Performance is a design decision, and we score it like one
Speed is one of the seven UXE criteria. Why perceived performance is a design problem rather than an engineering one, what field data editors look at, and the five design choices that make products feel fast.
Most design awards don't score performance at all, which is strange: a slow interface is a bad interface, and almost every decision that makes a product slow was made by a designer. Image weight, font count, how much has to load before anything is useful, whether the layout moves while it settles — those are design decisions with engineering consequences, not the other way round.
It is one of our seven criteria, scored on every entry. Here's what that means in practice.
Field data, not a lab score
A lab score measured on a fast laptop on office wifi says very little. What matters is what real users experienced, at the 75th percentile — which is to say, what the slowest quarter of your audience actually got.
- ≤ 2.5s
Largest Contentful Paint
When the main content appears. The single best proxy for "has this page started?"
- ≤ 200ms
Interaction to Next Paint
How long the interface takes to visibly respond to a tap or a click.
- ≤ 0.1
Cumulative Layout Shift
How much the page moves under the reader. The cause of most mis-taps.
For apps and products behind a login, where field vitals may not exist, send what you have: cold-start times, time-to-interactive on a mid-range device, or your own instrumentation. Method matters more than the number.
Perceived speed is the one that counts
Two products with identical load times can feel completely different, because perception is shaped by what the interface does while waiting. This is where design does the work.
- Respond within 100ms5Not finish — respond. A pressed state within 100ms buys you seconds of patience.
- Show structure first5Navigation, headings and layout before data. Never a blank rectangle.
- Reserve the space4Skeletons at final dimensions. Nothing may move once content arrives.
- Be optimistic4Show the result immediately where you can safely reconcile later.
- Stream, don't block3One slow query must not hold the whole page.
The five design decisions that cost the most
- A hero image nobody measured. A 3MB photograph at the top of the page is the most common single cause of a failed LCP.
- Fonts that block text. Four weights of a custom face, loaded before anything can be read. Two weights and a sensible fallback are almost always enough.
- Third-party tags in the critical path. Analytics, chat, consent, heatmaps — each one is someone else's code deciding how fast your product feels.
- Layout that settles. A banner that arrives late and pushes the page down is how people click the wrong thing.
- Animating the wrong properties. Transitions on layout properties cause work on every frame; transforms and opacity don't.
Speed as an equity question
Performance is not only comfort. A heavy product costs more on a metered connection, fails sooner on an older device, and drains a battery faster — which means the cost falls hardest on the people with the least. A 4MB page is a decision about who your product is for.
The fastest interface is the one that didn't need to load. The second fastest is the one that told you what it was doing while it did.
What to send in an entry
Field Core Web Vitals if you have them, with the date range and the population. If not: cold-start times on a named mid-range device, your page-weight budget, or a before-and-after from a performance project.
What earns the most is showing the decision: "we cut the hero video because LCP at p75 was 4.1s on mobile, and accepted a less cinematic homepage." That's a designer weighing an experience against a number, which is exactly what this criterion is for.
Enter a project — and if performance is where your year went, say so in the case study. It's rare enough to stand out.
Budgets, and who owns them
A performance budget is a design artefact: a stated limit on page weight, request count and time-to-content that every subsequent decision has to fit inside. Teams that have one make different decisions — they choose two font weights instead of five, and they ask what a third-party script is worth before adding it.
- Images45%Usually the largest share, and the easiest to halve with modern formats and correct sizes.
- Scripts30%Including third parties, which are someone else's decision about your product.
- Fonts15%Two weights is nearly always enough. Every extra one blocks text.
- Everything else10%
The third-party problem
Analytics, chat widgets, consent managers, heatmaps, A/B tools and tag managers are each defensible individually and ruinous collectively. They also tend to be added by people who don't see the performance consequence.
- Count them. Most teams are surprised by the number.
- Load them after interactivity, never in the critical path.
- Give each one an owner and a review date — an orphaned tag runs forever.
- Measure with and without. If the difference is large, that's a business decision, not a technical one.
Perceived speed on a slow connection
Throttle your own product to a slow connection and use it for five minutes. It is the single most useful performance exercise available, and it changes design decisions rather than engineering ones: what should render first, what can be cached, what should be optimistic, and which animation is now unbearable.
Questions we get asked
Do you use Lighthouse scores?
As context only. Field data at the 75th percentile is far more meaningful — it describes what your slowest quarter of users actually experienced, rather than one run on one machine.
What if our product has no public traffic to measure?
Send what you have: cold-start times on a named device, your own instrumentation, or a before-and-after from a performance project. The method matters more than the number.
Is a heavy, beautiful site automatically penalised?
Not automatically — it is scored on the trade-off. A rich experience that stays under 2.5 seconds at p75 is impressive. The same experience at six seconds is a decision made at the user's expense.
How much can design actually change performance?
Most of it. Image choices, font count, what loads before content, whether the layout is stable and whether the interface responds immediately are all design decisions with engineering consequences.
Assessed work that scores highest on performance
Chosen by score from the work the editors have assessed, not by hand.
- performance
- core web vitals
- scoring
Written by
The UXE editors
The team that selects the work, uses it, scores the seven criteria and writes every verdict on this site.
Keep reading
Craft & practice6 min read
What separates the best SaaS dashboard UX design from the rest
The patterns behind the best SaaS dashboard UX design: default views that answer a question, density that respects real data, states that are designed, and the performance budget nobody talks about.
The UXE editors
Craft & practice5 min read
E-commerce UX design awards: what separates a good store from a scoring one
How retail, marketplace, D2C and subscription experiences are judged: findability, product pages that answer real questions, honest availability, and the post-purchase stretch everyone forgets.
The UXE editors
Craft & practice5 min read
Interaction design that earns its motion
Interaction and feedback is one of the seven UXE criteria — and the one most often mistaken for animation. What editors score: response times, state changes, undo over confirmation, and motion that does a job.
The UXE editors