Your About paragraph has never had one. Go and find the word carrying no load, and then build the page on the word you define.

Common questions

How do I know if my architecture firm's website is losing leads?
Start with three measurements rather than an opinion. Run Lighthouse or PageSpeed Insights on the homepage and look at Largest Contentful Paint, the moment a visitor can actually read the page, which should land under 2.5 seconds. Check whether your title tag and meta description fit inside roughly 60 and 155 characters. Then read your own opening paragraph and ask whether a developer or client could tell within one sentence who the practice serves. Speed and metadata are measurable. Only you can judge who the page is aimed at.
What should a website analysis actually check?
Measurement first, then structure, then copy. Performance numbers from a real tool rather than an impression. The code a browser receives, including heading order, link text and image descriptions, because none of those show up in a design review. Anything written for a fixed container, like titles and descriptions. And last, whether the page names its audience. An analysis stopping at page weight has only checked the part a platform may already be handling.
Does page speed matter for an architecture firm's website?
Yes, though rarely in the way firms expect. One number matters: Largest Contentful Paint, the moment a visitor sees the main content rather than the moment the page starts loading. A site can put a logo and a background on screen in under two seconds and still leave the real content missing for ten. Firms miss the delay because they visit their own site constantly with everything cached, which is the one condition a prospective client is never in.
Why does my website look good but bring in no enquiries?
Usually because the page is designed well and aimed at nobody. Visual quality and positioning are separate problems, and no audit tool scores the second one. If the opening paragraph could be swapped onto another studio's site without anyone noticing, the page is competing on price in a market where you meant to compete on expertise.
Should I fix my website or rebuild it?
Fix your website when the findings are technical: heading structure, metadata, image handling, slow content. Those are an afternoon of work on any platform and rarely justify a rebuild. Rebuild when the problem is who the site is for, because repositioning changes the copy, the structure, the project selection and usually the brand underneath. Run the cheap fixes first. The fixes will tell you quickly whether the expensive problem is the one you actually have.

Sources

The numbers above come from measuring the original Omni2 homepage, republished to a temporary address in October 2026, rather than from an impression of it.

Christian Cueva

Written by

Christian Cueva

Founder of Omni2 Design, building websites and web apps for AEC firms. Architect by training. About the studio