Numbers tell you where, but people tell you why
Picture a dashboard with one alarming figure on it: 29% of the people who begin your sign-up never finish it. The number is precise, repeatable, and completely silent about the cause. You can stare at 29% all afternoon and it will not tell you whether the form is too long, the password rule is hostile, the submit button sits off-screen on mobile, or the error text appears in a colour nobody with low vision can read. The metric found the wound. It has no idea how it got there.
That's the standing tension in measuring anything, and websites included. There are two kinds of evidence, and each is nearly useless at the other's job.
Two questions, two kinds of evidence
- Quantitative measurement answers where and how many. It counts things. p75 LCP is 4.1 seconds; 900 pages share one render-blocking script; 38% abandon the form; rage-clicks cluster on a thing that looks like a button but isn't. Numbers scale without fatigue, they settle taste arguments, and they tell you the size of a problem, which is how you work out what deserves your week.
- Qualitative measurement answers why and how it feels. It observes and asks. You sit beside five people trying to register and watch three of them hunt for password rules that only appear after a failed attempt. No dashboard produced that sentence. A person did, by paying attention to another person.
A second split hides inside this one, the gap between what people say and what they do. Ask users whether they'd use a feature and they'll be polite. Watch them, and the truth is in their hands. Strong research programmes cover both axes at once, how much (quantitative to qualitative) and stated versus observed (attitudinal to behavioural).
The research methods, from one extreme to the other
Lay the methods along those two axes and the whole toolkit appears. Most teams own one corner of it and forget the rest exists.
- Quantitative and behavioural (large numbers, real actions): analytics and flow analysis, web-performance metrics such as Lighthouse scores and Core Web Vitals field data (CrUX), A/B and multivariate tests, server logs, and the aggregate signals of frustration, rage-clicks, dead-clicks and pogo-sticking (the visitor bouncing back and forth because nothing was right). This is where an automated audit lives.
- Quantitative and attitudinal (large numbers, stated opinion): surveys at scale, NPS (Net Promoter Score), CSAT (Customer Satisfaction Score), the recurring satisfaction pulse. Cheap to run on thousands of people, and heir to everything people misreport about themselves.
- Qualitative and behavioural (few people, real actions): moderated usability testing, session replays, first-click and tree tests, card sorts, field and diary studies that catch the site in the wild. Small samples, but you watch the failure actually happen.
- Qualitative and attitudinal (few people, stated experience): user interviews, contextual inquiry, reading your own support tickets, and the plain, humbling exercise of reading a page aloud or handing it to someone who navigates by keyboard or screen reader. This is where you learn the why a metric can only gesture at.
You don't need all of it. You need to know which corner your current evidence comes from, and therefore which corner it's blind to.
Where a quality tool sits, and what it can't see
Webperf Cloud lives firmly in the first corner: quantitative and behavioural, at scale. That's the point of it. It will tell you, across two thousand pages, which ones are slow, which set cookies before consent, which fail an accessibility check, and how many real visitors feel it in the field. It is tireless where humans get bored, and it speaks for the visitors on cheap phones and poor connections that nobody in your office is using. Measurement does things human attention simply can't.
What it can't tell you is whether a slow page frustrated anyone, or why they gave up, or whether the accessible-in-theory form is usable in practice for the person it was built for. That's not a flaw to apologise for. It's the reason the tool points, deliberately, at the other kind of evidence. An automated check can prove your alt text exists; only a person can tell you it's meaningful.
Run them as a loop, not a rivalry
The mistake is treating this as a choice of sides. The teams who get the most from either method use them in sequence, each covering the other's blind spot.
- Start wide, quantitatively. Let the audit or the analytics find where and how big. 29% abandonment, or 900 pages with the same fault, is your map. It stops you spending scarce qualitative time on a problem that touches four people.
- Form a hypothesis you can be wrong about.
People abandon because the password rules stay hidden until they fail.
Write it down before you look, so the session tests an idea instead of confirming a mood. - Go deep, qualitatively. Watch five people. Read the support tickets about that form. Replay a dozen sessions. Five users is famously enough to surface most usability problems, precisely because you're explaining a cause, not measuring a rate.
- Fix, then measure again quantitatively. The figure you started from is also your proof. Did abandonment move, tested under the same conditions? Quant scoped the problem, qual explained it, quant confirms the repair.
Quantitative to find it, qualitative to understand it, quantitative to prove you fixed it. The loop is the product, not any single number inside it.
Each one lies when it's alone
Left to itself, quantitative evidence drifts toward whatever is easy to count, and a roadmap made only of measurable things quietly stops being a roadmap of important ones. Chase a score long enough and a team learns to satisfy the test rather than the visitor. Left to itself, qualitative evidence hands you five vivid stories that may or may not generalise, and the danger is rebuilding the checkout around the one articulate person in the room. A number without a story is trivia; a story without a number is an anecdote. Put them side by side and each keeps the other roughly truthful.
Collect the best of both. The measurable half, at scale and continuously, so nothing hides. And the human half, in small, deliberate doses, so the numbers mean something.
We build the quantitative half, continuously and across every page, in a form you can inspect and export. Pair it with an hour spent watching real people. See the plans, or email hello@webperf.cloud.