Baymard UX Teardown: Selling Evidence to People Who Audit Evidence

Baymard's business is telling ecommerce teams what's wrong with their interfaces. That makes its own marketing site an unusually exposed piece of work: every visitor is professionally equipped to audit it back. A vague hero, an unsourced number, a stock illustration standing in for a product: the audience catches all of it, because catching it is the job they're shopping for. Every page in this teardown answers the same way. Put the evidence first, and keep the product visible while you do it.

The number arrives before the promise

Stripe, Hotjar and most of the SaaS pages in this teardown series sell an outcome first and back-fill the proof somewhere below the fold. Baymard runs that order backwards. The subhead names a quantity — 200,000+ hours of validated research — before a single capability appears, and the only supporting line above the logo wall is another count: 29,000+ professionals, 71% of all Fortune 500 ecommerce companies.

Step through the three pages that carry the pitch:

Sitting beside the headline is the product itself: a panel of numbered guidelines, each scored with a red or green marker, stacked against a live product page. So the fold reads proof, product, proof. The whole sequence depends on the number being large enough to carry a headline on its own.

One screenshot argues better than the headline above it

The research overview reuses the same hero skeleton and swaps the illustration. It shows a real retailer's page with guideline verdicts pinned to specific regions: one green pass, two fails in red and orange, each dotted back to the element it judges.

The headline over it says "The UX Intelligence Platform for Ecommerce Teams," a sentence made entirely of words that constrain nothing. Platform could mean a dashboard, a PDF library, or a Slack bot. The illustration settles it: what you get is a verdict that lands on a coordinate. That's recognition doing work recall can't (NN/g on recognition vs recall). One look at a red marker clipped to a search field, and the deliverable is understood.

The paywall that gives away the distribution

The public benchmark is where the site stops arguing and hands over product. As of this August 2026 capture: 335 ecommerce sites plotted as coloured dots across five bands running POOR to PERFECT, split by surface into Overall UX Performance, Desktop Web, Homepage & Navigation and On-Site Search, with the guideline count printed under every row label.

Look at where the dots sit. They pile up left of centre in every row, and in the two page-level rows there's a hard red cluster jammed against the poor end. The dots are the pitch. Nobody had to write the sentence.

What you can't read is any individual site's per-guideline scores; the values stay premium. That trade is sharper than the usual gated whitepaper, because the free half is the half that creates the anxiety: most major ecommerce sites are mediocre at this, yours is probably among them, and here is the ruler that says so.

The banding is a fudge, though. Five labelled buckets turn a continuous score into a verdict, and somebody decided where DECENT ends without saying who or on what basis. That conversion is precisely what makes the chart survivable in a stakeholder deck. A dot at 62 starts an argument; the word GOOD ends one, so the fudge is doing the work.

Length as a deliberate strategy

The homepage is very long. Use-case grid, a stat row, five testimonials with employers attached, an industry cloud, plans by company size, a feature index, and a footer that doubles as a site map.

Length is usually a symptom of a pitch that hasn't resolved. Here it reads as a decision. The page serves at least three buyers who need different evidence: a researcher who wants the method, a design director who wants to see Nike and Adidas in the logo wall, and a VP who wants the 5% conversion-uplift figure Baymard prints in its own stat row. Baymard declines to choose between them, and the near-monochrome palette with one reserved blue is what keeps that much material from reading as noise.

Memory is what it costs. Ask someone what this homepage said and they'll describe a feeling of authority; nobody quotes a line from it. For an institution selling credibility, that's a fair trade.

What a smaller catalogue can't do

The move worth stealing is the free distribution: show the shape of your dataset, charge for the values inside it. But it only works at Baymard's scale. The chart is persuasive because 335 sites is enough for the clustering to read as an industry verdict — at thirty sites the same picture is an anecdote with error bars, and a buyer who evaluates evidence for a living will say so out loud.

So before you gate anything, check whether your dataset is large enough that its shape indicts the reader. If it isn't, giving it away costs you the asset and buys you nothing.