The UX-ROI Curve: Why Some Sites Print Money and Others Bleed It

Two businesses of similar size, in similar markets, with similar traffic. One website generates a steady stream of enquiries. The other generates almost nothing. The design budgets were not very different, and neither site is obviously bad.

The usual explanation is a curve: make many small improvements, watch them compound, and returns accelerate. It is a satisfying story and it is how a great deal of conversion work is sold. It is also not something anyone has shown to be true.

The compounding curve does not exist

Retiring it costs less than you would think, because the honest version of the argument is more useful anyway.

Interface improvements do not multiply. If you move a button and it helps, and then you change the button’s wording and that helps too, you have two independent small gains that add. They do not feed each other. And the second change is almost always worth less than the first, because the first one fixed the more obvious problem. That is diminishing returns, which is the normal shape of optimisation work and the opposite of a compounding curve.

What actually separates the two sites in the opening paragraph is rarely an accumulation of refinements. It is usually a small number of large faults — one thing that makes the site slow, one thing that makes it unclear, one thing that makes it hard to buy from. Fix those and the difference is dramatic. Polish the rest and the difference is difficult to detect at all.

The one lever with a price tag attached

If you want a UX change with a documented effect on revenue, there is essentially one: make the site faster.

In Milliseconds Make Millions, research commissioned by Google and carried out by 55 and Deloitte across 37 brands and more than 30 million sessions, a 0.1 second improvement in mobile site speed moved 3.2% more retail shoppers from the product listing to the product page and 9.1% more from the product page to the basket. Retail consumers spent 9.2% more, and travel bookings rose 10%. Read the framing carefully: those are progression rates between funnel steps in retail and travel on mobile, not a universal conversion uplift. It is still the strongest speed-to-money evidence in public.

Google also publishes what “fast enough” means, which removes the argument. Its Core Web Vitals guidance asks for a Largest Contentful Paint within 2.5 seconds, an Interaction to Next Paint of 200 milliseconds or less, and a Cumulative Layout Shift of 0.1 or less, all measured at the 75th percentile of real page loads. If your site misses those, you have found a large fault, and it has a known value.

Everything else is a hunt for failures

Beyond speed, the productive work is not optimisation. It is diagnosis — finding the places where the site simply does not work, which are almost never where the team assumes.

The cheapest diagnostic is also the oldest. Jakob Nielsen of Nielsen Norman Group argued in March 2000 that a first study with five participants finds 85% of the usability problems in a design, and that at least 15 users are needed to find all of them. Five people, one real task each, someone watching without helping. It costs an afternoon and it consistently outperforms months of speculative tweaking, because it finds the large faults rather than the small ones.

Three places are worth checking before any others, because they are where money leaves most reliably:

1. The first screen: does a stranger know what you sell, to whom, and where, without scrolling?
2. The moment of commitment: the form, the basket, the booking step — how much are you demanding right at the point of highest hesitation?
3. The people you are excluding: can the site be used on a keyboard, at speed, on a mid-range phone, by someone with low vision?

On the first: writing in September 2011 on an analysis of more than 205,000 pages, Nielsen Norman Group concluded that users often leave within 10-20 seconds, and that holding attention beyond that requires communicating your value proposition within 10 seconds. A page that spends its opening on ambition rather than substance is losing people before any interface decision matters.

On the second, the clearest numbers come from ecommerce and should be read as ecommerce. Baymard Institute puts the average documented online shopping cart abandonment rate at 70.22%, averaged across 50 separate studies. That figure does not transfer to a contact form on a professional services site — but it does establish that abandonment at the point of commitment is normal, and that anything you add to that moment is expensive.

On the third: the WebAIM Million survey for 2026 found detected WCAG 2 failures on 95.9% of the top one million home pages, averaging 56.1 errors per page, and automated testing only catches part of the problem. An inaccessible site is not a compliance issue with a revenue side-effect. It is a portion of your traffic that arrived, could not proceed, and left without appearing in any funnel report as anything other than a bounce.

How to spend, in order

If the compounding curve is not real, the sensible strategy is not “improve everything continuously”. It is “remove the largest faults, then stop and check whether anything else is worth doing”.

Most agencies sell that list in the opposite order, because the fifth item is the one that produces an ongoing retainer. The first four are the ones that produce results.

The uncomfortable bit about proving any of it

There is a reason the compounding story survives: it is very hard to disprove on a normal website.

A/B testing needs a great deal of traffic before a small difference can be told apart from noise. A business getting a few hundred visits a week to a key page will never accumulate enough of them to distinguish a genuine improvement from a fortnight of ordinary variation. Which means a great many “we tested this and it lifted conversions” claims rest on samples too small to support the conclusion. The exceptions are the large multi-brand studies run across millions of sessions, which is exactly why the speed research above is worth citing and a single agency case study usually is not.

So be honest about what you can and cannot know at the traffic you actually have.

Treat anyone who says otherwise with the scepticism they have earned.

The sites that turn traffic into money are not the ones running the most experiments. They are the ones where somebody found the three things that were badly broken, fixed them properly, and then had the discipline to leave the rest alone.